[{"data":1,"prerenderedAt":450},["ShallowReactive",2],{"blog-posts-en":3},[4,230,346],{"_path":5,"_dir":6,"_draft":7,"_partial":7,"_locale":8,"title":9,"description":10,"slug":11,"locale":12,"date":13,"readingTime":14,"summaryUseful":15,"summaryTakeaway":16,"coverLabel":17,"coverAlt":18,"socialImage":19,"socialImageAlt":20,"tags":21,"body":25,"_type":224,"_id":225,"_source":226,"_file":227,"_stem":228,"_extension":229},"\u002Fposts\u002Fwhy-most-ai-projects-die-between-the-pilot-and-production","posts",false,"","Why AI projects get stuck between pilot and operations","A working AI pilot does not change a way of working on its own. The critical question is who owns the move into day-to-day operations.","why-most-ai-projects-die-between-the-pilot-and-production","en","2026-05-13","6","a pilot works, but has not yet become an established way of working.","Adoption does not begin after implementation. It is part of implementation.","FROM PILOT TO OPERATIONS","A paper ribbon passes through a precise checkpoint on its way into day-to-day operations.","\u002Fgraphics\u002Fribbon-checkpoint-dark-v1.png","A paper ribbon passing through a precise checkpoint.",[22,23,24],"AI","Adoption","Implementation",{"type":26,"children":27,"toc":211},"root",[28,36,41,46,53,58,63,68,74,79,86,91,96,102,107,112,118,123,129,134,139,145,150,155,160,165,170,176,181,206],{"type":29,"tag":30,"props":31,"children":32},"element","p",{},[33],{"type":34,"value":35},"text","Many AI initiatives I see begin in a similar way: a workshop identifies a plausible use case, a first prototype follows, and the demonstration works. That is a useful milestone — but it is not yet a changed way of working.",{"type":29,"tag":30,"props":37,"children":38},{},[39],{"type":34,"value":40},"Later, the underlying process is still largely manual, the old spreadsheet remains in use, and the new solution is used irregularly. Operational accountability and the transition into day-to-day operations are often still unresolved.",{"type":29,"tag":30,"props":42,"children":43},{},[44],{"type":34,"value":45},"The difficult task is therefore not always the first prototype. It is turning that prototype into a dependable operational practice. That is where many initiatives get stuck.",{"type":29,"tag":47,"props":48,"children":50},"h2",{"id":49},"the-handover-without-clear-accountability",[51],{"type":34,"value":52},"The handover without clear accountability",{"type":29,"tag":30,"props":54,"children":55},{},[56],{"type":34,"value":57},"In many organisations, the pilot phase has an owner and day-to-day operations have an owner, but nobody feels responsible for the transition between them.",{"type":29,"tag":30,"props":59,"children":60},{},[61],{"type":34,"value":62},"Innovation teams get the work started. External partners build the prototype. Department heads want an outcome. IT wants stability. Legal wants clarity. Every concern is legitimate. Yet when the question becomes who will ensure that the solution is used every day, the answer quickly becomes vague.",{"type":29,"tag":30,"props":64,"children":65},{},[66],{"type":34,"value":67},"The software exists, but the new way of working is not sufficiently embedded. As long as the previous process remains available in parallel without an agreed end date, people can return to it under pressure.",{"type":29,"tag":47,"props":69,"children":71},{"id":70},"why-working-pilots-still-get-stuck",[72],{"type":34,"value":73},"Why working pilots still get stuck",{"type":29,"tag":30,"props":75,"children":76},{},[77],{"type":34,"value":78},"Four problems recur in the stalled AI rollouts I see.",{"type":29,"tag":80,"props":81,"children":83},"h3",{"id":82},"_1-the-use-case-remains-too-imprecise",[84],{"type":34,"value":85},"1. The use case remains too imprecise",{"type":29,"tag":30,"props":87,"children":88},{},[89],{"type":34,"value":90},"When an initiative begins with a broad ambition such as “support sales with AI” or “automate reporting,” the team often skips the difficult work of identifying the exact step in the process that should change.",{"type":29,"tag":30,"props":92,"children":93},{},[94],{"type":34,"value":95},"A sound use case names a concrete point of friction, the role affected, the trigger, and the decision required. Without that foundation, the implementation may look impressive but has no stable place in the organisation.",{"type":29,"tag":80,"props":97,"children":99},{"id":98},"_2-the-previous-process-remains-available-in-parallel",[100],{"type":34,"value":101},"2. The previous process remains available in parallel",{"type":29,"tag":30,"props":103,"children":104},{},[105],{"type":34,"value":106},"The new AI-supported process is introduced, but the previous version remains “just in case.” That appears cautious, yet often weakens adoption. Under pressure, people return to the familiar process while it is still available.",{"type":29,"tag":30,"props":108,"children":109},{},[110],{"type":34,"value":111},"When both processes remain in parallel without a clear end date, the transition is easily deferred.",{"type":29,"tag":80,"props":113,"children":115},{"id":114},"_3-the-prototype-is-treated-as-the-outcome",[116],{"type":34,"value":117},"3. The prototype is treated as the outcome",{"type":29,"tag":30,"props":119,"children":120},{},[121],{"type":34,"value":122},"A working prototype is an important technical milestone. The operational outcome only becomes visible when the new way of working is used reliably over an agreed period.",{"type":29,"tag":80,"props":124,"children":126},{"id":125},"_4-enablement-is-planned-only-at-the-end",[127],{"type":34,"value":128},"4. Enablement is planned only at the end",{"type":29,"tag":30,"props":130,"children":131},{},[132],{"type":34,"value":133},"Many initiatives treat enablement as a communication task at the end. Shortly before rollout, there is a training session, a presentation or a document, and then the project moves on.",{"type":29,"tag":30,"props":135,"children":136},{},[137],{"type":34,"value":138},"That is usually too late. If people do not understand how the process changes, which quality standard applies, when they can trust the system, and whom to ask when problems occur, confidence does not develop quickly enough. They return to the previous way of working.",{"type":29,"tag":47,"props":140,"children":142},{"id":141},"what-makes-the-transition-hold",[143],{"type":34,"value":144},"What makes the transition hold",{"type":29,"tag":30,"props":146,"children":147},{},[148],{"type":34,"value":149},"Initiatives that reach day-to-day operations plan rollout and adoption as early as the technical implementation.",{"type":29,"tag":30,"props":151,"children":152},{},[153],{"type":34,"value":154},"First, the new way of working needs clearly named operational accountability after rollout — for quality, changes, and questions, not only for the technology.",{"type":29,"tag":30,"props":156,"children":157},{},[158],{"type":34,"value":159},"Second, it must be clear which way of working should change and which previous process will end. Third, the solution has to reflect daily work: inputs, handovers, approvals, error cases, and situations in which an output is only partly correct.",{"type":29,"tag":30,"props":161,"children":162},{},[163],{"type":34,"value":164},"Fourth, the team agrees operational signals in advance, such as regular usage, manual effort, adherence to the process, and necessary escalations. These signals show whether the solution has actually become part of operations.",{"type":29,"tag":30,"props":166,"children":167},{},[168],{"type":34,"value":169},"A responsibly run AI initiative therefore connects assessment, technical implementation, and adoption in daily work. Together, these elements turn a prototype into a solution the organisation can operate responsibly.",{"type":29,"tag":47,"props":171,"children":173},{"id":172},"what-to-clarify-before-the-next-rollout",[174],{"type":34,"value":175},"What to clarify before the next rollout",{"type":29,"tag":30,"props":177,"children":178},{},[179],{"type":34,"value":180},"Before starting the next pilot, answer these questions in writing:",{"type":29,"tag":182,"props":183,"children":184},"ol",{},[185,191,196,201],{"type":29,"tag":186,"props":187,"children":188},"li",{},[189],{"type":34,"value":190},"Which concrete way of working should change?",{"type":29,"tag":186,"props":192,"children":193},{},[194],{"type":34,"value":195},"Who carries operational accountability after rollout?",{"type":29,"tag":186,"props":197,"children":198},{},[199],{"type":34,"value":200},"Which previous process ends when — and which remains, for what reason?",{"type":29,"tag":186,"props":202,"children":203},{},[204],{"type":34,"value":205},"Which agreed signals will show that the solution is being used reliably?",{"type":29,"tag":30,"props":207,"children":208},{},[209],{"type":34,"value":210},"If these answers are unclear, the initiative can work technically and still remain a pilot. Plan rollout and adoption as part of implementation from the beginning — not as a communication or training task added afterwards.",{"title":8,"searchDepth":212,"depth":212,"links":213},3,[214,216,222,223],{"id":49,"depth":215,"text":52},2,{"id":70,"depth":215,"text":73,"children":217},[218,219,220,221],{"id":82,"depth":212,"text":85},{"id":98,"depth":212,"text":101},{"id":114,"depth":212,"text":117},{"id":125,"depth":212,"text":128},{"id":141,"depth":215,"text":144},{"id":172,"depth":215,"text":175},"markdown","content:posts:why-most-ai-projects-die-between-the-pilot-and-production.md","content","posts\u002Fwhy-most-ai-projects-die-between-the-pilot-and-production.md","posts\u002Fwhy-most-ai-projects-die-between-the-pilot-and-production","md",{"_path":231,"_dir":6,"_draft":7,"_partial":7,"_locale":8,"title":232,"description":233,"slug":234,"locale":12,"date":235,"readingTime":236,"summaryUseful":237,"summaryTakeaway":238,"coverLabel":239,"coverAlt":240,"socialImage":241,"socialImageAlt":242,"tags":243,"body":246,"_type":224,"_id":343,"_source":226,"_file":344,"_stem":345,"_extension":229},"\u002Fposts\u002Fwhen-an-ai-use-case-should-not-be-built","When an AI use case should deliberately not be built","Not every idea improves when you implement it with AI. Sometimes the most robust decision is to choose not to build.","when-an-ai-use-case-should-not-be-built","2026-04-08","5","a plausible use case exists, but the prerequisites for responsible operation are still missing.","The decision against an initiative can be as valuable as delivery.","DECISION OVER DEMO","A paper ribbon reaches a clear fork in the path.","\u002Fgraphics\u002Fribbon-decision-dark-v1.png","A paper ribbon at a decision fork.",[22,244,245],"Decision","Governance",{"type":26,"children":247,"toc":334},[248,253,258,264,269,275,280,286,291,297,302,308,313,319,324,329],{"type":29,"tag":30,"props":249,"children":250},{},[251],{"type":34,"value":252},"Many organisations come to me with a concrete idea: a document should be analysed automatically, an assistant should prepare answers, or a process should become AI-supported. The motivation is understandable. Not every one of these ideas should become an initiative.",{"type":29,"tag":30,"props":254,"children":255},{},[256],{"type":34,"value":257},"A sound use case needs more than an impressive workshop. It needs a clear point of friction, a role that benefits, and an environment in which the result can be operated responsibly.",{"type":29,"tag":47,"props":259,"children":261},{"id":260},"when-no-is-more-useful-than-a-pilot",[262],{"type":34,"value":263},"When no is more useful than a pilot",{"type":29,"tag":30,"props":265,"children":266},{},[267],{"type":34,"value":268},"I more often advise against building when one of these signals dominates:",{"type":29,"tag":80,"props":270,"children":272},{"id":271},"_1-the-process-is-not-stable-enough-yet",[273],{"type":34,"value":274},"1. The process is not stable enough yet",{"type":29,"tag":30,"props":276,"children":277},{},[278],{"type":34,"value":279},"When the underlying workflow is still unclear, AI tends to amplify ambiguity rather than quality. Process work should come first — not model selection.",{"type":29,"tag":80,"props":281,"children":283},{"id":282},"_2-the-data-basis-is-not-approved",[284],{"type":34,"value":285},"2. The data basis is not approved",{"type":29,"tag":30,"props":287,"children":288},{},[289],{"type":34,"value":290},"Without clear permissions, sources, and freshness, you quickly get a solution that works technically but cannot be owned operationally.",{"type":29,"tag":80,"props":292,"children":294},{"id":293},"_3-the-expected-benefit-depends-on-perfection",[295],{"type":34,"value":296},"3. The expected benefit depends on perfection",{"type":29,"tag":30,"props":298,"children":299},{},[300],{"type":34,"value":301},"If the initiative only makes sense when the system is right in almost every case, the use case is usually too broad or too risky as a starting point.",{"type":29,"tag":80,"props":303,"children":305},{"id":304},"_4-nobody-owns-operations",[306],{"type":34,"value":307},"4. Nobody owns operations",{"type":29,"tag":30,"props":309,"children":310},{},[311],{"type":34,"value":312},"A prototype without later accountability for quality, adjustments, and questions remains an experiment — regardless of model quality.",{"type":29,"tag":47,"props":314,"children":316},{"id":315},"what-helps-instead",[317],{"type":34,"value":318},"What helps instead",{"type":29,"tag":30,"props":320,"children":321},{},[322],{"type":34,"value":323},"Rather than building immediately, a tighter step often pays off: define the use case in writing, name the role affected, review the data sources, and agree which signal would show that a pilot is worthwhile.",{"type":29,"tag":30,"props":325,"children":326},{},[327],{"type":34,"value":328},"Sometimes that leads to a smaller, more viable initiative. Sometimes to the insight that the right next step is not AI at all.",{"type":29,"tag":30,"props":330,"children":331},{},[332],{"type":34,"value":333},"Both are useful outcomes.",{"title":8,"searchDepth":212,"depth":212,"links":335},[336,342],{"id":260,"depth":215,"text":263,"children":337},[338,339,340,341],{"id":271,"depth":212,"text":274},{"id":282,"depth":212,"text":285},{"id":293,"depth":212,"text":296},{"id":304,"depth":212,"text":307},{"id":315,"depth":215,"text":318},"content:posts:when-an-ai-use-case-should-not-be-built.md","posts\u002Fwhen-an-ai-use-case-should-not-be-built.md","posts\u002Fwhen-an-ai-use-case-should-not-be-built",{"_path":347,"_dir":6,"_draft":7,"_partial":7,"_locale":8,"title":348,"description":349,"slug":350,"locale":12,"date":351,"readingTime":236,"summaryUseful":352,"summaryTakeaway":353,"coverLabel":354,"coverAlt":355,"socialImage":356,"socialImageAlt":357,"tags":358,"body":360,"_type":224,"_id":447,"_source":226,"_file":448,"_stem":449,"_extension":229},"\u002Fposts\u002Fwhat-good-ai-governance-looks-like-in-practice","What good AI governance looks like in practice","Governance is not a slide deck for approval. It decides whether an AI solution can be used responsibly in day-to-day work.","what-good-ai-governance-looks-like-in-practice","2026-03-22","an AI initiative is about to start, but roles, approvals, and accountability are still unclear.","Good governance makes decisions traceable — not only documented.","ROLES · APPROVALS · USE","A paper ribbon connects several layers into one continuous flow.","\u002Fgraphics\u002Fribbon-structure-dark-v1.png","A paper ribbon with clear structure.",[22,245,359],"Operations",{"type":26,"children":361,"toc":438},[362,367,372,378,383,389,395,400,406,411,417,422,428,433],{"type":29,"tag":30,"props":363,"children":364},{},[365],{"type":34,"value":366},"Governance is often handled too late. Only once an assistant works do questions about permissions, data use, approvals, and accountability appear. Then governance quickly feels like a brake.",{"type":29,"tag":30,"props":368,"children":369},{},[370],{"type":34,"value":371},"In practice, it is closer to the opposite: it makes clear what may be built, tested, and introduced.",{"type":29,"tag":47,"props":373,"children":375},{"id":374},"what-governance-does-not-mean-here",[376],{"type":34,"value":377},"What governance does not mean here",{"type":29,"tag":30,"props":379,"children":380},{},[381],{"type":34,"value":382},"Good governance is not a forty-page concept that grows in parallel to the project and nobody uses in daily work. It has to fit the concrete process: who may use which source, which outputs need review, and who decides on changes.",{"type":29,"tag":47,"props":384,"children":386},{"id":385},"three-layers-that-have-to-align",[387],{"type":34,"value":388},"Three layers that have to align",{"type":29,"tag":80,"props":390,"children":392},{"id":391},"_1-data-access-and-permissions",[393],{"type":34,"value":394},"1. Data access and permissions",{"type":29,"tag":30,"props":396,"children":397},{},[398],{"type":34,"value":399},"A solution may only access information approved for the task. That applies equally to SharePoint, email, CRM data, and internal systems.",{"type":29,"tag":80,"props":401,"children":403},{"id":402},"_2-quality-and-approval",[404],{"type":34,"value":405},"2. Quality and approval",{"type":29,"tag":30,"props":407,"children":408},{},[409],{"type":34,"value":410},"Not every output should continue automatically. Many tasks need defined review points — especially where errors have follow-on cost.",{"type":29,"tag":80,"props":412,"children":414},{"id":413},"_3-operations-and-ownership",[415],{"type":34,"value":416},"3. Operations and ownership",{"type":29,"tag":30,"props":418,"children":419},{},[420],{"type":34,"value":421},"After rollout, it must be clear who owns adjustments, how errors become visible, and when a process is stopped or rolled back.",{"type":29,"tag":47,"props":423,"children":425},{"id":424},"the-practical-test",[426],{"type":34,"value":427},"The practical test",{"type":29,"tag":30,"props":429,"children":430},{},[431],{"type":34,"value":432},"A simple test: can someone outside the project team explain in ten minutes who is accountable for the solution, which data it uses, and when an output must not be passed on automatically?",{"type":29,"tag":30,"props":434,"children":435},{},[436],{"type":34,"value":437},"If those answers are unclear, what is missing is not more technology — but the governance that makes responsible operation possible.",{"title":8,"searchDepth":212,"depth":212,"links":439},[440,441,446],{"id":374,"depth":215,"text":377},{"id":385,"depth":215,"text":388,"children":442},[443,444,445],{"id":391,"depth":212,"text":394},{"id":402,"depth":212,"text":405},{"id":413,"depth":212,"text":416},{"id":424,"depth":215,"text":427},"content:posts:what-good-ai-governance-looks-like-in-practice.md","posts\u002Fwhat-good-ai-governance-looks-like-in-practice.md","posts\u002Fwhat-good-ai-governance-looks-like-in-practice",1784773582995]