“Which AI tool should we buy?” It is often the first question a leader asks us. It is rarely the right one. It puts technology at the center, when the success of an AI project almost always depends on something else: how clear the problem is, the state of the data, and how the organization will decide on and take responsibility for what the system produces.
So before any project, we ask three questions. Always in the same order.
1. What problem, measured how?
A good AI project starts with a business pain point you can put a number on: hours spent consolidating spreadsheets, an error rate, a customer response time, a cost. “Doing AI” is not a problem. “Halving the time it takes to produce the weekly report” is.
That baseline is used three times: to choose between several use cases, to decide whether the project is worth the investment, and to know, at the end, whether it worked.
- Who suffers from the problem today, and how many times a week?
- What does it cost, in time, errors or money?
- What would success look like in six months, in a single number?
2. What data, and who owns it?
An AI system is only as good as the data it receives. Three checks are enough to avoid most bad surprises.
- Existence: is the data actually collected, or only assumed to be?
- Quality: is it complete, up to date and consistent from one source to another?
- Access: who is responsible for it, and can it be accessed legally and technically?
Sometimes the honest answer is “not yet”. In that case, the first workstream is not AI but data. It is a valuable conclusion, and a much cheaper one when it comes before the project rather than halfway through.
3. What governance?
An AI system will eventually make mistakes. The point is not to prevent that, but to know what happens when it does.
- Who decides to deploy the system, and who can stop it?
- Who answers for its results to management, customers or the regulator?
- How is an error detected, corrected, then prevented the next time?
These answers are not an IT matter. They belong to leadership. That is why governance comes before the first prototype, not after the first incident.
If one of the three answers is missing, it is not an AI project yet. And it is very good news to find out early.
An example: reporting at a major industrial group
For a major industrial group, the three answers were clear from the start. The problem: operational reports consolidated by hand every week, at the cost of many hours and copy errors. The data: already produced by the teams’ tools, but scattered. The governance: an identified business team, owner of the report and of its validation.
The automated reporting system could therefore be scoped, built and documented without detours. The result: more than 70% of time saved on report production, and fewer errors.
Why this order
Starting with the data risks building a solution in search of a problem. Starting with governance means framing a project whose value is not yet known. The problem comes first because it gives meaning to the other two questions.
On video
These three questions, presented in an interview with Malt about why so many corporate AI projects fail.
Key takeaways
- A quantified business problem, before any choice of tool.
- Data that exists, is reliable and is accessible.
- Clear governance: who decides, who is accountable, what to do when it goes wrong.
Let’s test these three questions on your project, in 30 minutes.
Book a callMore notes