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

A shopping list of features was a slow way to waste money. A shopping list of AI ideas is a fast one, because the tools now say yes to almost anything you point them at. The discipline has to come from the questions you ask before you start.

Most requirements gathering starts with the wrong question. Someone asks the business what reports it wants, or what the dashboard should show, or which fields belong on the form. You write the answers down, and you leave with a list. The list feels like progress. It is usually the start of trouble.

In practice, this is where our AI work comes in, and MCP, AI Agents And The Governance You Now Need covers useful related ground.

That was always true. AI makes it expensive. A shopping list of features was a slow way to waste money. A shopping list of AI ideas is a fast one, because the tools now say yes to almost anything you point them at, whether or not it was worth building. When the tools stop saying no, the discipline has to come from the questions you ask before you start.

So before we talk about Power BI, or Fabric, or an agent, we ask a small number of plain questions about the business itself. Not what it wants built. What it is trying to answer.

The eight questions

We get a leadership team to ask these about their own operation. They sit above the detail, and they do not change much from one year to the next.

1. 1Are our people happy? Whether the team is effective, and whether the systems help them or grind them down.
2. 2Are our customers happy? Complaints, repeat problems, and the experience behind the numbers.
3. 3Is our supply chain ready? Not how efficient it is, but whether it is ready for what is coming. That reframe matters, because readiness is the thing the business loses sleep over.
4. 4Are our internal processes working? Where the handoffs break, where work stalls, where effort is spent twice.
5. 5Are we performing financially where we expect to? Which parts of the operation make money, and which quietly do not.
6. 6Where are we wasting spend? The money leaking out in ways nobody has put a number to yet.
7. 7Are we winning more than we are losing? New sales, retention, and growth in the installed base.
8. 8Can we trust the numbers we already have? Data integrity, and whether anyone believes the figures enough to act on them.

We put that last question first in practice, not last. On one engagement the previous system had left the business with no confidence in its own data, so the trust bar had to be cleared before anything built on top would be believed. This matters more, not less, as AI enters the picture. A model built on numbers you do not trust does not fix the problem. It gives you confident nonsense faster.

The discipline underneath the soft question

Left there, these would be warm words, and warm words build nothing. The value is in what happens next. Each question has to decompose into something you can measure, and then into the unglamorous data capture that has to exist before you can measure it at all.

Take “are our customers happy” on a recent field service engagement. To answer it properly, complaints had to be captured as a proper case type. Jobs needed fault codes. Reschedules and drop jobs needed their own reason codes, because they are different events and lumping them together hides the truth. None of that is exciting. None of it shows up in a demo. But without it, the question cannot be answered, no matter how polished the dashboard or how clever the model sitting on top.

So a good session does two things at once. It keeps sight of the human question, and it is honest about the plumbing that question depends on. Naming the plumbing early is part of the job, so nobody is surprised later by what a simple-sounding question really requires.

How we run the room

The harder part is rarely the technical content. It is the people. A few principles we hold to.

• Ten small irritants matter more than one big technical issue. The thing that grinds a team down is usually not the headline problem on the project plan. It is the daily friction: the extra click, the view that never fits, the report nobody trusts. We go looking for those, because fixing them buys goodwill and shows us where the real pain sits.
• We stay light touch on people. Short sessions, one or two people from a team at a time, scheduled around their operational rhythm rather than ours. We record where people are comfortable, so we never make a busy team explain the same thing twice.
• We try to earn a quick win early. A tidied form, a better view, a queue that finally makes sense. Trust in a room is built on evidence, not promises, and a small visible improvement in the first week does more than any amount of reassurance.
• And we are honest about what can move quickly and what cannot. Some things are a same-week change. Some things need their own governance and sign-off. Confusing the two is how projects lose credibility, so we keep the line clear from the start.

Requirements are built, not extracted

The old model treats requirements as something you extract from people, like taking a statement. The better model treats them as a shared understanding you build with them. By the end of a good session the client understands their own operation slightly better than when they walked in, and can carry the thinking forward without us in the room.

That is the point, and it holds whether the work ends in a dashboard or an agent. The technology at the end changes. The eight questions at the start do not. Get them right and everything downstream, data or AI, is built for a reason someone can name.

We are expanding this into a fuller playbook on how we understand data and AI requirements, with the full framework and the session structure. If you would like an early copy when it lands, get in touch at hello@hoptonanalytics.com.

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

Related reading

• AI Does Not Fix Bad Data. It Amplifies It.
• What a Semantic Layer Is, and Why AI Needs One
• AI Will Not Replace Your Analysts. It Changes The Job.