Working through a puzzling finding

One slide in a deck I was working on this week was unusually difficult.
A research team had completed a strong analysis and asked me to help translate the findings into an executive presentation. But one finding contradicted both intuition and everything we’d seen in previous analyses.

The first question in that situation is always the same: is this a real signal, or a problem with the data?

If it’s a real signal, you have a problem. Include a surprising finding you can’t explain, and you risk undermining the credibility of everything else in the deck. Include it with a few speculative hypotheses, and leaders may find it interesting — but not something they feel confident acting on.

This kind of moment comes up in a lot of work, not just data analysis. You’re deep into something and you hit a result, a response, a number that doesn’t fit. You can paper over it or you can work through it. Working through it is harder.

Too often, people complete strong work and stop just short of this step. Not because they don’t care about the answer — but because this stage can feel overwhelming. The space of possible explanations is large. Progress is slow. And there’s a nagging worry that you might be forcing meaning onto something that doesn’t have any.

So you move on. The puzzling finding gets a footnote, or gets cut, or becomes one of those lingering questions that nobody quite resolves.

It’s an uncomfortable place to work. You’re expending effort with no guarantee it’s going anywhere.

I asked: what could plausibly explain this finding without contradicting what we already know? Then: how could we test those explanations?

One by one, I checked them against deeper cuts of the data. Some fell apart immediately. Others got closer but introduced new contradictions. One explanation kept holding up, and as I pushed on it, the picture got clearer rather than messier.

This took longer than it would have if I’d just flagged the anomaly and moved on. But each exchange gave me a concrete next question. I was never staring at a blank screen wondering what to try. The work felt like it was moving — and because it felt like it was moving, I kept going.

Eventually the pattern held. What had looked like a problematic anomaly turned out to be one of the most compelling insights in the entire deck — one we could explain with confidence, not speculation.

The insight didn’t come from the first AI response. It emerged from the rounds of thinking that followed. And those rounds happened because I never hit the wall where the effort stops feeling worth it.

There’s a temptation to treat AI like a vending machine: put in a prompt, expect an answer. When you’re working through something genuinely unresolved, that doesn’t work. The answer isn’t in there yet — it’s still being built.

Use it like a sparring partner instead. Ask it to propose explanations. Pressure-test them. Adjust your thinking. Try again.

The goal isn’t to get AI to do the thinking. It’s to stay in the thinking long enough for something real to emerge. That’s harder than it sounds when the problem is messy and progress feels uncertain. A good sparring partner makes it possible to keep going.

The difference between asking AI for an answer and using it to work through a problem isn’t a small one. It’s usually the difference between an insight that moves people to act and one that is quickly forgotten.