Overview
Most AI pilots in BeLux mid-market companies do not fail because the technology is bad. They fail because the foundation underneath the technology is bad. The licences are active, the tenant is configured, the training session was held. From the procurement side, the project looks successful. From inside the work, nothing has changed, or the work has gotten harder.
This article is about why that happens, and what the unglamorous foundation work underneath looks like in practice.
The Pilot Graveyard
The pattern repeats with little variation. We deployed it but nobody uses it. We deployed it and now audit is asking questions. The vendor said it would change everything, it changed nothing. We are pausing while we figure out what to do next.
Most of these projects had a clean technical setup. By every project management measure, the pilot was on time and on budget. The failure was never in the deployment. It was in everything underneath the deployment.
What sits underneath. Data that lives in three or four places, with no clear source of truth. Processes with tribal knowledge baked into them, where three people on the team know unwritten rules. Approval flows in email threads, Teams chats, and the heads of senior staff. Processes that nobody owns end-to-end.
When you drop AI on top of that landscape, you do not get clarity. You get the same chaos, paraphrased confidently.
What 'Foundation' Actually Means
The word 'foundation' lands as something abstract in most boardrooms. So let me make it concrete.
Foundation is the answer to four questions, asked about every process you want AI to touch. Where does the master version of this data live, and who is allowed to change it. What are the steps of this process, written down somewhere people can read. Who approves what, with what evidence, in what system. Who owns this process when something goes wrong.
If those four answers exist and they are clear, you have a foundation. If any of them is a shrug, you do not. AI will not give it to you. It will amplify whatever clarity already exists, and amplify whatever confusion already exists.
This is unromantic work. It is governance work. It is process documentation work. It is data ownership work. It looks nothing like the demo videos the vendor showed your board. And it is the entire difference between a Copilot deployment that lands and one that quietly stalls.
The Three Layers Below AI
Pick a process you would like AI to help with. Customer onboarding. Vendor intake. Procurement approvals. Service-desk triage. Operational reporting. Whatever lives in your head right now.
Underneath the AI layer, there are three layers that need to exist before AI can do anything useful.
Layer one is data integrity. For the process you chose, there is a set of data that drives it. For AI to be useful, that data needs to live in one system, with one definition, with one set of permissions, with one update mechanism. Not three spreadsheets, two CRM exports, and a SharePoint folder. If the data lives in multiple places, AI does not pick the right one. It blends them. Then it speaks with confidence about an answer that came from a blend you cannot audit.
Layer two is process clarity. The steps of the process need to be written down. Not in someone's head. Not in a Word document last updated three years ago. In a system, ideally one that connects to the data layer above. Power Automate flows, Dynamics business processes, Power Apps with built-in stages. The key is that someone outside the team can read the process and understand what happens next, in what order, under what conditions.
Layer three is decision and approval traceability. For every decision in the process, there should be a record. Who decided, when, on what basis. Not a Teams message. Not an email thread. A record inside the system. This is what makes AI useful for audit and regulated work. The AI does not just help with the decision. It does so on top of a traceable audit trail you can defend.
How To Know If You Are Ready
The honest test, for any given process you want AI to touch.
Can you point to one system that holds the master data for this process. Can you show the process steps inside a system, not a document. Can you tell who approved the last three big decisions, and find the audit trail in under two minutes. Can you tell who owns this process today, and what their accountability looks like.
Four yeses, and you are ready to bring AI into this process. The Copilot deployment will land. The agent will earn its place.
Three yeses, and you have foundation work to do first. Maybe two weeks of work. Maybe two months. It depends on what is missing.
One or two yeses, and AI is not the next investment for this process. Foundation is.
This is the conversation the AMPLIFY IT Maturity Assessment is designed to have. Not 'are you ready for AI.' That question is too broad. The real question is which of your processes are ready for AI today, which need foundation work first, and which are not even close. Three different conversations, each with a different next step.
The companies that get this right do not pick their AI use cases by what is fashionable. They pick them by what is ready. They start with the modules already at Level 3 and Level 4 on the maturity ladder. They run foundation projects in parallel on the Level 1 and Level 2 modules. Twelve months later, they have a portfolio of AI use cases in production and a clear plan for the rest.
The companies that get it wrong pick the most exciting use case in the room and hope. They deploy. They quietly stall. They tell the next vendor 'we tried AI, it did not work for us.'
It worked. The foundation underneath did not.
AMPLIFY IT helps Mid-Market companies in Belgium and Luxembourg identify, prioritize, and deliver high-impact workflow automation on Microsoft Dynamics 365 and Power Platform, starting with the workflow that funds the rest.
