A team uses AI to write its meeting minutes. Everyone saves ten minutes. But the minutes still go through the same approval circuit, and someone still re-enters the information in another tool. The team works faster on one task. The process itself has stayed exactly the same.
The tool is there, the results are slow to come
A 2026 Kyndryl study, reported by Le Monde Informatique, surveyed more than a thousand executives on their AI readiness. A large majority say they have integrated AI into their processes. Only a minority consider their teams ready, and few organizations reach the goals they had set themselves.
Look at what your dashboards track today: the number of licenses, of active users, sometimes of requests. Almost never what those uses have changed in a process. And a few months later, you are the one who has to explain to the executive committee what the spending has produced.
Before automating a step, ask whether it should exist
Before automating a step, ask yourself whether it should still exist. A double approval that no longer serves a purpose, re-entry between two applications, a document produced out of habit: these are sometimes better candidates for removal than for automation.
For the other tasks, the question becomes: should it be simplified, assisted by AI, or left under human responsibility? In the meeting minutes example, removing the re-entry brings more than writing faster. The time saved then changes in nature: it is no longer a few minutes on a task, it is a step that disappears.
These decisions are made in short iterations, with feedback from users. No tool will make them for you.
What we did on our own inbound request pipeline
We followed this path on our own handling of inbound requests. It all started far from the code: the idea, the plan and the instructions, prepared with conversational AIs such as Fable 5 or ChatGPT.
The prototype was built around a knowledge base describing the business, its rules and its way of working, surrounded by commands that automate entire steps. We test it, correct it, enrich it.
Once the tool had proved its value, it moved beyond a single person's use: a web application connected to an AI model that the whole team uses, then linked to the company chat (Teams or Slack) to meet people where they already work. Every real use refines a rule or fixes a forgotten case.
The last step adds supervised automation. Agents analyze a request, propose a decision, carry it out within a bounded scope, then check the result. Humans set the rules, approve sensitive actions and take responsibility for them.
The removal test
To see what AI has changed for you, pick a process and imagine it removed tomorrow. If only one task becomes a little slower, you have mostly gained speed. If the whole organization of the work becomes more complicated again, the transformation runs deeper.
It is not a measurement, just a quick way to see where you stand.
