Not wrong, but that's not how we do it here
Many AI results are technically correct and still go unused, because they do not look the way the company expects. For AI to land, it has to know how you work. Whoever teaches it that also decides: keep the process as it is today, or change it now?

Picture an engineer receiving the draft of a technical statement, written by an AI. She checks the values and compares them with the specification. On substance, she finds little to fault. Then comes the sentence we hear again and again in rollout projects: "Not wrong, but that's not how we do it here."
Sometimes the usual structure is missing, sometimes the reference to the in-house test instruction. Sometimes a deviation goes straight to the customer, where here the project manager decides first. So she rewrites the draft, or it is left lying. The time the AI was meant to save is gone, and so is some trust in the next result.
The model is less and less the problem
For a long time the question was whether a model could read a specification sensibly at all. Today many results are technically correct. What is missing is the knowledge of how you work.
Analyses from one of the large AI labs reach a similar conclusion: demanding use of AI models may be constrained more by access to information than by the models' capabilities.1
What the AI needs to know about your company
People expect of an AI what they expect of a new project colleague: that after a while it has understood how things work here. For that it needs four things.
- Responsibilities: who reviews, who approves and who talks to the customer.
- Templates: structure, numbering and revision fields as the company uses them.
- Standard procedures: the test instructions and procedures people work by.
- Guardrails: standards, customer requirements and rules from outside, such as the classification society's plan review.2
This knowledge rarely sits in one place. It lives in documents, in systems and in people's heads. The AI therefore needs access to your files and systems, and some of it has to be written down for the first time. Otherwise every result stays a draft to rewrite.
Writing it down forces a decision
Whoever teaches the AI their way of working often writes it down completely for the first time. You see which steps are needed and which exist only out of habit. That raises a question nobody asked before: should the AI work exactly as you do today, or is this the moment to change the process?
Both are legitimate. Which way fits depends on the individual process.
Map it or rebuild it
Trust speaks for mapping. The result looks the way the team expects from day one, it is reviewed as usual, and the time saved is there at once. The process can be changed later at relatively low cost, because with AI an adjusted instruction or a new form is often enough.
The greater value speaks for rebuilding. Superfluous steps go at once. In McKinsey's survey, the leaders who get real results from AI are nearly three times as likely as other companies to have fundamentally redesigned workflows.3 Such projects are more demanding, though, because other systems around the process are often affected.
of companies have not yet redesigned jobs around AI capabilities
Deloitte, January 20264What meets the expectation gets used
Mapped or rebuilt: a result is accepted only once it looks like your own work. Then the AI takes over the tedious small tasks, gathering details and filling in templates, and there is time for the work that is fun: finding and weighing technical solutions.
Our observation: people who have experienced this once often ask for the next use case on their own. This is how we approach rollouts at Engera: weigh it up per process whether to map first or rebuild straight away, together with the team.
What to do now
- 01
Capture how you work
Write down who does what and which templates, standard procedures and guardrails apply.
- 02
Give the AI access
Connect it to the files and systems where this knowledge lives.
- 03
Decide per process
Weigh whether to map the process first or rebuild it straight away.
- 04
Build trust
Have results reviewed as usual until the team trusts them, then remove review steps bit by bit. The final sign-off stays with people.
Sources
- Anthropic: Anthropic Economic Index report. Uneven geographic and enterprise AI adoption, 15 September 2025
- IACS: Classification societies, their key role (Information Paper)
- McKinsey: The state of AI in 2025. Agents, innovation, and transformation, 5 November 2025
- Deloitte: State of AI in the Enterprise. The untapped edge, January 2026
Read on
Which challenge is keeping
your team busy right now?
Tell us, and in the demo we show you how Engera makes it easier.

