Choose a question your team already needs to answer. Give the experiment a decision, an owner and a boundary.
We care about the question you are still carrying home after the meeting. The margin nobody can explain. The cash number that changes when a different spreadsheet opens. The customer commitment that needs an answer before your team has finished assembling the facts. That is where we want an AI conversation to begin. You already have a business to run; another open-ended technology project has to earn its place.
Our view is that a good first engagement should reduce uncertainty about a decision you actually own. We would ask you to name that decision before showing you a tool. If we cannot explain who will use the answer and what they might do differently, we have more work to do on the brief.
Define the work
Consider an illustrative business whose monthly meeting repeatedly returns to a margin decline. The first useful AI task might be to organise the evidence behind that movement. It could bring together the ledger, a product mix view and management notes, then distinguish supported explanations from open questions. That is a much more specific ambition than automating finance. It also gives the team something concrete to assess.
A good starting question contains a decision, an owner and a boundary. The decision might concern pricing or a collection priority. The owner is someone who can recognise whether the answer is useful. The boundary identifies the period, business unit and information the workflow should use. This turns experimentation into a manageable piece of work rather than an open-ended search for impressive output.
Our view is that a good first engagement should reduce uncertainty about a decision you actually own.
There is an emotional benefit to this precision. People can approach AI with curiosity when success is understandable. They do not have to accept a sweeping claim about the future of their role. They can inspect a familiar task and ask whether a different method helps. Their experience becomes part of the evaluation, which gives them an active place in the change.
Make success inspectable
We also care about the person who currently holds this process together. The colleague who knows which export is reliable, who remembers the adjustment, who catches the mistake before it reaches you. We want their judgment in the design. Asking them to help define success is a better start than presenting them with a finished answer they are expected to trust.
Once the question is chosen, agree what evidence would support success. That could include reconciliation to a known report, a clear list of exceptions and an explanation of unresolved assumptions. Measure the time to produce an accepted result, including review. A fast first draft is encouraging; the real value arrives when the responsible person can use it.
The starting point also shapes the next investment. A focused experiment can reveal that the main obstacle is a missing account mapping, an unclear approval or a report nobody actually uses. Those discoveries are useful outcomes. They make a later integration more informed and may prevent investment in the wrong process.
Our starting question for you is simple: what keeps coming back to your desk because nobody has a clear enough answer? Tell us what is at stake, who knows the work and what a useful result would let you do. We can build a sensible first scope from that.
Examples in this essay are illustrative, rather than accounts of client engagements.