Module 6Academy

From pilot to production

Why most pilots do not land, and what a team must arrange before something runs daily.

From pilot to production is a half-day session for a team that has run a trial and wants to know why it does not stick. The team learns the four reasons pilots stall, decides what a pilot must prove before it scales, and records ownership, measurement and maintenance. The result is a go/no-go framework the team can use itself.

After this module the team can
  • 01The team names the four reasons pilots stall and tests its own pilot against them
  • 02The team writes down the evidence a pilot must deliver before it scales, with a threshold
  • 03Every production AI has a named owner, a measure and a maintenance agreement
  • 04The team uses a go/no-go framework it applies itself after the module
Agenda
  1. 01

    Why pilots stall

    Four patterns: no owner, no measure, a demo that does not fit the work, and a build nobody maintains. We recognise them in the team’s own pilot.

  2. 02

    The evidence

    What must a pilot show? We write the thresholds down before we look at the results, so the decision is not coloured afterwards.

  3. 03

    Ownership and maintenance

    Who decides, who watches, who repairs and who stops it. With an agreement on monthly cost and on what happens when the model changes.

  4. 04

    The go/no-go framework

    One page with the questions, the thresholds and the decision-maker. The team applies it to its own pilot and then decides.

The pilot is the easy part

A pilot is small, manageable and often started with enthusiasm. Someone builds something in a corner of the organisation, it works on a few test cases, and the presentation is good. The difficult part starts afterwards: the system must run every day, the exceptions must be handled, and someone must feel responsible for the errors. That part was not built in the pilot, which is why the pilot stalls.

The MIT NANDA study from 2025 found that most of the organisations it studied saw no measurable return from their generative AI initiatives. The study does not point at the models as the cause but at the link to daily work and at the question of who manages the system afterwards. That is why this module starts with the organisation, not the technology.

Four reasons pilots stall

First, there is no owner. The pilot has a maker, but nobody decides what happens with the outcome. Second, there is no measure: the pilot shows that it works, but not how much it delivers compared with today. Third, the demo does not fit the work: the test cases are clean and the real work is messy. Fourth, there is no maintenance. Models change, vendors change prices and terms, and a link breaks when a system gets an update.

In the module the team tests its own pilot against these four. That takes a short questionnaire, and the answers are often uncomfortable. That is useful: a pilot that scores three of the four is not yet production, and that is a decision the team can take without shaming the project.

Write the evidence before you look

The biggest mistake with a pilot is looking at the result and only then deciding what is good enough. Every result then becomes a success, or every result a failure, depending on who is in the room. The module therefore asks the team to write three things in advance: which task the system must take over, which quality that requires, and which number translates that into hours or cost.

For example: a system that takes orders from PDFs must reach a fixed error ratio on a control sample, and the team must agree how many manual corrections per week are acceptable. Those two numbers decide, not the impression the demo made.

Ownership, measurement and maintenance

A production system has three names. An owner who decides what it does and may do. A manager who watches it and takes up alerts. And an agreement on maintenance: who repairs it when a link breaks, how fast, and what that costs per month. Most pilots do not record those three, and that is the moment a pilot gets dropped again.

Measurement belongs with it. Every production AI gets a small set of numbers that the team reviews every month: how many tasks, how many manual corrections, and how much time the team still spends on it. If nobody reviews those numbers, nobody knows when the system stops working.

The go/no-go framework

The module closes with a page of four questions. Is the threshold of evidence met? Is there an owner with the time to do it? Is there a measure the team uses every month? Is there a maintenance agreement with a cost? A yes on all four is a go. A no on one or two is often a deferral with a clear assignment. A no on more than two is a decision to stop, and that is a good decision.

Frequently asked questions

  • What if the pilot works but nobody uses it?

    That is an adoption problem, not a technology problem. The team then looks at the work around it: is the system where the work arrives, and is the work instruction updated? That comes back in the adoption programme.

  • How long does a pilot take to reach production?

    That depends on the build and access to the systems, not on the module. The module makes sure you know what must happen first, and what you may leave alone.

  • Does this replace a blueprint?

    No. A blueprint describes the architecture and return of a specific system. This module decides whether you should go on, and what you then arrange.

Which team needs this module?

For half day €2,450 excl. VAT. In the conversation we say whether the module fits, or whether a build day or the adoption programme is better.