[{"data":1,"prerenderedAt":2962},["ShallowReactive",2],{"news-list-en":3},[4,290,366,434,502,727,982,1083,1202,1590,1976,2518,2729],{"id":5,"title":6,"author":7,"body":8,"category":271,"date":272,"description":273,"extension":274,"image":275,"meta":276,"navigation":277,"path":278,"readTime":279,"seo":280,"stem":281,"tags":282,"translationId":288,"__hash__":289},"news\u002Fnews\u002F2026-04-27-ai-agent-deployment-without-vendor-lock-in.md","AI Agent Deployment Without Vendor Lock-In","Gabriel Lagerström de Jong",{"type":9,"value":10,"toc":257},"minimark",[11,15,18,21,24,29,32,67,70,74,77,80,83,90,94,97,100,103,106,109,112,116,119,122,125,128,139,142,146,149,152,155,158,162,165,168,171,180,183,187,190,193,196,199,203,206,209,212,215,219,222,225,254],[12,13,14],"p",{},"AI agents are quickly becoming part of how modern companies work. Coding agents, productivity agents, research agents and document agents can help employees move faster, automate repetitive work and get more value from company knowledge.",[12,16,17],{},"But for many organisations, AI agent deployment quickly becomes messy.",[12,19,20],{},"Different teams start using different tools. One team uses Claude Code. Another experiments with Codex. Someone else connects directly to an API. Finance receives several invoices. IT loses control over access. Management cannot see who is using what, how much it costs or whether the tools are being used in a secure and structured way.",[12,22,23],{},"Walma AI solves this by giving companies one platform for AI agent deployment, with central control, usage tracking, cost management and freedom to choose between different AI models.",[25,26,28],"h2",{"id":27},"the-simple-structure-companies-need","The simple structure companies need",[12,30,31],{},"With Walma, AI agent deployment becomes easier to understand, manage and scale:",[33,34,35,43,49,55,61],"ul",{},[36,37,38,42],"li",{},[39,40,41],"strong",{},"1 platform"," for all AI agents",[36,44,45,48],{},[39,46,47],{},"1 invoice"," for all AI agent usage",[36,50,51,54],{},[39,52,53],{},"100% visibility"," per employee, team and model",[36,56,57,60],{},[39,58,59],{},"0 forced lock-in"," to one AI vendor or model",[36,62,63,66],{},[39,64,65],{},"Real-time control"," over usage, budgets and cost",[12,68,69],{},"This is what makes AI adoption practical at company level. Employees get access to powerful AI agents. Management gets control. Finance gets clarity. IT gets structure.",[25,71,73],{"id":72},"a-central-platform-for-all-ai-agents","A central platform for all AI agents",[12,75,76],{},"Walma AI helps companies deploy and manage AI agents across the organisation from one place.",[12,78,79],{},"Instead of being locked into one AI vendor, one model or one specific tool, companies can use several leading AI agents through the same structure. This can include coding agents such as Claude Code, Codex and other market-leading AI tools. The key difference is that Walma provides the layer for governance, administration and follow-up.",[12,81,82],{},"That means your company can deploy AI agents across teams, manage users centrally, track usage per employee, control costs before they grow, switch between AI models over time and keep one invoice for all AI agent usage.",[12,84,85],{},[86,87],"img",{"alt":88,"src":89},"Walma AI Agents Dashboard","\u002Fimages\u002Fnews\u002Fwalma-agents-dashboard.jpg",[25,91,93],{"id":92},"one-invoice-for-all-ai-agent-usage","One invoice for all AI agent usage",[12,95,96],{},"Managing several AI tools often creates unnecessary administration. Separate subscriptions, vendor accounts, API payments and employee licences make it harder to understand the real cost of AI across the company.",[12,98,99],{},"With Walma, companies get one invoice for AI agents.",[12,101,102],{},"This makes life easier for finance, IT and management. Instead of chasing costs across different vendors, your organisation gets one clear commercial setup for AI usage.",[12,104,105],{},"As AI adoption grows, this becomes increasingly important. A company may start with 5-10 users testing coding agents. A few months later, the same company may have 50 or 100 employees using agents for analysis, documentation, support, sales or internal workflows.",[12,107,108],{},"Without a central structure, this quickly becomes hard to manage.",[12,110,111],{},"With Walma, scaling AI usage does not mean scaling the administration around it.",[25,113,115],{"id":114},"track-ai-usage-per-employee","Track AI usage per employee",[12,117,118],{},"One of the biggest challenges with AI agents is visibility.",[12,120,121],{},"Without proper tracking, it is difficult to know who is using AI tools, how often they are being used and what the actual cost is per employee or team.",[12,123,124],{},"Walma gives companies a central dashboard where usage can be followed per employee, team and model. This creates a clear picture of AI adoption across the organisation.",[12,126,127],{},"You can see which employees are using AI agents, which models and tools they are using, how much each user consumes, where costs are increasing and which teams are getting the most value from AI.",[12,129,130,131,135,136],{},"Instead of asking ",[132,133,134],"em",{},"\"Why did our AI costs increase this month?\""," you can ask ",[132,137,138],{},"\"Which teams are using AI the most, which models are driving the cost, and where is the value being created?\"",[12,140,141],{},"That is a much better conversation.",[25,143,145],{"id":144},"control-ai-costs-before-they-become-a-problem","Control AI costs before they become a problem",[12,147,148],{},"Powerful AI agents can create real value, but they can also generate unpredictable costs if usage is not controlled.",[12,150,151],{},"Walma gives administrators the ability to manage budgets, monitor usage and set limits. This makes it possible to give employees access to advanced AI tools while still keeping cost under control.",[12,153,154],{},"For example, companies can set monthly budgets per user, team or model. Admins can receive alerts when usage increases and stop usage when limits are reached.",[12,156,157],{},"This matters because not every task requires the most powerful model. Some users may need access to advanced models for complex development work. Others may only need lighter models for everyday tasks. Walma makes that structure manageable.",[25,159,161],{"id":160},"avoid-vendor-lock-in-with-flexible-ai-model-access","Avoid vendor lock-in with flexible AI model access",[12,163,164],{},"The AI market changes fast. The best model today may not be the best model six months from now.",[12,166,167],{},"That is why companies should avoid building their entire AI strategy around one vendor.",[12,169,170],{},"With Walma, organisations are not tied to one AI model or one provider. Companies can switch, combine and update their AI agents as the market evolves. If a better model is released, your organisation can adapt without rebuilding the entire setup.",[12,172,130,173,176,177],{},[132,174,175],{},"\"Which model should we bet everything on?\""," the better question becomes ",[132,178,179],{},"\"How do we build an AI structure that lets us use the best models over time?\"",[12,181,182],{},"That is the structure Walma provides.",[25,184,186],{"id":185},"secure-and-scalable-ai-agent-deployment","Secure and scalable AI agent deployment",[12,188,189],{},"AI agent deployment is not only about productivity. It also needs to be secure, structured and scalable.",[12,191,192],{},"Walma supports companies with a more controlled setup for data handling, access management and administration. The platform can be run through dedicated and regionally placed servers in Azure, with the possibility to choose region, for example within the EU, for safer and more compliant data handling.",[12,194,195],{},"This gives companies a stronger foundation than simply sharing API keys or letting employees create separate accounts with different AI vendors.",[12,197,198],{},"It also makes onboarding and offboarding easier. New users can be added through the dashboard. Access can be removed when someone leaves the company or changes role. Usage history is kept, while future access is closed.",[25,200,202],{"id":201},"from-scattered-ai-tools-to-managed-ai-infrastructure","From scattered AI tools to managed AI infrastructure",[12,204,205],{},"The first phase of AI adoption was experimentation. A few people tried ChatGPT. A few developers tested Claude Code. Different teams found their own tools.",[12,207,208],{},"That phase was useful. But it is not enough for serious company-wide deployment.",[12,210,211],{},"The next phase is managed AI infrastructure. That means one place to handle access, usage, cost, models, security and reporting. It means treating AI agents as a real part of the company's digital environment, not as separate tools floating around in different departments.",[12,213,214],{},"This is where Walma comes in.",[25,216,218],{"id":217},"why-companies-choose-walma-for-ai-agent-deployment","Why companies choose Walma for AI agent deployment",[12,220,221],{},"Walma AI is built for companies that want to use AI agents seriously, not just test them in isolated pockets.",[12,223,224],{},"With Walma, your company gets:",[33,226,227,230,233,236,239,242,245,248,251],{},[36,228,229],{},"1 platform for AI agent deployment",[36,231,232],{},"1 invoice for all AI agents",[36,234,235],{},"100% usage visibility per employee, team and model",[36,237,238],{},"Budget control before costs grow",[36,240,241],{},"Flexible access to different AI models",[36,243,244],{},"No forced vendor lock-in",[36,246,247],{},"Central dashboard for administration",[36,249,250],{},"Secure infrastructure through Azure",[36,252,253],{},"Scalable setup for teams and organisations",[12,255,256],{},"The result is simple: companies can give employees access to powerful AI agents while keeping control over cost, security and long-term flexibility.",{"title":258,"searchDepth":259,"depth":260,"links":261},"",2,3,[262,263,264,265,266,267,268,269,270],{"id":27,"depth":259,"text":28},{"id":72,"depth":259,"text":73},{"id":92,"depth":259,"text":93},{"id":114,"depth":259,"text":115},{"id":144,"depth":259,"text":145},{"id":160,"depth":259,"text":161},{"id":185,"depth":259,"text":186},{"id":201,"depth":259,"text":202},{"id":217,"depth":259,"text":218},"Insight","2026-04-27","One platform. One invoice. Full control over usage, cost and AI models. How Walma helps companies scale AI agents without being locked to one vendor.","md","\u002Fimages\u002Fnews\u002Fai-agent-deployment.jpg",{},true,"\u002Fnews\u002F2026-04-27-ai-agent-deployment-without-vendor-lock-in","8 min read",{"title":6,"description":273},"news\u002F2026-04-27-ai-agent-deployment-without-vendor-lock-in",[283,284,285,286,287],"AI agents","enterprise AI","vendor lock-in","cost management","security","ai-agent-deployment-without-vendor-lock-in","igQYjV4BU9jZmS2eicivlCV8AJd-uOLdmLVSWpJ_8Uo",{"id":291,"title":292,"author":293,"body":294,"category":271,"date":351,"description":352,"extension":274,"image":353,"meta":354,"navigation":277,"path":355,"readTime":356,"seo":357,"stem":358,"tags":359,"translationId":364,"__hash__":365},"news\u002Fnews\u002F2026-04-24-ai-as-revenue-stream.md","AI as a revenue stream, not just cost savings","Johan Holmström",{"type":9,"value":295,"toc":345},[296,299,302,305,309,312,316,319,322,326,329,332,335,338,342],[12,297,298],{},"Most people I talk to justify their AI investment with cost savings. Faster processes, fewer manual steps, more efficient support. That's fine. But it's only half the picture.",[12,300,301],{},"What I see among our customers, particularly in the security industry, is that AI is starting to create entirely new revenue streams.",[12,303,304],{},"IDC has now put numbers on it. Their Frontier Firms study (4,000 organisations, November 2025) shows that GenAI investments deliver an average of 2.8x ROI globally. In Western Europe, the figure is 2.7x. Agentic AI — AI that doesn't just answer but acts — already delivers 2.4x in Europe despite most implementations still being in pilot or testing phase.",[25,306,308],{"id":307},"what-the-returns-actually-consist-of","What the returns actually consist of",[12,310,311],{},"What's interesting is what the returns consist of. 51% of organisations cite improved accuracy and consistency as the most important effect. 45% mention time savings. 40% improved customer experience. It's not just about doing the same things cheaper. It's about doing better things.",[25,313,315],{"id":314},"concrete-examples","Concrete examples",[12,317,318],{},"A security company with international operations used Walma's platform to build a sales support agent. The agent helps salespeople with product information, quote support and argumentation, adapted by market and customer. The result wasn't just that salespeople worked faster. They started selling products they hadn't sold before, because the agent knew more about the range than any individual salesperson could.",[12,320,321],{},"Another example. A technical knowledge agent for installers. Collects product documentation, manuals and installation data in one place. Installers get faster and more accurate answers in the field. It reduces errors and saves time, yes. But it also means the company can take on more jobs per week.",[25,323,325],{"id":324},"the-market-confirms-the-shift","The market confirms the shift",[12,327,328],{},"Microsoft's Data Security Index 2026 shows that 92% of surveyed decision-makers have confidence in using GenAI to strengthen data security. Confidence is growing. And with it, the willingness to do more.",[12,330,331],{},"IDC shows the same trend: 67% of Frontier firms are already monetising their GenAI investment, and 44% of all organisations plan to use industry-specific AI use cases to increase revenue within 24 months. The shift from \"saving money\" to \"making money\" is happening now.",[12,333,334],{},"When the support team handles 40% of cases automatically, time is freed up for proactive customer contact. When salespeople have an AI that knows the product range better than they do, new markets open up. When installers work faster, the company can take on more customers without hiring.",[12,336,337],{},"That's where we at Walma see the real change. Not AI that saves money. AI that makes money.",[25,339,341],{"id":340},"the-question-you-should-be-asking","The question you should be asking",[12,343,344],{},"The question for your company isn't whether AI can make your operations more efficient. It can. The question is what new offerings become possible when your employees have access to the right information in real time.",{"title":258,"searchDepth":259,"depth":260,"links":346},[347,348,349,350],{"id":307,"depth":259,"text":308},{"id":314,"depth":259,"text":315},{"id":324,"depth":259,"text":325},{"id":340,"depth":259,"text":341},"2026-04-24","Most justify their AI investment with efficiency gains. But the real change happens when AI creates new revenue. IDC shows GenAI delivers 2.8x ROI globally.","\u002Fimages\u002Fnews\u002Fai-som-intaktsstrom.jpg",{},"\u002Fnews\u002F2026-04-24-ai-as-revenue-stream","5 min read",{"title":292,"description":352},"news\u002F2026-04-24-ai-as-revenue-stream",[360,361,362,284,363],"AI","ROI","revenue","business development","ai-as-revenue-stream","Q5opv4CmE6Vw2MdpQ650D1t0KLnIC1nnrf4tCoxtSiw",{"id":367,"title":368,"author":293,"body":369,"category":271,"date":351,"description":423,"extension":274,"image":424,"meta":425,"navigation":277,"path":426,"readTime":356,"seo":427,"stem":428,"tags":429,"translationId":432,"__hash__":433},"news\u002Fnews\u002F2026-04-24-from-experiment-to-production.md","From experiment to production: how to know if your organisation is ready",{"type":9,"value":370,"toc":417},[371,374,377,380,383,387,390,394,397,401,404,407,411,414],[12,372,373],{},"82% of organisations have developed plans to use generative AI in their data security measures. That's up from 64% just a year ago, according to Microsoft's Data Security Index 2026.",[12,375,376],{},"Yet most get stuck. The pilot went well, but then nothing happens.",[12,378,379],{},"IDC's Frontier Firms study confirms the pattern. Only 22% of organisations globally are classified as \"Frontier firms\" — those already realising measurable returns on their AI investments. In Europe, the figure is even lower: 18%. Half of European companies fall into the \"laggards\" category, still in the exploration or planning phase. The gap between those who test and those who deliver is growing.",[12,381,382],{},"I've seen this pattern for 30 years, across different industries and technologies. It's rarely about the technology. It comes down to three things.",[25,384,386],{"id":385},"_1-ownership","1. Ownership",[12,388,389],{},"The pilot is driven by an enthusiast in the IT department. But when it's time to scale, the business needs to own the question. Who is the sponsor? Which process should be improved? If the answer is \"we want to test AI\" instead of \"we want to cut support response times by 40%\", the connection to business outcomes needed to move forward is missing.",[25,391,393],{"id":392},"_2-data","2. Data",[12,395,396],{},"Or rather, a lack of order in data. 29% of decision-makers in Microsoft's report cite poor integration between data security and data management platforms as their biggest challenge. 25% lack a unified overview. You can't build an AI assistant that answers questions about your policies if those policies are spread across 14 different systems, three of which haven't been updated since 2019.",[25,398,400],{"id":399},"_3-expectation-management","3. Expectation management",[12,402,403],{},"AI doesn't solve everything on day one. The first version of your AI assistant might answer correctly 70% of the time. That means it gets it wrong 30% of the time. Organisations that survive that phase — that iterate and improve rather than shut it down — are the ones that end up with a tool that actually transforms operations.",[12,405,406],{},"IDC's data shows that Frontier firms realise ROI within an average of 15 months. Not day one. Not even after six months. But those who persist and iterate see the returns come.",[25,408,410],{"id":409},"what-determines-success","What determines success?",[12,412,413],{},"At Walma, we've onboarded municipalities, property companies and security firms. What always determines success isn't our technology. It's that the customer has a person who owns the question, data that can be worked with, and patience to iterate.",[12,415,416],{},"If you're sitting with a successful pilot that hasn't moved forward, don't start with the technology. Start with the question: who in the business wakes up in the morning and cares that this works?",{"title":258,"searchDepth":259,"depth":260,"links":418},[419,420,421,422],{"id":385,"depth":259,"text":386},{"id":392,"depth":259,"text":393},{"id":399,"depth":259,"text":400},{"id":409,"depth":259,"text":410},"82% have plans to use generative AI in their data security measures. Yet most get stuck after the pilot. Here's what separates those who deliver from those who test.","\u002Fimages\u002Fnews\u002Fexperiment-till-produktion.png",{},"\u002Fnews\u002F2026-04-24-from-experiment-to-production",{"title":368,"description":423},"news\u002F2026-04-24-from-experiment-to-production",[360,284,430,361,431],"implementation","digital transformation","from-experiment-to-production","EgF9RhKR5_H2IxsNL1JHZK4BqLqJmxVoNCV5EzqVMkc",{"id":435,"title":436,"author":293,"body":437,"category":271,"date":489,"description":490,"extension":274,"image":491,"meta":492,"navigation":277,"path":493,"readTime":356,"seo":494,"stem":495,"tags":496,"translationId":500,"__hash__":501},"news\u002Fnews\u002F2026-04-20-shadow-ai-at-work.md","Shadow AI at work: one third of incidents linked to generative AI",{"type":9,"value":438,"toc":484},[439,442,445,448,452,455,458,461,464,468,471,474,477,481],[12,440,441],{},"I meet IT managers and security officers every week. Over the past year, one question has come up in almost every meeting: \"How do we know what our employees are doing with AI?\"",[12,443,444],{},"The question is justified. Microsoft's Data Security Index 2026 shows that 32% of organisations' data security incidents are linked to the use of generative AI tools. Over 70% of employees bring their own AI tools to work. 58% use personal devices to access GenAI at work.",[12,446,447],{},"The numbers don't surprise me. They confirm what I hear in the conversations.",[25,449,451],{"id":450},"the-problem-is-called-shadow-ai","The problem is called shadow AI",[12,453,454],{},"Employees use ChatGPT, Copilot or other tools without the company knowing. They log in with personal accounts. They paste in customer data, contract texts, internal communications. Not out of malice, but because the tools actually help them work faster.",[12,456,457],{},"But every time someone pastes a customer list into an uncontrolled AI tool, data leaves the company's firewall. Nobody knows where it ends up. Nobody can trace it.",[12,459,460],{},"IDC's Frontier Firms study (November 2025, 4,000 organisations globally) confirms this isn't just a Swedish problem. 31% of all organisations cite security and privacy as the biggest barrier to implementing GenAI. 29% highlight governance, sovereignty and quality. The problem grows as more tools reach the market.",[12,462,463],{},"In the security industry, where I've worked for over 30 years, this is a direct business risk. Our customers handle CCTV surveillance, access control systems, alarm data. The information is sensitive by definition.",[25,465,467],{"id":466},"the-solution-is-not-to-ban-ai","The solution is not to ban AI",[12,469,470],{},"35% of organisations expect more incidents next year, but the answer isn't to shut things down. The answer is to give employees approved tools that are at least as good as the uncontrolled alternatives.",[12,472,473],{},"That's exactly what we do at Walma. We build AI assistants that run in the customer's own, isolated environment in Azure Sweden. All data stays in Sweden. Each customer has their own database, their own URL, their own access controls. Employees get an AI that actually answers their questions, with source references to internal documents. And the IT department can see exactly what's happening.",[12,475,476],{},"47% of organisations are now implementing specific GenAI controls, an increase of 8 percentage points since 2024. The trend is clear. Those who succeed best are those who give employees better alternatives, not more bans.",[25,478,480],{"id":479},"what-are-you-doing","What are you doing?",[12,482,483],{},"Next time you hear that an employee is using ChatGPT to search your policy documents, ask yourself: why don't we have an internal tool that does the same thing, but securely?",{"title":258,"searchDepth":259,"depth":260,"links":485},[486,487,488],{"id":450,"depth":259,"text":451},{"id":466,"depth":259,"text":467},{"id":479,"depth":259,"text":480},"2026-04-20","32% of organisations' data security incidents are linked to generative AI tools. Over 70% of employees bring their own AI tools to work. Here's the reality, and how we solve it.","\u002Fimages\u002Fnews\u002Fshadow-ai.jpg",{},"\u002Fnews\u002F2026-04-20-shadow-ai-at-work",{"title":436,"description":490},"news\u002F2026-04-20-shadow-ai-at-work",[360,497,498,499,284],"data security","shadow AI","GenAI","shadow-ai-at-work","9xKZ3zQtrX6JE_0oTGA1-etSWoMfUSnX2W9fKYb9tnc",{"id":503,"title":504,"author":505,"body":506,"category":271,"date":715,"description":510,"extension":274,"image":716,"meta":717,"navigation":277,"path":718,"readTime":279,"seo":719,"stem":720,"tags":721,"translationId":725,"__hash__":726},"news\u002Fnews\u002F2026-04-13-top-5-things-for-successful-ai-adoption.md","Top 5 Things to Consider for Successful AI Adoption in Your Organisation","Walma",{"type":9,"value":507,"toc":706},[508,511,515,518,521,524,527,530,534,537,540,543,546,552,555,559,562,565,568,571,581,584,588,591,594,597,600,606,609,613,616,619,622,625,630,633,637,640,643,646,649,654,658,661,664,667,670,675,680],[12,509,510],{},"95% of corporate AI pilots fail. Here is what the successful 5% do differently — and how to make sure you are in that group.",[25,512,514],{"id":513},"the-ai-adoption-gap-is-real","The AI adoption gap is real",[12,516,517],{},"Most organisations are investing in AI. Few are getting results.",[12,519,520],{},"According to MIT's 2025 NANDA report, 95% of generative AI pilots at companies fail to deliver measurable business impact. A broader look at enterprise AI projects puts the failure rate at somewhere between 70% and 85%. And yet, 87% of large enterprises are actively implementing AI solutions, spending an average of $6.5 million per organisation per year.",[12,522,523],{},"The gap between investment and outcome is not a technology problem. It is an implementation problem.",[12,525,526],{},"The organisations that succeed — those reporting $3.70 in value for every dollar invested, with top performers reaching $10.30 — do not have better AI tools. They have a better approach.",[12,528,529],{},"Here are the five things they get right.",[25,531,533],{"id":532},"_1-start-with-real-work-not-theory","1. Start with real work, not theory",[12,535,536],{},"The most common mistake organisations make is treating AI adoption as an awareness exercise. A presentation. A demo. A seminar about what AI could do.",[12,538,539],{},"It does not work.",[12,541,542],{},"Research shows that generic AI training programs achieve only 23% sustained adoption rates. Role-specific training that addresses actual job functions? 67% sustained adoption. The difference is whether people get to use AI on problems they actually have — or sit through slides about problems they do not.",[12,544,545],{},"48% of employees rank hands-on training as the single most crucial factor for successful AI adoption. And participants who go through practical training engage in 25.8% more interactions with AI and write 30% more in their prompts — signals of real, embedded use rather than surface-level experimentation.",[12,547,548,551],{},[39,549,550],{},"What this means in practice:"," Do not roll out AI company-wide before people have had the chance to connect it to their daily work. Start with a structured session where everyone solves a real problem with AI — not a hypothetical one.",[12,553,554],{},"This is exactly what Walma's AI workshop is built around. Every participant works on their own actual tasks, coached hands-on, with AI tools configured and ready. 9 out of 10 participants report taking meaningful steps forward in their AI use — not after months of adoption campaigns, but after a single day.",[25,556,558],{"id":557},"_2-identify-your-low-hanging-fruit-first","2. Identify your low-hanging fruit first",[12,560,561],{},"One of the biggest reasons AI projects fail is that organisations try to boil the ocean. They launch broad transformation initiatives before establishing where AI actually creates value for them specifically.",[12,563,564],{},"The result: diffuse efforts, unclear ownership, hard-to-measure outcomes, and eventual abandonment.",[12,566,567],{},"The organisations that succeed start narrow and specific. They identify the tasks that are high-frequency, time-consuming, and well-defined — where the cost of AI implementation is low and the efficiency gain is immediate and measurable. Then they prove value there before expanding.",[12,569,570],{},"74% of executives who achieved ROI from AI did so within the first year — and they did it by focusing on concrete productivity wins, not organisation-wide transformation.",[12,572,573,576,577,580],{},[39,574,575],{},"The right question is not"," \"How do we become an AI-first organisation?\" ",[39,578,579],{},"It is"," \"What three things are we doing manually today that AI could handle by next week?\"",[12,582,583],{},"In Walma's workshops and strategy engagements, identifying low-hanging fruit is a core deliverable. Participants leave with a clear picture of what they can do themselves immediately, what is worth building into a system, and what is not worth pursuing at all. This clarity is what turns AI interest into AI action.",[25,585,587],{"id":586},"_3-take-data-security-seriously-from-day-one-not-as-an-afterthought","3. Take data security seriously from day one — not as an afterthought",[12,589,590],{},"Here is a number that should concern any leadership team: 39.7% of employee AI interactions involve sensitive data.",[12,592,593],{},"When employees start using AI tools — and they will, with or without official guidance — sensitive corporate information starts flowing through systems that may not be approved, audited, or even known to IT. Contracts. Customer data. Financial reports. Internal communications.",[12,595,596],{},"Data security is now the number one concern among executives when choosing AI infrastructure. And rightly so. Nearly 60% of AI leaders report that integrating AI with legacy systems and addressing risk and compliance concerns are their primary adoption challenges.",[12,598,599],{},"The organisations that get AI adoption right do not wait for a data breach to think about governance. They make security and data residency part of the foundation — not a retrofit.",[12,601,602,605],{},[39,603,604],{},"What good looks like:"," All AI interactions happening within a controlled, audited environment. Data that never leaves your approved infrastructure. AI models running on your data, not feeding it back into public training sets.",[12,607,608],{},"Noda, Walma's enterprise AI platform, is built on this principle. All data is hosted in Sweden, fully GDPR-compliant. Employees work within a secure environment where customer data can be used without compliance risk — which is precisely what makes the AI useful. You cannot solve real work problems with AI if you cannot feed it real data.",[25,610,612],{"id":611},"_4-make-sure-your-leaders-have-real-ai-literacy-not-just-awareness","4. Make sure your leaders have real AI literacy — not just awareness",[12,614,615],{},"Only 8% of enterprise leaders have a sufficient level of AI literacy, according to recent studies. Yet over 90% of C-suite executives claim to be knowledgeable about AI's capabilities.",[12,617,618],{},"That gap is dangerous.",[12,620,621],{},"When leaders do not truly understand what AI can and cannot do, two things happen. Either they underinvest — treating AI as a novelty rather than an operational lever. Or they overcommit — launching initiatives that are technically unfeasible or poorly scoped, draining resources and destroying internal credibility for future efforts.",[12,623,624],{},"43% of businesses point to a lack of vision among managers and leaders as a top barrier to AI adoption. Leaders do not need to be technical. But they need to be able to ask the right questions, evaluate proposals critically, and recognise where AI creates genuine leverage versus where it adds complexity.",[12,626,627,629],{},[39,628,604],{}," Leadership teams that have spent time working with AI themselves — not just receiving briefings about it. Executives who understand the difference between an AI chatbot and a retrieval-augmented system. Decision-makers who can read an AI vendor proposal and know what to scrutinise.",[12,631,632],{},"Walma's workshops are designed with key personnel in mind, not just frontline employees. When leadership experiences AI hands-on — working through real strategic and operational problems with AI tools — they develop the instincts needed to make better investment decisions and lead adoption with credibility.",[25,634,636],{"id":635},"_5-build-a-system-not-a-one-off-project","5. Build a system, not a one-off project",[12,638,639],{},"The final and perhaps most important distinction between organisations that succeed with AI and those that do not: successful ones treat AI adoption as infrastructure, not as a project with a start and end date.",[12,641,642],{},"One workshop is not enough. One tool rollout is not enough. AI adoption compounds when it is embedded into how work actually gets done — in the tools people use every day, the knowledge they can access, the processes that run in the background.",[12,644,645],{},"The data supports this. Companies that purchased AI from specialised vendors and built partnerships succeeded 67% of the time, compared to one-third success rate for internal builds attempted in isolation. The difference is sustained support, continuous improvement, and systems that are built to last beyond the initial rollout.",[12,647,648],{},"96% of organisations investing in AI experience productivity gains — but the ones reporting significant gains are those who moved from experimentation to embedded systems. Not those who ran pilots and moved on.",[12,650,651,653],{},[39,652,604],{}," Employees with a single interface to access your organisation's entire knowledge base. Processes that run automatically in the background — invoice review, document analysis, report generation — without manual intervention. An AI layer that improves as more data flows through it.",[25,655,657],{"id":656},"the-takeaway","The takeaway",[12,659,660],{},"Successful AI adoption is not about choosing the right model or the biggest budget. It is about sequence, specificity, and systems.",[12,662,663],{},"Start with real work. Find your quick wins. Secure your data from day one. Build leadership literacy through experience, not briefings. And then build the infrastructure that makes AI part of how your organisation operates — permanently.",[12,665,666],{},"The 5% that succeed are not smarter or better resourced. They just do these five things in the right order.",[668,669],"hr",{},[12,671,672],{},[132,673,674],{},"Walma is a Swedish AI consultancy that helps organisations go from interest to implementation. We run hands-on workshops, build enterprise AI systems, and help leadership teams make better decisions about where AI creates real value.",[12,676,677],{},[39,678,679],{},"Sources:",[33,681,682,685,688,691,694,697,700,703],{},[36,683,684],{},"MIT report: 95% of generative AI pilots at companies are failing",[36,686,687],{},"Why AI Adoption Stalls, According to Industry Data – HBR",[36,689,690],{},"AI Adoption Statistics 2026 – Netguru",[36,692,693],{},"AI Adoption Benchmarks 2025 – Worklytics",[36,695,696],{},"Data in the Wild: 40% of Employee AI Use Involves Sensitive Info",[36,698,699],{},"The ROI of AI 2025",[36,701,702],{},"Human Factors as Drivers of Success in Generative AI – UMU",[36,704,705],{},"AI Adoption Challenges 2025 – IBM",{"title":258,"searchDepth":259,"depth":260,"links":707},[708,709,710,711,712,713,714],{"id":513,"depth":259,"text":514},{"id":532,"depth":259,"text":533},{"id":557,"depth":259,"text":558},{"id":586,"depth":259,"text":587},{"id":611,"depth":259,"text":612},{"id":635,"depth":259,"text":636},{"id":656,"depth":259,"text":657},"2026-04-13","\u002Fimages\u002Fnews\u002Fai-adoption-strategy.jpg",{},"\u002Fnews\u002F2026-04-13-top-5-things-for-successful-ai-adoption",{"title":504,"description":510},"news\u002F2026-04-13-top-5-things-for-successful-ai-adoption",[360,722,723,724,430],"adoption","strategy","leadership","top-5-ai-adoption","-UW3VAdZUH9bosJWoS2wk-JMZ-XcQsN8txsbiKFrArA",{"id":728,"title":729,"author":505,"body":730,"category":271,"date":715,"description":968,"extension":274,"image":969,"meta":970,"navigation":277,"path":971,"readTime":972,"seo":973,"stem":974,"tags":975,"translationId":980,"__hash__":981},"news\u002Fnews\u002F2026-04-13-you-already-have-all-the-data-you-need.md","You already have all the data you need. The problem is that nobody can find it.",{"type":9,"value":731,"toc":959},[732,738,741,744,750,754,757,760,766,769,772,776,779,782,814,817,820,824,827,830,833,839,842,845,848,852,855,860,863,870,873,877,880,886,889,893,896,923,926,930,948,951,953],[12,733,734,735],{},"Every week we talk to companies that say the same thing: ",[132,736,737],{},"\"We're not really ready for AI yet. We need to sort out our systems first.\"",[12,739,740],{},"And every time we wonder: what do they actually mean?",[12,742,743],{},"Because when we dig deeper, it turns out they have loads of data. Customer history in old Excel files. Service reports in PDFs. Meeting notes in Word documents. Presentations with sales data. It's all there — just scattered, unstructured and impossible to search.",[12,745,746,747],{},"That's not a sign that you're not ready for AI. ",[39,748,749],{},"That's precisely the starting point AI is built to solve.",[25,751,753],{"id":752},"the-hidden-gold-in-your-folders","The hidden gold in your folders",[12,755,756],{},"Let's take a concrete example that we see over and over again.",[12,758,759],{},"A company sells and services equipment. Over the years they've created thousands of documents — installation reports, service protocols, warranty cases, customer correspondence. Everything sits in folders, in Dropbox, on a server, in an email archive.",[12,761,762,763],{},"The question they can never answer is: ",[132,764,765],{},"which of our customers are ready for a new service agreement?",[12,767,768],{},"The answer is in the documents. It's always been there. But nobody has the time to go through a thousand PDFs to find it.",[12,770,771],{},"That's exactly the problem we solve.",[25,773,775],{"id":774},"what-we-actually-do","What we actually do",[12,777,778],{},"We built a pipeline — an automated flow — that takes your files exactly as they are and transforms them into an AI system that understands the content.",[12,780,781],{},"It doesn't matter whether it's:",[33,783,784,790,796,802,808],{},[36,785,786,789],{},[39,787,788],{},"Excel files"," with customer data, sales figures or service history",[36,791,792,795],{},[39,793,794],{},"PDFs"," with protocols, reports, contracts or manuals",[36,797,798,801],{},[39,799,800],{},"Word documents"," with notes, quotes or internal guidelines",[36,803,804,807],{},[39,805,806],{},"PowerPoint presentations"," with sales material or product info",[36,809,810,813],{},[39,811,812],{},"Images and scanned documents"," — even old papers that have been photographed or scanned",[12,815,816],{},"In short: if the information exists somewhere in your folders, we can make it searchable and accessible to AI.",[12,818,819],{},"And you don't need to move a single document. You give us access to where the files already are — Dropbox, SharePoint, a network drive, whatever — and we handle the rest.",[25,821,823],{"id":822},"back-to-the-service-agreements","Back to the service agreements",[12,825,826],{},"Here's what happened when one of our clients gave us access to their document archive.",[12,828,829],{},"They had roughly a thousand PDF files with customer protocols. Installation dates, service history, warranty periods, technician notes. No structure. No search capability. Just a folder with files.",[12,831,832],{},"We ran everything through our pipeline. Three weeks later, the sales team could ask questions like:",[834,835,836],"blockquote",{},[12,837,838],{},"\"Which customers have equipment installed more than five years ago and haven't had a service visit in over two years?\"",[12,840,841],{},"The system answered in seconds — with a list of customers, with source references to the right protocols, with contact details.",[12,843,844],{},"These are leads that were hidden in a document archive. The salespeople knew they should be in there. They just never had the ability to find them.",[12,846,847],{},"Now they do.",[25,849,851],{"id":850},"the-impact-isnt-a-better-system-its-money","The impact isn't a better system — it's money",[12,853,854],{},"It's easy to talk about AI in terms of technology and structure. But what we actually deliver is quite simple:",[12,856,857],{},[39,858,859],{},"Customers you're missing today, you'll find tomorrow.",[12,861,862],{},"Time spent searching through old files is instead spent calling the right customers. Salespeople who didn't know what leads they had suddenly get a list. Managers who manually summarise data every month let the system do it.",[12,864,865,866,869],{},"Another client had their sales data in one Excel and their marketing data in another. The monthly report was done manually by one person — and only that person knew how the files connected. We got access to their folders and built a system where everything is shown together. Management can now ask ",[132,867,868],{},"\"How did the campaign in March compare to sales in April?\""," and get an answer immediately, without involving anyone.",[12,871,872],{},"A third client had everything in Dropbox — documents, presentations, protocols from several years. They didn't even know where to start. They gave us access on a Monday. The following week they had a working system.",[25,874,876],{"id":875},"why-it-doesnt-require-a-big-it-project","Why it doesn't require a big IT project",[12,878,879],{},"The classic path to a new system looks like this: map the current state → procure a new system → migrate all data → train everyone → hope it works. Plan for at least a year.",[12,881,882,885],{},[39,883,884],{},"Our approach:"," give us access → we fix the structure → you start using the system.",[12,887,888],{},"It works because we don't ask you to change how you work. Your Excels can remain your Excels. Your PDFs stay where they are. What we do is add an AI layer on top that understands what's in them.",[25,890,892],{"id":891},"what-can-you-ask-the-system","What can you ask the system?",[12,894,895],{},"It depends on what's in your documents — but examples from our clients:",[33,897,898,903,908,913,918],{},[36,899,900],{},[132,901,902],{},"\"Which customers haven't we heard from in over a year?\"",[36,904,905],{},[132,906,907],{},"\"What did the protocol from the installation at Lundgren's Industry in 2021 say?\"",[36,909,910],{},[132,911,912],{},"\"Show all cases where the warranty expired during 2024.\"",[36,914,915],{},[132,916,917],{},"\"What does the sales trend look like compared to the same period last year?\"",[36,919,920],{},[132,921,922],{},"\"Which customers bought model X and might be interested in the upgrade?\"",[12,924,925],{},"Questions that previously took hours to answer — if they could be answered at all — now take seconds.",[25,927,929],{"id":928},"how-do-you-get-started","How do you get started?",[931,932,933,936,939,942,945],"ol",{},[36,934,935],{},"You give us access to where the files already are — Dropbox, SharePoint, a server, a network drive.",[36,937,938],{},"We map what's there and tell you what we can build from the material.",[36,940,941],{},"We run the files through our pipeline and build an initial system.",[36,943,944],{},"You test — ask questions, see what the system finds, and tell us what's missing.",[36,946,947],{},"We iterate until it works exactly as you want.",[12,949,950],{},"In most cases we have a working system ready within a couple of weeks.",[668,952],{},[12,954,955,958],{},[39,956,957],{},"Do you have documents, files and presentations with information that's hard to find?"," Chances are your next service agreements are already hidden in there.",{"title":258,"searchDepth":259,"depth":260,"links":960},[961,962,963,964,965,966,967],{"id":752,"depth":259,"text":753},{"id":774,"depth":259,"text":775},{"id":822,"depth":259,"text":823},{"id":850,"depth":259,"text":851},{"id":875,"depth":259,"text":876},{"id":891,"depth":259,"text":892},{"id":928,"depth":259,"text":929},"Every week we talk to companies that say they're not ready for AI yet. But the truth is their data already exists – scattered across Excel files, PDFs and email archives. That's precisely the problem AI is built to solve.","\u002Fimages\u002Fnews\u002Fhidden-data-folders.jpg",{},"\u002Fnews\u002F2026-04-13-you-already-have-all-the-data-you-need","6 min read",{"title":729,"description":968},"news\u002F2026-04-13-you-already-have-all-the-data-you-need",[360,976,977,978,979],"data","documents","automation","business value","you-already-have-all-the-data","aHAGjrkkOVscXJ61T9gq68xKTU8Kp16pkQ7MFZh2Jik",{"id":983,"title":984,"author":505,"body":985,"category":1068,"date":1069,"description":1070,"extension":274,"image":1071,"meta":1072,"navigation":277,"path":1073,"readTime":1074,"seo":1075,"stem":1076,"tags":1077,"translationId":1081,"__hash__":1082},"news\u002Fnews\u002F2026-03-19-johan-holmstrom-vp-of-sales.md","Walma Strengthens Market Position – Recruits Johan Holmström as Vice President of Sales",{"type":9,"value":986,"toc":1063},[987,990,993,996,1000,1006,1012,1018,1024,1027,1031,1034,1037,1040,1044,1054],[12,988,989],{},"Walma, a fast-growing player in Enterprise AI, is now strengthening its commercial focus by bringing on Johan Holmström as Vice President of Sales. The recruitment marks a strategic step in the company's continued expansion – with the goal of taking a leading position in secure, business-driven AI in the Nordics and internationally.",[12,991,992],{},"Johan Holmström brings over 30 years of experience from the security and technology industry, where he has helped build and internationalize companies within digitalization and service-based business models. Together, Walma and Johan have already established several Enterprise AI implementations with customers in the security and installation industry – with clear results in both efficiency and business development.",[12,994,995],{},"– We are now seeing how AI is truly beginning to create business value – not just through efficiency gains, but as a driver for growth and new offerings. It's about connecting AI directly to business processes, says Johan Holmström.",[25,997,999],{"id":998},"concrete-examples-of-how-walmas-enterprise-ai-creates-value","Concrete Examples of How Walma's Enterprise AI Creates Value",[12,1001,1002,1005],{},[39,1003,1004],{},"AI-driven support agent"," – Automates and quality-assures support cases by giving technicians and support teams access to the right information in real time. By analyzing large volumes of historical support cases, the agent can quickly produce relevant answers, ready-made formulations, and concrete solution proposals directly in the workflow. The agent reduces manual handling and streamlines the support process.",[12,1007,1008,1011],{},[39,1009,1010],{},"Sales support agent for international organizations"," – Supports salespeople with product information, quote assistance, and argumentation – adapted to market, customer, and context. Particularly valuable for companies with complex offerings and multiple geographies.",[12,1013,1014,1017],{},[39,1015,1016],{},"Analytics and decision support linked to business and financial systems"," – Enables deeper insights into sales, production, and profitability by connecting data from multiple systems – and making it accessible to management and operations in real time.",[12,1019,1020,1023],{},[39,1021,1022],{},"Technical knowledge agent for installers and field personnel"," – Collects and structures product documentation, manuals, and installation data to provide faster and more accurate answers in the field – reducing errors and saving time.",[12,1025,1026],{},"Walma's Enterprise AI and AI agents are currently used by companies in both the private and public sectors. Customers include players in the security industry, installation companies, as well as municipalities and property companies.",[25,1028,1030],{"id":1029},"strategic-recruitment-for-the-next-phase","Strategic Recruitment for the Next Phase",[12,1032,1033],{},"– Bringing Johan into a more operational and commercial role is an important step for us. The combination of his industry experience, international network, and business focus means we can now accelerate our growth even further, says Gabriel Lagerström de Jong, CEO of Walma.",[12,1035,1036],{},"In his new role, Johan Holmström will focus on strengthening Walma's position in Sweden, expanding into new industries, and driving the company's international growth. Dialogues with several international product and service companies are already underway, where demand for secure and scalable Enterprise AI is growing rapidly.",[12,1038,1039],{},"– We are at a turning point where many companies are moving from experimentation to implementation. Those who succeed are the ones who connect AI to business – where AI either saves money and time or becomes part of new revenue streams for companies. That's where we make the difference together with our customers, concludes Johan Holmström.",[25,1041,1043],{"id":1042},"contact","Contact",[12,1045,1046,1048,1049],{},[39,1047,293],{},"\nVice President of Sales, Walma\n+46 763 10 00 60\n",[1050,1051,1053],"a",{"href":1052},"mailto:johan@walma.ai","johan@walma.ai",[12,1055,1056,1058,1059],{},[39,1057,7],{},"\nCEO, Walma\n+46 73 666 73 17\n",[1050,1060,1062],{"href":1061},"mailto:gabriel@walma.ai","gabriel@walma.ai",{"title":258,"searchDepth":259,"depth":260,"links":1064},[1065,1066,1067],{"id":998,"depth":259,"text":999},{"id":1029,"depth":259,"text":1030},{"id":1042,"depth":259,"text":1043},"Company","2026-03-19","Walma brings on Johan Holmström as VP of Sales to accelerate Enterprise AI in the security industry and internationally.","\u002Fimages\u002Fnews\u002Fjohan-holmstrom-vp-sales.jpg",{},"\u002Fnews\u002F2026-03-19-johan-holmstrom-vp-of-sales","4 min read",{"title":984,"description":1070},"news\u002F2026-03-19-johan-holmstrom-vp-of-sales",[1078,287,1079,1080,283],"Enterprise AI","recruitment","international expansion","johan-holmstrom-vp-sales","q2dosYksc3XxLx3RkMDPgoUD_1lcL7GcpuuKtou2jfg",{"id":1084,"title":1085,"author":7,"body":1086,"category":271,"date":1187,"description":1188,"extension":274,"image":1189,"meta":1190,"navigation":277,"path":1191,"readTime":356,"seo":1192,"stem":1193,"tags":1194,"translationId":1200,"__hash__":1201},"news\u002Fnews\u002F2026-03-17-chatgpt-business-data-export.md","Trying to Export Your ChatGPT Business Data? Here's the Uncomfortable Truth: You Can't",{"type":9,"value":1087,"toc":1181},[1088,1091,1094,1097,1100,1103,1107,1110,1113,1116,1119,1123,1126,1129,1132,1135,1138,1141,1145,1148,1151,1154,1157,1160,1163,1167,1170,1173,1176],[12,1089,1090],{},"A lot of companies are searching for the same thing right now: how to export ChatGPT Business data.",[12,1092,1093],{},"They go into settings.\nThey look for the button.\nThey assume they have missed something.",[12,1095,1096],{},"They haven't.",[12,1098,1099],{},"OpenAI says export is available for consumer ChatGPT accounts on Free, Plus, and Pro. But OpenAI also states that you can't export data from a ChatGPT Business account. That is what many companies are only discovering now, at the exact moment they try to leave.",[12,1101,1102],{},"And that should alarm business owners.",[25,1104,1106],{"id":1105},"this-is-not-just-chat-history","This Is Not Just Chat History",[12,1108,1109],{},"Because this is not just chat history. This is strategy work, drafts, internal processes, research, prompting logic, and months or years of accumulated company knowledge.",[12,1111,1112],{},"Data is the new gold.",[12,1114,1115],{},"And if a provider lets you put your gold in, but not take it with you in any clean way, that is not a minor product flaw. That is lock-in.",[12,1117,1118],{},"At Walma, we think those are unacceptable terms for businesses.",[25,1120,1122],{"id":1121},"the-bigger-picture","The Bigger Picture",[12,1124,1125],{},"And OpenAI is not alone. Anthropic may offer more export options in some cases, but that is still nowhere near enough. In a serious business setup, portability should not depend on special roles, limited eligibility, or vendor goodwill. It should be standard.",[12,1127,1128],{},"That is the deeper problem with big global AI model providers and the way they let companies use their services.",[12,1130,1131],{},"These are not neutral tools floating above the market. They are giant corporations under enormous pressure. They have made massive commitments to investors. They are spending at extreme levels. They are entering military and defense structures. And they need revenue at a scale that matches those ambitions.",[12,1133,1134],{},"That is their problem.",[12,1136,1137],{},"It should not become yours.",[12,1139,1140],{},"But the moment your workflows, internal reasoning, company knowledge, and operational memory are locked into their environment, their strategy becomes your problem too. Because now you are not just using the service. You are helping carry the weight of their business model.",[25,1142,1144],{"id":1143},"think-differently-about-ai-infrastructure","Think Differently About AI Infrastructure",[12,1146,1147],{},"That is exactly why businesses should think differently about AI infrastructure.",[12,1149,1150],{},"Your AI provider should work more like a bank you trust.",[12,1152,1153],{},"You should be able to store something valuable there.\nYou should be able to work with it securely.\nAnd above all, you should know it is still yours.",[12,1155,1156],{},"That is how we think about it at Walma.",[12,1158,1159],{},"We can store, structure, and help manage your data, but we know the difference between managing something and owning it. You own it. Always. We take pride in being careful guardians of that value, not gatekeepers standing in front of the exit.",[12,1161,1162],{},"Because models will change. The market will change. New leaders will emerge. And if your setup is sound, that should not be a threat. You should be able to change model provider without losing the value you have built up over time.",[25,1164,1166],{"id":1165},"what-companies-are-really-finding","What Companies Are Really Finding",[12,1168,1169],{},"So yes, a lot of companies are now searching for a guide to export their ChatGPT Business data.",[12,1171,1172],{},"What they are finding instead is something much bigger than a missing button.",[12,1174,1175],{},"They are discovering a business model built to keep them in.",[834,1177,1178],{},[12,1179,1180],{},"At Walma, we believe businesses deserve better: AI where the data stays yours, the infrastructure stays under your control, and the choice of model remains open. That should not be a premium feature. That should be the baseline.",{"title":258,"searchDepth":259,"depth":260,"links":1182},[1183,1184,1185,1186],{"id":1105,"depth":259,"text":1106},{"id":1121,"depth":259,"text":1122},{"id":1143,"depth":259,"text":1144},{"id":1165,"depth":259,"text":1166},"2026-03-17","Why that matters for AI lock-in, data ownership, and business control.","\u002Fimages\u002Fnews\u002Fchatgpt-data-export.jpeg",{},"\u002Fnews\u002F2026-03-17-chatgpt-business-data-export",{"title":1085,"description":1188},"news\u002F2026-03-17-chatgpt-business-data-export",[360,1195,1196,1197,1198,1199],"ChatGPT","data ownership","lock-in","OpenAI","data portability","chatgpt-business-data-export","MahsuYejwi9g29fwPmz24QIXRc2VbL0YAzWC4CcSXzE",{"id":1203,"title":1204,"author":505,"body":1205,"category":271,"date":1579,"description":1580,"extension":274,"image":1581,"meta":1582,"navigation":277,"path":1583,"readTime":972,"seo":1584,"stem":1585,"tags":1586,"translationId":1588,"__hash__":1589},"news\u002Fnews\u002F2026-03-06-gpt-5-4-whats-new.md","GPT-5.4: What's new in OpenAI's latest AI model?",{"type":9,"value":1206,"toc":1568},[1207,1210,1213,1216,1220,1223,1237,1240,1244,1251,1254,1257,1271,1274,1278,1281,1284,1298,1301,1305,1308,1315,1318,1329,1332,1336,1341,1344,1355,1358,1362,1365,1431,1434,1438,1441,1445,1453,1457,1465,1470,1478,1483,1490,1494,1497,1500,1520,1523,1527,1533,1539,1545,1556,1562],[12,1208,1209],{},"OpenAI continues to develop artificial intelligence at a rapid pace. The latest model, GPT-5.4, is a major update to ChatGPT, focusing on improved reasoning, integration with work tools and the ability to automate tasks across different applications.",[12,1211,1212],{},"The new model is designed to help users with more advanced tasks — from programming and analysis to document work and data management.",[12,1214,1215],{},"In this article, we cover what GPT-5.4 is, what new features it includes and how it differs from previous models.",[25,1217,1219],{"id":1218},"what-is-gpt-54","What is GPT-5.4?",[12,1221,1222],{},"GPT-5.4 is an updated version of OpenAI's language model that powers ChatGPT. It builds on previous generations but introduces several improvements in:",[33,1224,1225,1228,1231,1234],{},[36,1226,1227],{},"Advanced reasoning",[36,1229,1230],{},"Handling large datasets",[36,1232,1233],{},"Integration with other applications",[36,1235,1236],{},"Automated workflows",[12,1238,1239],{},"The goal of the model is to make AI a more powerful tool for work, analysis and productivity.",[25,1241,1243],{"id":1242},"thinking-mode-better-ai-reasoning","Thinking Mode – better AI reasoning",[12,1245,1246,1247,1250],{},"One of the biggest innovations in GPT-5.4 is ",[39,1248,1249],{},"Thinking Mode",".",[12,1252,1253],{},"This mode allows the model to analyse a problem more systematically before providing an answer. The result is often more accurate and well-considered responses, especially for complex tasks.",[12,1255,1256],{},"Thinking Mode is particularly useful for:",[33,1258,1259,1262,1265,1268],{},[36,1260,1261],{},"Programming",[36,1263,1264],{},"Mathematics",[36,1266,1267],{},"Logical problems",[36,1269,1270],{},"Advanced analysis",[12,1272,1273],{},"For developers and analysts, this means AI can function more as a problem solver than a regular chatbot.",[25,1275,1277],{"id":1276},"integration-with-excel-and-google-sheets","Integration with Excel and Google Sheets",[12,1279,1280],{},"GPT-5.4 also features improved integration with spreadsheet tools such as Microsoft Excel and Google Sheets.",[12,1282,1283],{},"This allows AI to assist with:",[33,1285,1286,1289,1292,1295],{},[36,1287,1288],{},"Data analysis",[36,1290,1291],{},"Formula creation",[36,1293,1294],{},"Automated reports",[36,1296,1297],{},"Data visualisation",[12,1299,1300],{},"These types of features make the model particularly attractive for businesses that work extensively with data.",[25,1302,1304],{"id":1303},"up-to-1-million-tokens-of-context","Up to 1 million tokens of context",[12,1306,1307],{},"Another major improvement is the increased context capacity.",[12,1309,1310,1311,1314],{},"GPT-5.4 can handle up to ",[39,1312,1313],{},"1 million tokens",", meaning the model can read and analyse much larger documents than before.",[12,1316,1317],{},"This makes it possible to:",[33,1319,1320,1323,1326],{},[36,1321,1322],{},"Analyse lengthy reports",[36,1324,1325],{},"Process large datasets",[36,1327,1328],{},"Work with entire projects in the same conversation",[12,1330,1331],{},"The larger context also means AI can better keep track of context in long workflows.",[25,1333,1335],{"id":1334},"ai-agents-and-automated-tasks","AI agents and automated tasks",[12,1337,1338,1339,1250],{},"GPT-5.4 is a step towards what is often called ",[39,1340,283],{},[12,1342,1343],{},"This means AI doesn't just answer questions but can also:",[33,1345,1346,1349,1352],{},[36,1347,1348],{},"Use applications",[36,1350,1351],{},"Perform tasks in programs",[36,1353,1354],{},"Automate workflows",[12,1356,1357],{},"In practice, AI can help with everything from analysing data to creating documents or code in other tools.",[25,1359,1361],{"id":1360},"gpt-54-vs-previous-models","GPT-5.4 vs previous models",[12,1363,1364],{},"Here are some of the most important differences between GPT-5.4 and previous models:",[1366,1367,1368,1384],"table",{},[1369,1370,1371],"thead",{},[1372,1373,1374,1378,1381],"tr",{},[1375,1376,1377],"th",{},"Feature",[1375,1379,1380],{},"GPT-4",[1375,1382,1383],{},"GPT-5.4",[1385,1386,1387,1399,1410,1421],"tbody",{},[1372,1388,1389,1393,1396],{},[1390,1391,1392],"td",{},"Reasoning",[1390,1394,1395],{},"Good",[1390,1397,1398],{},"Significantly better",[1372,1400,1401,1404,1407],{},[1390,1402,1403],{},"Context",[1390,1405,1406],{},"Smaller",[1390,1408,1409],{},"Up to 1M tokens",[1372,1411,1412,1415,1418],{},[1390,1413,1414],{},"Integrations",[1390,1416,1417],{},"Limited",[1390,1419,1420],{},"Excel, Sheets etc.",[1372,1422,1423,1426,1429],{},[1390,1424,1425],{},"Automation",[1390,1427,1428],{},"Basic",[1390,1430,283],{},[12,1432,1433],{},"The result is an AI model better suited for professional work and complex tasks.",[25,1435,1437],{"id":1436},"how-gpt-54-can-be-used","How GPT-5.4 can be used",[12,1439,1440],{},"The new model can be used across many areas, for example:",[12,1442,1443],{},[39,1444,1261],{},[33,1446,1447,1450],{},[36,1448,1449],{},"Writing and analysing code",[36,1451,1452],{},"Debugging programs",[12,1454,1455],{},[39,1456,1288],{},[33,1458,1459,1462],{},[36,1460,1461],{},"Analysing large datasets",[36,1463,1464],{},"Creating reports and dashboards",[12,1466,1467],{},[39,1468,1469],{},"Content creation",[33,1471,1472,1475],{},[36,1473,1474],{},"Writing articles",[36,1476,1477],{},"Creating marketing copy",[12,1479,1480],{},[39,1481,1482],{},"Business work",[33,1484,1485,1488],{},[36,1486,1487],{},"Document management",[36,1489,1236],{},[25,1491,1493],{"id":1492},"summary","Summary",[12,1495,1496],{},"GPT-5.4 is a major step forward for artificial intelligence. With better reasoning, larger context and integration with work tools, the model becomes more useful in both everyday life and work.",[12,1498,1499],{},"The key improvements are:",[33,1501,1502,1507,1511,1515],{},[36,1503,1504,1506],{},[39,1505,1249],{}," for better problem solving",[36,1508,1509],{},[39,1510,1277],{},[36,1512,1513],{},[39,1514,1304],{},[36,1516,1517],{},[39,1518,1519],{},"Support for AI agents and automated workflows",[12,1521,1522],{},"The development clearly shows that AI is moving from simple chatbots to powerful digital work assistants.",[25,1524,1526],{"id":1525},"frequently-asked-questions-about-gpt-54","Frequently asked questions about GPT-5.4",[12,1528,1529,1532],{},[39,1530,1531],{},"How much does GPT-5.4 cost?","\nGPT-5.4 is available through ChatGPT Plus and Enterprise plans. The pricing model varies depending on usage and which plan you choose.",[12,1534,1535,1538],{},[39,1536,1537],{},"Can I use GPT-5.4 for free?","\nBasic features are available in the free version of ChatGPT, but the most advanced features like Thinking Mode and extended context require a paid plan.",[12,1540,1541,1544],{},[39,1542,1543],{},"Is GPT-5.4 better than Claude?","\nIt depends on the use case. GPT-5.4 has strong integration capabilities while Claude often excels in reasoning and coding. The choice depends on your specific needs.",[12,1546,1547,1550,1551,1555],{},[39,1548,1549],{},"Can GPT-5.4 replace an enterprise AI platform?","\nGPT-5.4 is a powerful tool, but lacks features that many businesses need — such as data security with Swedish jurisdiction, RAG with source citations, and integration with internal systems. A dedicated enterprise platform like ",[1050,1552,1554],{"href":1553},"\u002Fprodukt-noda","Noda"," is better suited for business-critical needs.",[12,1557,1558,1561],{},[39,1559,1560],{},"How does GPT-5.4 differ from GPT-4?","\nThe biggest differences are better reasoning (Thinking Mode), a much larger context window (1M tokens), deeper integrations with work tools and support for AI agents that can perform tasks automatically.",[12,1563,1564,1567],{},[39,1565,1566],{},"Is GPT-5.4 safe to use for businesses?","\nOpenAI has improved security, but data is sent to servers in the US. Businesses with high requirements for data security and GDPR should consider European alternatives or complement with a platform that handles data within the EU\u002FSweden.",{"title":258,"searchDepth":259,"depth":260,"links":1569},[1570,1571,1572,1573,1574,1575,1576,1577,1578],{"id":1218,"depth":259,"text":1219},{"id":1242,"depth":259,"text":1243},{"id":1276,"depth":259,"text":1277},{"id":1303,"depth":259,"text":1304},{"id":1334,"depth":259,"text":1335},{"id":1360,"depth":259,"text":1361},{"id":1436,"depth":259,"text":1437},{"id":1492,"depth":259,"text":1493},{"id":1525,"depth":259,"text":1526},"2026-03-06","GPT-5.4 introduces better reasoning, AI agents and powerful data analysis. Here's everything you need to know about the new ChatGPT model.","\u002Fimages\u002Fnews\u002Fopen-ai-gpt-5.4.jpeg",{},"\u002Fnews\u002F2026-03-06-gpt-5-4-whats-new",{"title":1204,"description":1580},"news\u002F2026-03-06-gpt-5-4-whats-new",[360,1383,1198,1195,283,1587],"LLM","gpt-5-4-whats-new","xDl3gHXk9xUqNCCv-OJrWvHu_129drjhxbGQdw9J5MY",{"id":1591,"title":1592,"author":7,"body":1593,"category":271,"date":1962,"description":1963,"extension":274,"image":1964,"meta":1965,"navigation":277,"path":1966,"readTime":279,"seo":1967,"stem":1968,"tags":1969,"translationId":1974,"__hash__":1975},"news\u002Fnews\u002F2026-03-03-ai-compliance-eu-ai-act.md","AI Compliance Under the EU AI Act: How to Get Ready",{"type":9,"value":1594,"toc":1931},[1595,1598,1601,1605,1608,1634,1638,1641,1646,1649,1663,1667,1670,1702,1705,1709,1716,1720,1723,1727,1730,1734,1737,1741,1744,1748,1751,1755,1758,1762,1765,1769,1772,1776,1779,1783,1786,1790,1793,1819,1823,1826,1830,1876,1880,1884,1887,1891,1894,1898,1901,1905,1908,1912,1915,1917,1920,1923],[12,1596,1597],{},"With the EU AI Act (Regulation (EU) 2024\u002F1689) in place, AI compliance has become a concrete reality for organisations across Europe. The regulation began its phased application during 2025, and in 2026 the most extensive requirements take effect – particularly for high-risk AI systems.",[12,1599,1600],{},"This article provides a practical walkthrough of what compliance means, which requirements apply, and how your organisation can prepare.",[25,1602,1604],{"id":1603},"what-does-ai-compliance-mean","What Does AI Compliance Mean?",[12,1606,1607],{},"AI compliance is about ensuring that AI systems developed, deployed, or used within the EU meet the requirements set by the AI Act. This involves:",[33,1609,1610,1616,1622,1628],{},[36,1611,1612,1615],{},[39,1613,1614],{},"Risk classification"," – identifying whether your AI system falls under unacceptable, high, limited, or minimal risk.",[36,1617,1618,1621],{},[39,1619,1620],{},"Documentation and transparency"," – being able to demonstrate how the system works, what data it was trained on, and what decisions it makes.",[36,1623,1624,1627],{},[39,1625,1626],{},"Human oversight"," – ensuring that human control and intervention are possible.",[36,1629,1630,1633],{},[39,1631,1632],{},"Data quality and governance"," – ensuring that training and test data meet quality standards.",[25,1635,1637],{"id":1636},"risk-classification-the-foundation-of-your-obligations","Risk Classification: The Foundation of Your Obligations",[12,1639,1640],{},"The AI Act is built on a risk-based model with four levels:",[1642,1643,1645],"h3",{"id":1644},"unacceptable-risk-prohibited","Unacceptable Risk (Prohibited)",[12,1647,1648],{},"Certain AI uses have been completely prohibited since February 2025. These include:",[33,1650,1651,1654,1657,1660],{},[36,1652,1653],{},"Social scoring by public authorities",[36,1655,1656],{},"Real-time remote biometric identification in public spaces (with limited exceptions)",[36,1658,1659],{},"AI that exploits vulnerabilities of specific groups",[36,1661,1662],{},"Subliminal manipulation that may cause harm",[1642,1664,1666],{"id":1665},"high-risk","High Risk",[12,1668,1669],{},"High-risk AI systems face the most extensive requirements. This includes systems used in:",[33,1671,1672,1678,1684,1690,1696],{},[36,1673,1674,1677],{},[39,1675,1676],{},"Critical infrastructure"," – energy, transport, water",[36,1679,1680,1683],{},[39,1681,1682],{},"Education"," – admissions, grading",[36,1685,1686,1689],{},[39,1687,1688],{},"Employment"," – recruitment, dismissal, performance evaluation",[36,1691,1692,1695],{},[39,1693,1694],{},"Law enforcement"," – risk assessment, evidence analysis",[36,1697,1698,1701],{},[39,1699,1700],{},"Migration and border control"," – application processing, risk profiling",[12,1703,1704],{},"Requirements include risk management systems, data quality standards, technical documentation, logging, transparency information, human oversight, and cybersecurity.",[1642,1706,1708],{"id":1707},"limited-risk","Limited Risk",[12,1710,1711,1712,1715],{},"Systems with limited risk primarily face ",[39,1713,1714],{},"transparency obligations"," – for example, chatbots and deepfakes must be clearly labelled so users know they are interacting with AI or that the content is AI-generated.",[1642,1717,1719],{"id":1718},"minimal-risk","Minimal Risk",[12,1721,1722],{},"Most AI systems fall under minimal risk and have no specific requirements, although voluntary codes of conduct are encouraged.",[25,1724,1726],{"id":1725},"high-risk-ai-the-key-requirements-in-practice","High-Risk AI: The Key Requirements in Practice",[12,1728,1729],{},"If your system is classified as high-risk, you need to meet a number of requirements that become fully applicable in August 2026:",[1642,1731,1733],{"id":1732},"_1-risk-management-system-art-9","1. Risk Management System (Art. 9)",[12,1735,1736],{},"A risk management system must be established, implemented, and documented throughout the system's lifecycle. It must identify and analyse known and foreseeable risks, and evaluate risks that may arise during intended use and reasonably foreseeable misuse.",[1642,1738,1740],{"id":1739},"_2-data-quality-art-10","2. Data Quality (Art. 10)",[12,1742,1743],{},"Training, validation, and test data must meet quality criteria regarding relevance, representativeness, accuracy, and completeness. Bias in data must be identified and addressed.",[1642,1745,1747],{"id":1746},"_3-technical-documentation-art-11","3. Technical Documentation (Art. 11)",[12,1749,1750],{},"Documentation must be drawn up before the system is placed on the market and kept up to date. It must provide sufficient information for authorities to assess the system's conformity.",[1642,1752,1754],{"id":1753},"_4-logging-art-12","4. Logging (Art. 12)",[12,1756,1757],{},"High-risk AI systems must have automatic logging that enables traceability. Logs must be retained for an appropriate period.",[1642,1759,1761],{"id":1760},"_5-transparency-and-information-art-13","5. Transparency and Information (Art. 13)",[12,1763,1764],{},"Users must receive clear information about the system's capabilities, limitations, intended purpose, and interpretability.",[1642,1766,1768],{"id":1767},"_6-human-oversight-art-14","6. Human Oversight (Art. 14)",[12,1770,1771],{},"The system must be designed so that it can be effectively overseen by humans. There must be the ability to intervene, interrupt, or correct.",[1642,1773,1775],{"id":1774},"_7-cybersecurity-art-15","7. Cybersecurity (Art. 15)",[12,1777,1778],{},"The system must achieve an appropriate level of security, resilience, and accuracy with regard to its intended purpose.",[25,1780,1782],{"id":1781},"the-connection-to-gdpr-and-nis2","The Connection to GDPR and NIS2",[12,1784,1785],{},"AI compliance does not exist in a vacuum. Two other key regulatory frameworks have a direct impact:",[1642,1787,1789],{"id":1788},"gdpr","GDPR",[12,1791,1792],{},"AI systems that process personal data must continue to comply with GDPR. This includes:",[33,1794,1795,1801,1807,1813],{},[36,1796,1797,1800],{},[39,1798,1799],{},"Legal basis"," for data processing (e.g. consent, legitimate interest)",[36,1802,1803,1806],{},[39,1804,1805],{},"Data Protection Impact Assessment (DPIA)"," for high-risk processing",[36,1808,1809,1812],{},[39,1810,1811],{},"Right to explanation"," in automated decision-making (Art. 22 GDPR)",[36,1814,1815,1818],{},[39,1816,1817],{},"Privacy by design"," in system development",[1642,1820,1822],{"id":1821},"nis2","NIS2",[12,1824,1825],{},"Organisations covered by the NIS2 Directive already have obligations around cybersecurity and incident reporting. The AI Act's cybersecurity requirements for high-risk AI overlap with NIS2, creating an opportunity for coordinated compliance.",[25,1827,1829],{"id":1828},"timeline-when-does-what-apply","Timeline: When Does What Apply?",[1366,1831,1832,1842],{},[1369,1833,1834],{},[1372,1835,1836,1839],{},[1375,1837,1838],{},"Date",[1375,1840,1841],{},"What Happens",[1385,1843,1844,1852,1860,1868],{},[1372,1845,1846,1849],{},[1390,1847,1848],{},"February 2025",[1390,1850,1851],{},"Prohibition on unacceptable-risk AI takes effect",[1372,1853,1854,1857],{},[1390,1855,1856],{},"August 2025",[1390,1858,1859],{},"Rules for GPAI models (General Purpose AI) apply",[1372,1861,1862,1865],{},[1390,1863,1864],{},"August 2026",[1390,1866,1867],{},"Full requirements for high-risk AI systems take effect",[1372,1869,1870,1873],{},[1390,1871,1872],{},"August 2027",[1390,1874,1875],{},"Requirements for high-risk AI embedded in regulated products",[25,1877,1879],{"id":1878},"how-to-prepare-your-organisation","How to Prepare Your Organisation",[1642,1881,1883],{"id":1882},"step-1-inventory-your-ai-systems","Step 1: Inventory Your AI Systems",[12,1885,1886],{},"Map out which AI systems you develop, procure, or use. Identify which may be classified as high-risk.",[1642,1888,1890],{"id":1889},"step-2-conduct-a-risk-assessment","Step 2: Conduct a Risk Assessment",[12,1892,1893],{},"Analyse the risk level for each system. Document the assessment that led to the classification.",[1642,1895,1897],{"id":1896},"step-3-establish-a-compliance-framework","Step 3: Establish a Compliance Framework",[12,1899,1900],{},"Implement processes for documentation, risk management, data quality, human oversight, and incident management.",[1642,1902,1904],{"id":1903},"step-4-connect-with-gdpr-and-nis2","Step 4: Connect with GDPR and NIS2",[12,1906,1907],{},"Review existing processes for data protection and cybersecurity. Identify synergies and gaps.",[1642,1909,1911],{"id":1910},"step-5-train-your-organisation","Step 5: Train Your Organisation",[12,1913,1914],{},"Ensure that key personnel – developers, legal teams, management – understand the requirements and their role in compliance. The AI Act requires \"AI literacy\" (Art. 4).",[25,1916,1493],{"id":1492},[12,1918,1919],{},"The EU AI Act changes the playing field for everyone working with AI in Europe. Compliance is not just a legal obligation – it is an opportunity to build trust, reduce risk, and create sustainable AI systems.",[12,1921,1922],{},"Organisations that start preparing now – with risk classification, documentation, and integrated processes – will have a clear advantage when the full requirements take effect in August 2026.",[834,1924,1925],{},[12,1926,1927,1930],{},[39,1928,1929],{},"Want to know where you stand?"," Walma helps organisations map their AI usage, assess risk levels, and build a compliance framework that meets the EU AI Act, GDPR, and NIS2 – in practice, not just on paper.",{"title":258,"searchDepth":259,"depth":260,"links":1932},[1933,1934,1940,1949,1953,1954,1961],{"id":1603,"depth":259,"text":1604},{"id":1636,"depth":259,"text":1637,"children":1935},[1936,1937,1938,1939],{"id":1644,"depth":260,"text":1645},{"id":1665,"depth":260,"text":1666},{"id":1707,"depth":260,"text":1708},{"id":1718,"depth":260,"text":1719},{"id":1725,"depth":259,"text":1726,"children":1941},[1942,1943,1944,1945,1946,1947,1948],{"id":1732,"depth":260,"text":1733},{"id":1739,"depth":260,"text":1740},{"id":1746,"depth":260,"text":1747},{"id":1753,"depth":260,"text":1754},{"id":1760,"depth":260,"text":1761},{"id":1767,"depth":260,"text":1768},{"id":1774,"depth":260,"text":1775},{"id":1781,"depth":259,"text":1782,"children":1950},[1951,1952],{"id":1788,"depth":260,"text":1789},{"id":1821,"depth":260,"text":1822},{"id":1828,"depth":259,"text":1829},{"id":1878,"depth":259,"text":1879,"children":1955},[1956,1957,1958,1959,1960],{"id":1882,"depth":260,"text":1883},{"id":1889,"depth":260,"text":1890},{"id":1896,"depth":260,"text":1897},{"id":1903,"depth":260,"text":1904},{"id":1910,"depth":260,"text":1911},{"id":1492,"depth":259,"text":1493},"2026-03-03","The EU AI Act introduces new requirements for organisations that develop or use AI. Here we walk through what compliance means in practice – from risk classification and high-risk requirements to connections with GDPR and NIS2.","\u002Fimages\u002Fnews\u002Feu-compliance.jpeg",{},"\u002Fnews\u002F2026-03-03-ai-compliance-eu-ai-act",{"title":1592,"description":1963},"news\u002F2026-03-03-ai-compliance-eu-ai-act",[360,1970,1971,1972,1973,1789,1822],"EU AI Act","compliance","high-risk AI","AI regulation","ai-compliance-eu-ai-act","p7mgRcmR1TAslsXUD5udg5qUc1-SDY1YbQ6LBwwf11I",{"id":1977,"title":1978,"author":505,"body":1979,"category":271,"date":2505,"description":2506,"extension":274,"image":2507,"meta":2508,"navigation":277,"path":2509,"readTime":2510,"seo":2511,"stem":2512,"tags":2513,"translationId":2516,"__hash__":2517},"news\u002Fnews\u002F2026-03-01-eu-ai-act-timeline-2026-2027.md","The EU AI Act: History, What Applies Now – and What Happens in 2026–2027",{"type":9,"value":1980,"toc":2467},[1981,1984,1987,1998,2002,2005,2013,2017,2021,2024,2028,2031,2035,2038,2042,2045,2049,2052,2058,2062,2065,2069,2072,2076,2079,2099,2102,2106,2109,2113,2116,2120,2123,2127,2134,2160,2166,2170,2176,2187,2191,2318,2322,2325,2329,2332,2336,2339,2343,2346,2384,2387,2391,2395,2404,2408,2411,2415,2418,2422,2425,2429,2432,2436,2439,2443,2446,2450,2453,2457,2460,2464],[12,1982,1983],{},"The EU's AI regulation, commonly known as the AI Act, is the world's first broad \"horizontal\" regulatory framework governing the development, market introduction, and use of AI systems in the EU. It is built on a risk-based model: certain AI uses are prohibited, others require transparency or extensive requirements (particularly for high-risk AI), and general-purpose AI models have their own obligations.",[12,1985,1986],{},"This article provides:",[33,1988,1989,1992,1995],{},[36,1990,1991],{},"a historical overview (how we got here),",[36,1993,1994],{},"a practical walkthrough of what applies, and",[36,1996,1997],{},"a clear timeline for 2026 and 2027.",[25,1999,2001],{"id":2000},"brief-background-what-is-the-ai-act","Brief Background: What Is the AI Act?",[12,2003,2004],{},"The AI Act is Regulation (EU) 2024\u002F1689 (\"Artificial Intelligence Act\"), published in the Official Journal of the EU in the summer of 2024. It creates a common framework within the EU to:",[33,2006,2007,2010],{},[36,2008,2009],{},"protect health, safety, and fundamental rights,",[36,2011,2012],{},"while enabling innovation through measures such as sandboxes and support mechanisms.",[25,2014,2016],{"id":2015},"the-history-key-milestones-20212024","The History: Key Milestones (2021–2024)",[1642,2018,2020],{"id":2019},"_1-the-european-commissions-proposal-2021","1. The European Commission's Proposal (2021)",[12,2022,2023],{},"The journey began when the European Commission presented a formal proposal for an AI regulation in April 2021 (COM(2021) 206).",[1642,2025,2027],{"id":2026},"_2-negotiations-and-provisional-agreement-2023","2. Negotiations and \"Provisional Agreement\" (2023)",[12,2029,2030],{},"After intensive trilogue negotiations, the Council and Parliament reached a provisional political agreement in December 2023.",[1642,2032,2034],{"id":2033},"_3-european-parliament-approval-march-2024","3. European Parliament Approval (March 2024)",[12,2036,2037],{},"The European Parliament adopted the text at its plenary session on 13 March 2024.",[1642,2039,2041],{"id":2040},"_4-the-councils-final-green-light-may-2024","4. The Council's Final \"Green Light\" (May 2024)",[12,2043,2044],{},"The EU Council gave final approval on 21 May 2024, effectively concluding the legislative process.",[1642,2046,2048],{"id":2047},"_5-publication-and-entry-into-force-summer-2024","5. Publication and Entry into Force (Summer 2024)",[12,2050,2051],{},"The regulation was published in the Official Journal of the EU and entered into force as EU law in the summer of 2024.",[12,2053,2054,2057],{},[39,2055,2056],{},"Important:"," entry into force does not mean all obligations apply immediately – the regulation has a phased application.",[25,2059,2061],{"id":2060},"what-applies-in-practice-risk-levels-and-key-requirements","What Applies in Practice: Risk Levels and Key Requirements",[12,2063,2064],{},"The AI Act is typically summarised across four levels:",[1642,2066,2068],{"id":2067},"prohibited-ai-practices","Prohibited AI Practices",[12,2070,2071],{},"Certain uses are prohibited, such as certain forms of manipulation, social scoring, and certain biometric scenarios. The European Commission has also published guidelines on prohibited practices to support uniform interpretation.",[1642,2073,2075],{"id":2074},"high-risk-ai","High-Risk AI",[12,2077,2078],{},"AI systems in certain areas (e.g. critical infrastructure, education, recruitment, credit scoring, etc.) are classified as high-risk and require, among other things:",[33,2080,2081,2084,2087,2090,2093,2096],{},[36,2082,2083],{},"risk management,",[36,2085,2086],{},"data quality,",[36,2088,2089],{},"technical documentation,",[36,2091,2092],{},"logging,",[36,2094,2095],{},"human oversight,",[36,2097,2098],{},"cybersecurity and robustness,",[12,2100,2101],{},"…plus obligations for both providers and deployers.",[1642,2103,2105],{"id":2104},"transparency-requirements","Transparency Requirements",[12,2107,2108],{},"Certain AI functions are subject to clear transparency requirements – for example when people interact with AI or in the case of synthetically generated content. The transparency rules in Article 50 take effect later in the rollout.",[1642,2110,2112],{"id":2111},"general-purpose-ai-gpai","General-Purpose AI \u002F GPAI",[12,2114,2115],{},"The regulation contains specific rules for general-purpose AI models. This is crucial for providers of broad models and for the ecosystem around them.",[25,2117,2119],{"id":2118},"the-timeline-what-happens-in-2026-and-2027","The Timeline: What Happens in 2026 and 2027?",[12,2121,2122],{},"The European Commission's AI Act Service Desk describes that the law is applied progressively with \"full roll-out\" by 2 August 2027 at the latest.",[1642,2124,2126],{"id":2125},"during-2026-the-majority-of-rules-take-effect","During 2026: \"The Majority of Rules\" Take Effect",[12,2128,2129,2130,2133],{},"The major milestone is ",[39,2131,2132],{},"2 August 2026",":",[33,2135,2136,2142,2148,2154],{},[36,2137,2138,2141],{},[39,2139,2140],{},"The majority of rules"," begin to apply and supervision\u002Fenforcement starts at national and EU level.",[36,2143,2144,2147],{},[39,2145,2146],{},"High-risk AI in Annex III"," becomes covered (e.g. high-risk uses across multiple societal sectors).",[36,2149,2150,2153],{},[39,2151,2152],{},"Transparency rules"," (Article 50) take effect.",[36,2155,2156,2159],{},[39,2157,2158],{},"Innovation support"," takes effect, and member states must have at least one regulatory AI sandbox per country established.",[12,2161,2162,2165],{},[39,2163,2164],{},"Implications for businesses in 2026:"," This is when many organisations \"go live\" with compliance: classification, risk and quality processes, documentation, supply chain management, incident handling, and governance need to be in place for systems covered by Annex III.",[1642,2167,2169],{"id":2168},"during-2027-rules-for-high-risk-ai-in-regulated-products-take-effect","During 2027: Rules for High-Risk AI in Regulated Products Take Effect",[12,2171,2172,2173,2133],{},"The next major milestone is ",[39,2174,2175],{},"2 August 2027",[33,2177,2178,2184],{},[36,2179,2180,2183],{},[39,2181,2182],{},"Rules for high-risk AI embedded in regulated products"," begin to apply.",[36,2185,2186],{},"This typically concerns AI that is part of product-regulated areas (e.g. certain machinery, vehicles, and medical devices – depending on classification and which EU regulatory framework applies to the product).",[25,2188,2190],{"id":2189},"table-history-what-applies-20262027","Table: History + What Applies 2026–2027",[1366,2192,2193,2207],{},[1369,2194,2195],{},[1372,2196,2197,2199,2201,2204],{},[1375,2198,1838],{},[1375,2200,1841],{},[1375,2202,2203],{},"Who Is Most Affected",[1375,2205,2206],{},"Practical Implication",[1385,2208,2209,2223,2236,2249,2262,2276,2290,2304],{},[1372,2210,2211,2214,2217,2220],{},[1390,2212,2213],{},"21 Apr 2021",[1390,2215,2216],{},"Commission's proposal (COM(2021) 206)",[1390,2218,2219],{},"Everyone",[1390,2221,2222],{},"Start of the legislative process",[1372,2224,2225,2228,2231,2233],{},[1390,2226,2227],{},"8 Dec 2023",[1390,2229,2230],{},"Provisional political agreement",[1390,2232,2219],{},[1390,2234,2235],{},"Text \"settles\" after trilogue",[1372,2237,2238,2241,2244,2246],{},[1390,2239,2240],{},"13 Mar 2024",[1390,2242,2243],{},"European Parliament adopts the text",[1390,2245,2219],{},[1390,2247,2248],{},"Democratic final step in EP",[1372,2250,2251,2254,2257,2259],{},[1390,2252,2253],{},"21 May 2024",[1390,2255,2256],{},"Council gives final green light",[1390,2258,2219],{},[1390,2260,2261],{},"The law is formally adopted",[1372,2263,2264,2267,2270,2273],{},[1390,2265,2266],{},"02 Feb 2025",[1390,2268,2269],{},"Definitions\u002FAI literacy + prohibitions take effect",[1390,2271,2272],{},"Most organisations",[1390,2274,2275],{},"Requirements on AI competence and avoiding prohibited practices",[1372,2277,2278,2281,2284,2287],{},[1390,2279,2280],{},"02 Aug 2025",[1390,2282,2283],{},"GPAI rules + governance must be in place",[1390,2285,2286],{},"Model\u002Fplatform providers + authorities",[1390,2288,2289],{},"Requirements on GPAI providers; national authorities and EU bodies established",[1372,2291,2292,2295,2298,2301],{},[1390,2293,2294],{},"02 Aug 2026",[1390,2296,2297],{},"Majority of rules + enforcement starts",[1390,2299,2300],{},"High-risk actors, most larger orgs",[1390,2302,2303],{},"High-risk Annex III takes effect, transparency (Art. 50), sandboxes, etc.",[1372,2305,2306,2309,2312,2315],{},[1390,2307,2308],{},"02 Aug 2027",[1390,2310,2311],{},"High-risk AI in regulated products takes effect",[1390,2313,2314],{},"Manufacturers + product ecosystem",[1390,2316,2317],{},"Product-embedded AI gets full high-risk requirements in regulated product areas",[25,2319,2321],{"id":2320},"can-the-timeline-change","Can the Timeline Change?",[12,2323,2324],{},"There are two \"moving parts\" to watch:",[1642,2326,2328],{"id":2327},"the-commissions-signal-regarding-standards","The Commission's Signal Regarding Standards",[12,2330,2331],{},"The AI Act Service Desk states that the Commission (in the context of a \"Digital Omnibus package\") has proposed linking the application of rules for high-risk AI to the availability of support tools, including harmonised standards. This suggests that practical application may be affected by how quickly standards and support materials are finalised.",[1642,2333,2335],{"id":2334},"political-discussion-on-adjustments","Political Discussion on Adjustments",[12,2337,2338],{},"There have been media reports that parts of the AI Act may be subject to adjustments or delays under pressure from industry and geopolitics, particularly related to \"high-risk\" parts and sanctions logic. This is not the same as the law \"not applying\", but it is a reason to follow the European Commission's official updates.",[25,2340,2342],{"id":2341},"practical-checklist-for-2026","Practical Checklist for 2026",[12,2344,2345],{},"If your organisation is affected by the AI Act – here is what businesses typically need to do during 2025–2026:",[931,2347,2348,2354,2360,2366,2372,2378],{},[36,2349,2350,2353],{},[39,2351,2352],{},"Inventory AI uses"," – including embedded models, third-party tools, and automated decisions.",[36,2355,2356,2359],{},[39,2357,2358],{},"Classify risk"," – prohibited, high-risk Annex III, transparency, or other.",[36,2361,2362,2365],{},[39,2363,2364],{},"Governance & roles"," – AI policy, ownership, and supplier management.",[36,2367,2368,2371],{},[39,2369,2370],{},"Documentation & traceability"," – technical documentation, logging, and data sources.",[36,2373,2374,2377],{},[39,2375,2376],{},"Incident and change processes"," – procedures for when the model, data, or usage changes.",[36,2379,2380,2383],{},[39,2381,2382],{},"AI literacy programme"," – training adapted to roles: product, IT, legal, procurement, and operations.",[12,2385,2386],{},"The exact requirements depend on role: provider, importer, distributor, or deployer.",[25,2388,2390],{"id":2389},"faq","FAQ",[1642,2392,2394],{"id":2393},"when-does-the-ai-act-really-take-effect","When does the AI Act \"really\" take effect?",[12,2396,2397,2398,2400,2401,2403],{},"It is already in force in the EU law sense, but is applied in phases. The key dates for broad business compliance are ",[39,2399,2132],{}," (majority of rules) and ",[39,2402,2175],{}," (high-risk AI in regulated products).",[1642,2405,2407],{"id":2406},"which-businesses-are-affected-by-the-ai-act","Which businesses are affected by the AI Act?",[12,2409,2410],{},"All organisations that develop, distribute, or use AI systems within the EU are affected. This also applies to companies outside the EU whose AI systems are used within the union. Particularly relevant are actors in high-risk sectors such as healthcare, education, recruitment, credit scoring, and critical infrastructure.",[1642,2412,2414],{"id":2413},"what-counts-as-high-risk-ai","What counts as high-risk AI?",[12,2416,2417],{},"High-risk AI is defined in Annex III and includes systems used in areas such as biometric identification, critical infrastructure, education and vocational training, employment and personnel management, access to public services, law enforcement, migration management, and the justice system. AI systems embedded in already-regulated products (e.g. medical devices, vehicles) may also be classified as high-risk.",[1642,2419,2421],{"id":2420},"what-is-most-important-during-2026","What is most important during 2026?",[12,2423,2424],{},"That high-risk AI in Annex III becomes subject to the regulation and that enforcement begins. Organisations need to have their risk and quality processes, documentation, and governance in place by 2 August 2026 at the latest.",[1642,2426,2428],{"id":2427},"what-is-the-difference-in-2027-compared-to-2026","What is the difference in 2027 compared to 2026?",[12,2430,2431],{},"2027 specifically targets high-risk AI embedded in regulated products – for example AI components in machinery, vehicles, or medical devices that are already covered by other EU product legislation.",[1642,2433,2435],{"id":2434},"what-do-the-gpai-rules-mean","What do the GPAI rules mean?",[12,2437,2438],{},"General-Purpose AI models (GPAI) have their own obligations that took effect on 2 August 2025. Providers of such models must, among other things, provide technical documentation, comply with copyright rules, and publish summaries of training data. GPAI models with \"systemic risk\" have additional requirements.",[1642,2440,2442],{"id":2441},"what-happens-if-you-dont-comply-with-the-ai-act","What happens if you don't comply with the AI Act?",[12,2444,2445],{},"Sanctions vary depending on the nature of the violation. Fines can amount to EUR 35 million or 7% of global annual turnover for violations of prohibited AI practices, and up to EUR 15 million or 3% of turnover for other violations. National supervisory authorities are responsible for enforcement.",[1642,2447,2449],{"id":2448},"does-the-ai-act-apply-to-swedish-public-authorities","Does the AI Act apply to Swedish public authorities?",[12,2451,2452],{},"Yes, the AI Act applies to all actors that develop or use AI systems – including the public sector and government authorities. Swedish authorities that use AI systems for decision-making, assessments, or automated processes must follow the same risk classification and requirements as private actors.",[1642,2454,2456],{"id":2455},"do-we-need-an-ai-policy","Do we need an AI policy?",[12,2458,2459],{},"It is not an explicit requirement in the regulation to have a separate \"AI policy\", but in practice organisations need documented governance, role allocation, and processes to meet the requirements. An AI policy is often the simplest way to consolidate this. The AI literacy requirement (Article 4) also means that personnel working with AI systems must have sufficient competence.",[1642,2461,2463],{"id":2462},"how-can-walma-help-with-the-ai-act","How can Walma help with the AI Act?",[12,2465,2466],{},"Walma offers AI solutions designed with compliance in mind. Our platform Noda provides full traceability, source references, and Swedish data sovereignty – three key requirements in the AI Act. We also offer AI training and workshops that help organisations build the AI literacy the regulation requires.",{"title":258,"searchDepth":259,"depth":260,"links":2468},[2469,2470,2477,2483,2487,2488,2492,2493],{"id":2000,"depth":259,"text":2001},{"id":2015,"depth":259,"text":2016,"children":2471},[2472,2473,2474,2475,2476],{"id":2019,"depth":260,"text":2020},{"id":2026,"depth":260,"text":2027},{"id":2033,"depth":260,"text":2034},{"id":2040,"depth":260,"text":2041},{"id":2047,"depth":260,"text":2048},{"id":2060,"depth":259,"text":2061,"children":2478},[2479,2480,2481,2482],{"id":2067,"depth":260,"text":2068},{"id":2074,"depth":260,"text":2075},{"id":2104,"depth":260,"text":2105},{"id":2111,"depth":260,"text":2112},{"id":2118,"depth":259,"text":2119,"children":2484},[2485,2486],{"id":2125,"depth":260,"text":2126},{"id":2168,"depth":260,"text":2169},{"id":2189,"depth":259,"text":2190},{"id":2320,"depth":259,"text":2321,"children":2489},[2490,2491],{"id":2327,"depth":260,"text":2328},{"id":2334,"depth":260,"text":2335},{"id":2341,"depth":259,"text":2342},{"id":2389,"depth":259,"text":2390,"children":2494},[2495,2496,2497,2498,2499,2500,2501,2502,2503,2504],{"id":2393,"depth":260,"text":2394},{"id":2406,"depth":260,"text":2407},{"id":2413,"depth":260,"text":2414},{"id":2420,"depth":260,"text":2421},{"id":2427,"depth":260,"text":2428},{"id":2434,"depth":260,"text":2435},{"id":2441,"depth":260,"text":2442},{"id":2448,"depth":260,"text":2449},{"id":2455,"depth":260,"text":2456},{"id":2462,"depth":260,"text":2463},"2026-03-01","The EU AI Act is the world's first comprehensive regulation for AI. Here we cover the history, risk levels, what applies today, and the full timeline for 2026–2027.","\u002Fimages\u002Fnews\u002Feu-act-walma.jpg",{},"\u002Fnews\u002F2026-03-01-eu-ai-act-timeline-2026-2027","10 min read",{"title":1978,"description":2506},"news\u002F2026-03-01-eu-ai-act-timeline-2026-2027",[360,1970,1971,2514,1972,2515],"regulation","GPAI","eu-ai-act-timeline","vqqwNj_498GPN2MjIOjvAwZcW4vY6Zc4JcwGFf0o71Q",{"id":2519,"title":2520,"author":505,"body":2521,"category":271,"date":2715,"description":2716,"extension":274,"image":2717,"meta":2718,"navigation":277,"path":2719,"readTime":2720,"seo":2721,"stem":2722,"tags":2723,"translationId":2727,"__hash__":2728},"news\u002Fnews\u002F2026-01-11-how-rag-works.md","How RAG Works – The Foundation of Reliable Enterprise AI",{"type":9,"value":2522,"toc":2694},[2523,2526,2531,2534,2537,2541,2544,2547,2550,2554,2557,2560,2563,2567,2570,2573,2576,2580,2583,2586,2589,2593,2596,2599,2602,2606,2610,2613,2617,2620,2624,2627,2631,2634,2638,2641,2645,2648,2652,2655,2659,2662,2666,2669,2673,2676,2678,2681,2684,2688,2691],[12,2524,2525],{},"Generative AI has quickly found its way into many organisations. The possibilities are obvious, but before long a more fundamental question arises:",[12,2527,2528],{},[39,2529,2530],{},"How can AI provide answers that are actually based on our own information – and that can be trusted?",[12,2532,2533],{},"For most organisations that want to use AI in a serious way, the answer lands on RAG, or Retrieval-Augmented Generation. It is not a new model, but a way of building AI solutions that are grounded in real content and real sources.",[12,2535,2536],{},"This article explains what RAG is, why the technique has become so central to enterprise AI, and what it takes to make it work in practice.",[25,2538,2540],{"id":2539},"what-does-rag-actually-mean","What Does RAG Actually Mean?",[12,2542,2543],{},"RAG is an architecture where a generative AI model is not left alone with its training data, but is given access to the organisation's own information sources in connection with every question.",[12,2545,2546],{},"In practice, this means the AI first looks up relevant information in your documents and systems. That information is then used as a basis when the answer is formulated. The result is an answer that does not just sound reasonable, but that is actually built on content you recognise and can review.",[12,2548,2549],{},"The difference from a standalone language model is decisive. Without RAG, the model guesses based on probability. With RAG, it responds with support from concrete material.",[25,2551,2553],{"id":2552},"why-isnt-a-standard-language-model-enough","Why Isn't a Standard Language Model Enough?",[12,2555,2556],{},"A generative model without access to your own sources has no understanding of how your particular organisation works. It does not know internal concepts, local rules, or how decisions are typically interpreted in practice.",[12,2558,2559],{},"This means the answers often sound convincing, but are still wrong. On top of that, there is no traceability. It is not possible to see where the information came from or why a particular answer was given.",[12,2561,2562],{},"In contexts where AI is to be used as decision support, in internal processes, or in regulated environments, this quickly becomes a problem. This is where RAG makes a real difference.",[25,2564,2566],{"id":2565},"how-rag-works-in-practice","How RAG Works in Practice",[12,2568,2569],{},"For RAG to work, the organisation's information first needs to be made searchable in a smart way. Documents, guidelines, decisions, and other text are broken down into smaller parts and stored so the system can find content based on meaning, not just exact words.",[12,2571,2572],{},"When a user then asks a question – for example about what applies for overtime during on-call duty – the system searches for the parts of the material that best match the meaning of the question. This might involve several text passages from different documents, often supplemented with metadata such as source and date.",[12,2574,2575],{},"The generative model then receives both the question and the retrieved material. Its task is to formulate an answer that stays true to the content and that can be traced back to the sources. This is what makes RAG answers both understandable and auditable.",[25,2577,2579],{"id":2578},"what-you-gain-from-using-rag","What You Gain from Using RAG",[12,2581,2582],{},"The big advantage of RAG is not that the answers become longer or more advanced, but that they become more reliable. When the content is updated, the answers change too, without any model needing to be retrained. This makes the solution easier to manage over time.",[12,2584,2585],{},"Traceability is another important aspect. When you know which sources were used, it becomes possible to understand why a particular answer was given and to discover when the underlying material needs improvement.",[12,2587,2588],{},"This is why RAG is now used as the foundation in everything from internal AI assistants and customer support to legal support, HR, and knowledge management in larger organisations.",[25,2590,2592],{"id":2591},"challenges-that-are-often-underestimated","Challenges That Are Often Underestimated",[12,2594,2595],{},"Even though the technology is mature, RAG is rarely a \"plug-and-play\" project. In practice, the challenges are less about AI and more about information.",[12,2597,2598],{},"Many organisations have unstructured or outdated content. It can be unclear who owns what information, which versions are current, and who is responsible for updates. Without clear structure and governance, even a good RAG solution risks giving unreliable answers.",[12,2600,2601],{},"This is why RAG is just as much a question of information management and governance as it is about technology.",[25,2603,2605],{"id":2604},"faq-common-questions-about-rag","FAQ – Common Questions About RAG",[1642,2607,2609],{"id":2608},"is-the-ai-model-trained-on-our-internal-content","Is the AI model trained on our internal content?",[12,2611,2612],{},"No. The information is only used as temporary context when a question is answered. It is not saved and is not used to further train the model.",[1642,2614,2616],{"id":2615},"is-rag-the-same-as-search","Is RAG the same as search?",[12,2618,2619],{},"Not quite. A search engine finds documents. RAG uses search as a foundation, but goes further and formulates a coherent answer based on the content.",[1642,2621,2623],{"id":2622},"how-up-to-date-are-the-answers","How up-to-date are the answers?",[12,2625,2626],{},"The timeliness of the answers depends entirely on the sources. When documents are updated, the answers are affected immediately, without anything else needing to change.",[1642,2628,2630],{"id":2629},"can-you-control-which-sources-are-used","Can you control which sources are used?",[12,2632,2633],{},"Yes. It is possible to set both permissions and priorities so that different users or contexts have access to different parts of the information.",[1642,2635,2637],{"id":2636},"is-rag-secure-in-sensitive-environments","Is RAG secure in sensitive environments?",[12,2639,2640],{},"Yes, provided the solution is properly built. Access controls, logging, and clear governance are essential for secure operation.",[1642,2642,2644],{"id":2643},"what-types-of-documents-can-rag-handle","What types of documents can RAG handle?",[12,2646,2647],{},"RAG can work with virtually any text-based content: PDFs, Word documents, web pages, emails, internal wikis, and more. The key is that the content can be extracted and broken into meaningful chunks. Some solutions also support structured formats and metadata for more precise retrieval.",[1642,2649,2651],{"id":2650},"how-does-rag-differ-from-fine-tuning-a-model","How does RAG differ from fine-tuning a model?",[12,2653,2654],{},"Fine-tuning permanently alters a model's behaviour by training it on specific data. RAG leaves the model unchanged and instead provides relevant information at query time. This means RAG is easier to update, audit, and control – and your data never becomes part of the model itself.",[1642,2656,2658],{"id":2657},"can-rag-handle-multiple-languages","Can RAG handle multiple languages?",[12,2660,2661],{},"Yes. Modern embedding models and language models support multilingual content. A RAG system can retrieve Swedish documents and answer in English, or vice versa. The quality depends on the models used and how the content is indexed.",[1642,2663,2665],{"id":2664},"how-do-you-measure-whether-a-rag-solution-is-working-well","How do you measure whether a RAG solution is working well?",[12,2667,2668],{},"Key metrics include answer relevance, source accuracy, retrieval precision, and user satisfaction. In practice, this means regularly reviewing whether the system retrieves the right documents and whether the generated answers faithfully reflect the source material.",[1642,2670,2672],{"id":2671},"what-does-it-take-to-get-started-with-rag","What does it take to get started with RAG?",[12,2674,2675],{},"The most important step is having well-structured, up-to-date content. From a technical standpoint, you need an embedding model, a vector database, and a generative model. But the real work lies in content curation, access control, and establishing processes for keeping the knowledge base current.",[25,2677,1493],{"id":1492},[12,2679,2680],{},"RAG is not an experiment or an add-on at the end. It is the foundation for using generative AI responsibly in an organisation.",[12,2682,2683],{},"By connecting AI to your own information sources, you create solutions that can be trusted, followed up, and developed over time.",[25,2685,2687],{"id":2686},"next-steps","Next Steps",[12,2689,2690],{},"Want to see how RAG can be built on top of your existing information structure, with the right level of control and follow-up?",[12,2692,2693],{},"Walma helps organisations go from AI ideas to solutions that work in practice.",{"title":258,"searchDepth":259,"depth":260,"links":2695},[2696,2697,2698,2699,2700,2701,2713,2714],{"id":2539,"depth":259,"text":2540},{"id":2552,"depth":259,"text":2553},{"id":2565,"depth":259,"text":2566},{"id":2578,"depth":259,"text":2579},{"id":2591,"depth":259,"text":2592},{"id":2604,"depth":259,"text":2605,"children":2702},[2703,2704,2705,2706,2707,2708,2709,2710,2711,2712],{"id":2608,"depth":260,"text":2609},{"id":2615,"depth":260,"text":2616},{"id":2622,"depth":260,"text":2623},{"id":2629,"depth":260,"text":2630},{"id":2636,"depth":260,"text":2637},{"id":2643,"depth":260,"text":2644},{"id":2650,"depth":260,"text":2651},{"id":2657,"depth":260,"text":2658},{"id":2664,"depth":260,"text":2665},{"id":2671,"depth":260,"text":2672},{"id":1492,"depth":259,"text":1493},{"id":2686,"depth":259,"text":2687},"2026-01-11","Retrieval-Augmented Generation (RAG) explained: how AI can use your own content in a way that is trustworthy, traceable, and manageable.","\u002Fimages\u002Fnews\u002Frag-database.jpg",{},"\u002Fnews\u002F2026-01-11-how-rag-works","7 min read",{"title":2520,"description":2716},"news\u002F2026-01-11-how-rag-works",[360,2724,2725,2726],"RAG","knowledge","architecture","how-rag-works","mT_0rRvgzngjggN0Uv3ZtqrzLTZysYlTihr49s5SnGk",{"id":2730,"title":2731,"author":505,"body":2732,"category":271,"date":2715,"description":2952,"extension":274,"image":2953,"meta":2954,"navigation":277,"path":2955,"readTime":972,"seo":2956,"stem":2957,"tags":2958,"translationId":2960,"__hash__":2961},"news\u002Fnews\u002F2026-01-11-when-to-use-rag-vs-prompt-to-sql.md","RAG vs Prompt-to-SQL – When to Use What?",{"type":9,"value":2733,"toc":2940},[2734,2737,2742,2745,2748,2752,2755,2758,2762,2765,2768,2771,2775,2778,2781,2784,2788,2791,2862,2866,2869,2872,2875,2879,2882,2885,2888,2892,2895,2898,2901,2908,2911,2915,2918,2921,2923,2926,2929,2932,2934,2937],[12,2735,2736],{},"When AI starts being used for real in organizations, the same question almost always comes up sooner or later:",[12,2738,2739],{},[39,2740,2741],{},"Should we use RAG or Prompt-to-SQL?",[12,2743,2744],{},"It's a reasonable question. Both techniques sound similar, both are based on generative AI, and both are often used in similar contexts. At the same time, they solve completely different problems. When they're confused, the result is solutions that look smart on the surface but give wrong answers, are difficult to control, and in the worst case, can't be trusted.",[12,2746,2747],{},"In this article, we clarify the differences. We go through when RAG is the right choice, when Prompt-to-SQL is better, and how they can be combined in a way that actually works in practice.",[25,2749,2751],{"id":2750},"two-techniques-with-different-missions","Two Techniques with Different Missions",[12,2753,2754],{},"A good way to think about it is to start with what type of question needs to be answered.",[12,2756,2757],{},"When the answer is found in text, documents, or formulations, RAG is usually the right path. When the answer is in numbers, tables, or measurement data, Prompt-to-SQL is the natural choice. Both use generative AI, but they work against completely different types of sources and place different demands on the architecture.",[25,2759,2761],{"id":2760},"when-rag-is-best-suited","When RAG Is Best Suited",[12,2763,2764],{},"RAG, or Retrieval-Augmented Generation, is used when the question requires interpretation and context rather than exact values. It's often about understanding what applies, how something should be applied, or how previous decisions have been reasoned.",[12,2766,2767],{},"Typical sources are policy documents, handbooks, guidelines, decisions, and other unstructured text. Questions might, for example, concern what applies to remote work, how a case should be handled, or which rules are relevant in a certain context.",[12,2769,2770],{},"This is where RAG is strong, because the technique can read in relevant material, weigh together information from multiple sources, and formulate a coherent answer.",[25,2772,2774],{"id":2773},"when-prompt-to-sql-is-the-right-choice","When Prompt-to-SQL Is the Right Choice",[12,2776,2777],{},"Prompt-to-SQL is used instead when the answer is in structured data and must be correct at the detail level. This might involve counts, averages, trends, or comparisons over time.",[12,2779,2780],{},"The sources are then databases, data warehouses, and business systems. Questions might, for example, concern how many cases exceeded SLA last week or how processing time differs between different teams.",[12,2782,2783],{},"In these cases, a question in natural language is translated to SQL that runs directly against the database. The result is exact numbers, rather than interpretations.",[25,2785,2787],{"id":2786},"a-clear-comparison","A Clear Comparison",[12,2789,2790],{},"The differences between RAG and Prompt-to-SQL become clear when you see them side by side.",[1366,2792,2793,2805],{},[1369,2794,2795],{},[1372,2796,2797,2800,2802],{},[1375,2798,2799],{},"Dimension",[1375,2801,2724],{},[1375,2803,2804],{},"Prompt-to-SQL",[1385,2806,2807,2818,2829,2840,2851],{},[1372,2808,2809,2812,2815],{},[1390,2810,2811],{},"Data type",[1390,2813,2814],{},"Unstructured text",[1390,2816,2817],{},"Structured data",[1372,2819,2820,2823,2826],{},[1390,2821,2822],{},"Type of answer",[1390,2824,2825],{},"Explanatory and summarizing",[1390,2827,2828],{},"Exact values",[1372,2830,2831,2834,2837],{},[1390,2832,2833],{},"Character",[1390,2835,2836],{},"Interpretation-based",[1390,2838,2839],{},"Deterministic",[1372,2841,2842,2845,2848],{},[1390,2843,2844],{},"Common sources",[1390,2846,2847],{},"Documents and text",[1390,2849,2850],{},"Tables and databases",[1372,2852,2853,2856,2859],{},[1390,2854,2855],{},"Typical risk",[1390,2857,2858],{},"Misleading reasoning",[1390,2860,2861],{},"Wrong numbers that look reasonable",[25,2863,2865],{"id":2864},"common-use-cases","Common Use Cases",[12,2867,2868],{},"RAG is often used in internal AI assistants, knowledge support, policy questions, and case management where context and formulations are crucial.",[12,2870,2871],{},"Prompt-to-SQL is instead used for analysis, follow-up, management questions, and operational control where the numbers are what count.",[12,2873,2874],{},"Problems arise when the techniques are used in the wrong context.",[25,2876,2878],{"id":2877},"common-architectural-mistakes","Common Architectural Mistakes",[12,2880,2881],{},"A common mistake is trying to use RAG for statistics and numbers. When answers are based on documents or exports, they quickly become outdated and often more interpretive than factual.",[12,2883,2884],{},"The opposite mistake is using Prompt-to-SQL for questions that are really about rules or interpretations. A database can say what happened, but not why something applies or what exceptions exist.",[12,2886,2887],{},"A third mistake is mixing everything in the same question without clear control. When text, numbers, and rules are handled simultaneously, answers become difficult to verify and even harder to troubleshoot.",[25,2889,2891],{"id":2890},"when-the-techniques-are-combined-the-right-way","When the Techniques Are Combined the Right Way",[12,2893,2894],{},"In mature AI solutions, RAG and Prompt-to-SQL are used together, but with clearly separate roles.",[12,2896,2897],{},"A common setup is that RAG first helps understand the question and put it in the right context. Then Prompt-to-SQL is used to fetch exact numbers. Finally, RAG is used again to explain what the numbers mean and how they should be interpreted.",[12,2899,2900],{},"An answer might, for example, look like this:",[834,2902,2903],{},[12,2904,2905],{},[132,2906,2907],{},"Last week, 14 cases exceeded SLA, mainly in process X. This is an increase compared to the previous week and coincides with the change introduced at the beginning of the month.",[12,2909,2910],{},"Here, facts are combined with context in a way that is both correct and understandable.",[25,2912,2914],{"id":2913},"why-control-matters","Why Control Matters",[12,2916,2917],{},"Both RAG and Prompt-to-SQL require clear control to work over time. This involves permissions, logging, answer validation, and quality follow-up.",[12,2919,2920],{},"Without this, the AI solution risks being fast and impressive, but at the same time difficult to trust and even harder to maintain.",[25,2922,1493],{"id":1492},[12,2924,2925],{},"RAG and Prompt-to-SQL are not competing techniques. They complement each other.",[12,2927,2928],{},"RAG helps the AI understand and reason. Prompt-to-SQL ensures that answers are correct at the detail level. Together, they create solutions that work in real organizations, not just in demos.",[12,2930,2931],{},"The key is to use the right technique for the right type of question and to build the architecture with this in mind.",[25,2933,2687],{"id":2686},[12,2935,2936],{},"Want to see how RAG and Prompt-to-SQL can be combined in your existing systems, with the right level of control, security, and follow-up?",[12,2938,2939],{},"Walma helps organizations move from AI experiments to solutions that hold up in production.",{"title":258,"searchDepth":259,"depth":260,"links":2941},[2942,2943,2944,2945,2946,2947,2948,2949,2950,2951],{"id":2750,"depth":259,"text":2751},{"id":2760,"depth":259,"text":2761},{"id":2773,"depth":259,"text":2774},{"id":2786,"depth":259,"text":2787},{"id":2864,"depth":259,"text":2865},{"id":2877,"depth":259,"text":2878},{"id":2890,"depth":259,"text":2891},{"id":2913,"depth":259,"text":2914},{"id":1492,"depth":259,"text":1493},{"id":2686,"depth":259,"text":2687},"Two AI techniques often confused. Here we clarify the differences, show when each technique is best suited, and why they often need to be used together.","\u002Fimages\u002Fnews\u002Frag-or-prompt-to-sql.jpg",{},"\u002Fnews\u002F2026-01-11-when-to-use-rag-vs-prompt-to-sql",{"title":2731,"description":2952},"news\u002F2026-01-11-when-to-use-rag-vs-prompt-to-sql",[360,2724,2804,2726,2959],"decision support","rag-vs-sql","UOjVOjB1Pfp15qBqRmJJ3umrE5ZR1bydEl4fNCiZYJg",1789134649359]