With AI helping your team write queries and build charts, the analysis that used to take a week now takes an afternoon. You can investigate the follow-up questions during the meeting, and have AI draft a summary before everyone has left the room.
Then you have essentially the same meeting again next week.
If that sounds familiar, I think it’s worth asking whether the analysis was actually the thing holding up the decision. Making it faster is useful, but there may be another part of the work that nobody has taken responsibility for.
A correct answer can leave a real disagreement#
Imagine a business considering a promotion to bring back inactive customers. Marketing wants growth, operations is worried about delivery capacity, and finance wants to protect margin. Each team asks its AI assistant to investigate.
For this example, assume they’ve already agreed the definitions. They’re using the same customer population, costs and assumptions about the promotion.
The analysis might show that the promotion would probably increase orders, put pressure on capacity and reduce margin per order. Those findings could support all three teams’ positions. None of them has to be misunderstanding the numbers for the disagreement to be real.
Choosing what to do requires a view on priorities, acceptable risk, timing and who will do the work. It also means accepting uncertainty about what will actually happen. There may be no option which makes every team better off.
An AI can recommend a choice if you give it an objective and constraints. The organisation still has to agree those priorities, or give someone the authority to choose between them.
An awkward meeting can be part of the work#
I think this is one of the parts of a data leadership role that’s easiest to miss if you describe the job as “producing analysis”.
Someone needs to notice when two people are arguing about different things. They need to make room for an objection that hasn’t been said clearly, test whether an apparent agreement will survive outside the meeting, and establish who can make the call when the teams can’t agree.
Sometimes that means telling a senior stakeholder that the next chart won’t resolve the problem. Sometimes it means making it explicit that one team’s target needs to give way to a more important business priority.
That takes relationships, judgement and enough organisational backing to have the conversation. The person facilitating it doesn’t necessarily own the final decision, but they need a route to someone who does.
AI can already help with parts of meeting facilitation. Meeting tools can keep notes, track an agenda and summarise decisions and open questions. It can also help prepare options and questions for the discussion.
I wouldn’t currently give it responsibility for getting a contentious business decision over the line. A convincing summary isn’t evidence that the people involved have accepted the trade-offs or committed the necessary resources. Someone needs to check that, and deal with what happens if they haven’t.
Work out what is actually blocking the decision#
Before commissioning another round of analysis, I’d try to make the unresolved question explicit:
- Do we disagree about the facts? Identify the missing evidence, whether it can realistically be obtained, and what finding would change the decision.
- Do we disagree about what we’re measuring? Agree the definition and comparison before producing more numbers.
- Do we disagree about what matters most? Make the competing objectives and consequences visible, then involve whoever can resolve that trade-off.
- Is nobody able or willing to decide? Establish who owns the decision and when it needs to be made.
These can look remarkably similar in a meeting. “I don’t trust that number” might mean there’s a genuine error, a different definition, or that someone doesn’t accept the proposed action. It deserves investigation, rather than an automatic assumption that the stakeholder is being difficult.
The distinction about definitions is the subject of why AI analysts can disagree about how many customers you have. Here, the important point is that even a shared definition and sound evidence don’t remove a disagreement about priorities.
Use AI to make the conversation easier#
There’s plenty of useful work for AI here. It can help collect the previous decisions, compare options under different assumptions, and prepare the strongest case for each team’s position. That can make it easier to understand why a reasonable person might disagree with your preferred answer.
I’d go into the conversation with a clear decision to make, the options being considered, the material uncertainties, and the person who owns the call. I’d also ask what happens if you wait: sometimes more evidence is worth the delay, and sometimes waiting is itself an expensive choice.
Use AI to reduce the preparation and administration around that discussion. Then spend the time you’ve gained working through the actual disagreement. Another set of charts is often a much more comfortable thing to produce, which is part of why this can be easy to avoid.
The aim doesn’t have to be unanimity. People can understand a decision, disagree with it, and still commit to carrying it out. Make the decision and any remaining disagreement explicit so the meeting doesn’t end with several different versions of what was agreed.
Close the decision and keep a basis for judging it#
A useful decision record can be quite short. It needs to say what was decided, who owns the action, why that option was chosen, and what happens next. Record the important assumptions, the measure of success, and when you’ll review it. The people responsible should confirm it reflects the commitment they made.
For the promotion example, agreeing to “improve retention” isn’t enough. Agree the customer group, the measure, the comparison period, and what level of cost or service impact you’re willing to accept. Keep the baseline and the evidence used to make the choice.
That also gives you a way to reopen the decision constructively. You can ask whether an assumption changed or the outcome fell outside the agreed bounds, instead of starting the original argument again. Keep the measurement consistent, or explicitly explain a change, so the review tests the decision you actually made.
This is one reason I don’t think automating the tasks of a data team removes all of its responsibilities. A good data leader can help the organisation work through uncertainty and reach a decision. Business leaders still own the priorities, commitments and consequences.
If your analysis is faster but decisions aren’t, I’d look at one stalled decision and identify exactly what’s left unresolved. The next useful piece of work might be a conversation that someone needs to lead.
