Will your next AI pilot scale, or stall? Find out before you spend anything.

Fourteen questions you can answer from your chair. Four about your situation, nine about your readiness, one about what would help. About six minutes. You get a score, the dimension that decides the outcome, and three things to do on Monday. Scored in your browser.

Built by Fredrik Holmboe / Dualia.

You are looking at a colleague's result. Answer yourself to compare. Their score sits next to yours once you have.

Your situation

1. What made this worth the next six minutes?If several apply, pick the one with a date on it.

A date: a leadership meeting, budget round, launch or event where AI has to be answered. Something we started has stalled, and I have to decide what happens to it. I have been asked to come back with a proposal on AI. A question I could not answer, from legal, compliance, IT, data protection or a customer. A vendor has proposed something, and I want a second opinion first. Nothing specific. I am keeping up.

2. What should the AI take over first?

Producing or adapting content: documents, campaigns, training material, approved wording. Answering questions from customers, patients or healthcare professionals. Follow-up after meetings: call reports, personalised follow-up, CRM updates. Orchestrating channels: who gets what, in which order, and measuring it. Internal analysis or reporting. We have not chosen a task yet.

3. The last time your organisation put AI on a real task, what happened?

It is in daily use, with someone accountable for it. It worked in the demo, then stopped when someone asked who signs off, what happens when it is wrong, or where the data goes. The output was not good enough, or the tool did not fit. Dropped. It never became the normal way of working. It faded, or the licences sit unused. Only individuals trying tools. Nothing on a shared task yet.

4. Your own role in this:

I would own the outcome. I would run it day to day. I would have to approve it: legal, compliance, medical, IT or data protection. I am looking on behalf of someone else, or out of interest.

Ownership

Pick the lower option when in doubt.

5. Is there one named person whose job it is to carry the outcome of this pilot, not just to coordinate it?

Yes. One person, named, and their manager knows it is theirs. Several people are involved, but no single owner. No one in particular. It would be somebody's side project.

6. When the pilot needs a decision that costs money or changes how people work, do you know today who signs it?

Yes. One name, and they have signed something similar before. Probably, but we would have to find out who. It would go up the chain, and that is where it would sit.

Inputs

7. The material the AI would work from, your documents, templates, records, approved wording: where does it live, and is it written down anywhere what a correct output looks like?

In one place, kept current by a named person, and what good looks like is written down. In one place and current, but what good looks like lives in people's heads. Across a few systems. Mostly current, nobody formally owns it. Spread across inboxes, drives and individual people.

The task

8. The task you want AI to take on: is it done the same way every time by the people doing it now?

Yes. There is a defined way, and people follow it. Roughly the same, with meaningful personal variation. Everyone does it their own way. We have not chosen the task yet.

Room to run

9. For a pilot of a few weeks, including outside help: is there money in this budget period, and real working hours for the people involved?

Both. Money allocated or under an existing line, and hours set aside with something else dropped. Hours, not money. It needs a case, or the next budget round. Money, not hours. Everyone is at capacity; it would sit on top. Neither yet.

Review

Not a legal review: who stands between the output and the outside world.

10. Before AI-assisted output reaches anyone outside the team, a customer, a patient, a healthcare professional or a regulator, is there a named person who reviews and approves it?

Yes. The review step exists, is named, and is documented. Someone would look at it, informally. No review step has been defined.

11. When the output is wrong, and some of it will be, is it decided who catches it, who corrects it, and who is accountable for what already went out?

Yes. Decided and written down, including how we would stop the pilot. We have talked about it. Nothing is written down. We have not thought about it.

Data

12. Is it decided which data may go into the AI tool, which may not, and who checked that against your data protection and information security rules?

Yes. Decided, checked, and written down. We have a general policy. Nobody has applied it to this pilot. No. We would find out as we go.

13. Has anyone classified your intended use under the EU AI Act, that is, decided which obligations apply to it, if any?

Yes. Classified, and the conclusion is written down. We have discussed it. Nothing is written down. We have not looked at it.

Last one

14. What are you looking for?

A pilot we run ourselves afterwards, with the governance built in. A partner to run this for us for the foreseeable future. A platform or tool to buy. Training, so the team can start. Not sure yet.

0 of 14 answered

0 / 24

Where this points

Book a 30-min intro

Three questions for any vendor

Who signs off on the output, on our side?
What happens when it is wrong, and who is told?
Where does the data go, and who checked?

Optional: send your answers to Fredrik, so a call starts from them instead of from scratch.

First name

Work email

Company

Job title

Organisation size

Your sector

When is it?

Fourteen questions. Nine of them are scored, across six dimensions, four points each. The score is a conversation starter, not an assessment. The dimension with the lowest score is the one worth acting on first, whatever the total says.