AI delivery is not magic. It is disciplined execution.

AI solution delivery still follows a familiar enterprise path: problem, requirements, build, test, launch, monitor. What changes is the number of leadership judgement calls packed into every stage. Six questions decide whether that discipline is actually there.

Many conversations about AI make the subject sound more mysterious than it actually is. Strip away the hype and, at one level, AI solution delivery still follows the same path as any other enterprise project: identify the problem, define the requirements, choose the right delivery model, build it, test it, launch it, then monitor and improve.

In practice, this is where our AI work comes in, and Are your people happy? The first question in any Data or AI project covers useful related ground on getting that first stage right.

What changes with AI is not the shape of that path. It is the number of leadership decisions packed into every stage of it. A traditional delivery project asks a leader to sign off scope, budget and a go-live date. An AI project asks for all of that, plus a running series of judgement calls that a project plan alone will not answer for you.

The six questions leadership can’t delegate

We ask a version of these six questions on every AI engagement we run, and we ask a leader to own the answer, not a vendor or a model.

1. 1Is the use case actually suitable for AI? Not every problem needs a model. Some need a rule, a lookup table, or a better report.
2. 2Is the data reliable enough for the decision it will inform? A plausible-sounding answer built on data nobody trusts is still not trustworthy.
3. 3How will model quality be tested, and against what standard? “It looks right” is not a test, and someone has to define what a correct answer looks like before the system ships.
4. 4Where is human review required, and where is it genuinely optional? Deciding which is which, in advance, is a leadership call.
5. 5What happens when the system is wrong? Every system will be wrong sometimes. The question is whether there is a defined path for catching it, correcting it, and telling the people affected.
6. 6Who owns the outcome once the system is live? Somebody has to own what it does next, indefinitely, in the same way they would own a member of staff’s decisions.

None of these questions are technical. They are the same questions good leaders have always asked of any team taking a decision on the business’s behalf. AI just asks them earlier, and more often.

Why data and testing carry more weight than they used to

The second and third questions above deserve more attention than they usually get, because they are where most AI projects actually go wrong. We have written before about why context, not the model, is usually the real bottleneck in an AI build, and about what a mid-market business should expect AI to do well and where it will let you down. Both are worth a leader’s time before a business case gets signed off.

Testing deserves the same discipline. Our AI Trust Framework sets out the rule we apply on every engagement: cite or do not conclude. If a system cannot show its working, it should not be trusted with a conclusion, and that standard has to be agreed and tested against before launch, not discovered afterwards.

Human review is a design decision, not an afterthought

The fourth question, on human review, is where a lot of AI governance conversations start and stop, usually far too late. We think of this as being in the loop versus on the loop, and we have set out what that distinction means in practice for analytics that AI has helped build. Deciding it upfront, as part of the design, is far cheaper than retrofitting a review step once something has already gone wrong.

The same is true of the fifth question. Getting this right is not only a technology question, it is a data protection and risk question too, and one that deserves a straight answer before data starts moving into a tool, not after.

Ownership does not end at launch

The sixth question is the one leadership teams most often assume someone else has covered. It is worth being specific about who that is, by name, before go-live, along with what they are accountable for. That is as much a budgeting decision as a governance one; knowing what a project realistically costs, including the ongoing ownership of it, is worth working through before you start. It is also exactly the kind of decision our strategy and leadership work is built to support.

Technology decides what is possible. Leadership decides what is safe to ship.

That is the difference between AI projects that compound value over time and the ones that quietly get switched off six months in. None of this requires reinventing how your business makes decisions. It requires applying the same discipline you already use for any project that touches customers or money, at the points where AI specifically adds risk.

What to do this quarter

Pick one AI use case already in flight, or under consideration, and run it through the six questions above with the people who will actually own the outcome. If you cannot answer all six with confidence, that is the work to do before the build starts, not after. Our Discovery Playbook sets out the questions we ask before building anything, and it is a reasonable place to start if you want a structured way to run that conversation.

AI projects cannot be managed only as technology projects. They need stronger leadership discipline, at every stage, from the people who are accountable for what the system does.

If any of this sounds familiar, talk to us about your data.

Related reading

Are your people happy? The first question in any Data or AI project

The Data Engineering Layer That Decides Everything

The real risk with AI isn't the model. It's how you use it.