AI projects usually stall on the people around the tool. The licences are bought and the pilot is running, but nobody owns the result, the decision paths are unclear and the people who will use the agent have not been part of shaping it. Start by finding out where it stopped, pick one team, name an owner and agree three measures before anything is built.
How it looks
The licences are bought and several pilots are running. In the review meeting three departments say they are waiting for each other, and each has a good reason. None of the reasons is about the tool.
It is a common situation and there is something you can do. The first step is to stop asking which tool is missing and to ask where the work stands still.
Five common causes
- Nobody owns the result. The pilot has a project lead, but nobody answers for something actually getting better in daily work.
- Decisions are made in one place and influenced in another. The person who formally decides is rarely the one who decides whether the proposal goes through. It shows only when something sits still.
- The people who will use the agent were not involved. What is built without them rarely fits how they work, and then it goes unused.
- Trust is missing. Worry over quality, over data ending up in the wrong place or over one's own role is seldom said out loud, but it shows in how little the tool is used.
- The pilot is too big and is not measured. When nobody knows what should get better, nobody can see whether it did.
What to do: four parts with a stop point after each
A bounded package for one or two teams works better than a large programme. Four parts are enough, and after each part you decide whether it is worth going on.
- Baseline. A working session with whoever owns the AI effort and whoever owns leadership and culture. You choose the team, name an owner and set the baseline for three measures.
- Map. Conversations at several levels, alongside an analysis of the workflow. It shows where ideas, information or trust get stuck.
- Shared work. The team shapes the way forward itself and picks the first use of the agent.
- Build and follow-up. The agent is built with the same team, in the tools you already have and inside your security rules. The sceptics see it work in their own week.
Three measures you agree first
Use three measures everyone can understand: how much the agent is used, how much time it saves and a short pulse on trust. Measure before and after. You then have a story to show the steering group, with a recommendation on how to go on.
First week: three questions you can ask yourselves
The questions can be taken on your own in a week. If you need someone who listens at several levels and builds the agent together with the team, we can help.
- Who answers for it getting better in daily work? One name, and not a project group. If you cannot write a name, that is the first thing missing.
- Where is someone waiting for someone else? Follow one case from start to finish and note every time it sits still. The answer is often there.
- What would the most sceptical person need to see? Ask them directly. The answer tells you what the pilot should show first.
Who you need on your side
Two sponsors, one from the AI side and one from HR or leadership, a pilot owner in the team and access to the people and tools in the engagement. It is a shared question for the AI team and HR, which is why the pilot is run together.
How we work with it
We call it AI Adoption. We run the mapping and build the agent, and together with facilitators from 21 Dimensions we lead the conversations and the workshops. The price is fixed per part and agreed before the part starts. We sign a data processing agreement before we touch any data. The length is set at the start, and you can count on weeks and not years.
Common questions
How do we know our AI project has stalled?
One sign is that the pilot runs but usage stands still, that departments wait for each other or that nobody can say what got better. Another is that the people who will use the agent have not been part of shaping it.
Do we need to change tools?
Usually not. The cause normally lies in ownership, decision paths and trust, and a new tool does not help with those.
How big should a first attempt be?
One or two teams. A small bounded attempt with three measures can be evaluated, and you can stop after each part.
Who should own the pilot?
Someone on your side who wants the result to be good. A supplier can help, and the owner sits with you.
How long does it take?
Count on weeks and not years. The length is set together at the start.
Can we do this ourselves?
The three questions in the first week can be taken on your own. If you need conversations at several levels and an agent built together with the team, we can help.