What if the data confirmed what I already thought

Then it is a weak example and you should find another. An analysis that agreed with you demonstrates that you can run a query, not that you use evidence. Look for a time the numbers contradicted your assumption, even slightly, and build the answer around the correction rather than the conclusion.

How do I answer if I used data to influence a decision rather than make one

That works well, provided you describe what the other person believed beforehand and what specifically changed their mind. Name the number that did the work. Answers that describe building a case without saying which figure moved the room tend to sound like presentation skills rather than analysis.

Do I need a big dataset for this answer

No. Five email open rates is enough if the finding was genuinely surprising. Interviewers are assessing whether you form hypotheses, test them honestly, and update when the result disagrees. Scale of data is far less interesting than whether you were ever wrong and noticed.

Tell Me About a Time You Used Data to Make a Decision

Why interviewers ask this

When an interviewer says tell me about a time you used data to make a decision, the answer they get is almost always a story in which the data agreed with the candidate. Somebody suspected a problem, pulled the numbers, the numbers confirmed the suspicion, a change was made, a metric improved. That is not a story about using data at all. That is a story about finding support for a conclusion, which is a different activity with a much worse reputation, and the two are indistinguishable from the outside unless somebody asks the right question. Data has only demonstrably changed your mind on the occasions when it told you something you did not want to hear. Every other time, it agreed with you, and agreement is available from cheaper sources. This is why the follow-up goes straight at the one moment the numbers pushed back.

An answer that sounds fine

We had a suspicion that our onboarding email sequence was too long, because we could see people dropping off partway through. I pulled the engagement figures across the five emails and open rates fell off a cliff after the third one, from around fifty percent down to twelve. So I cut the sequence from five emails down to three, moved the genuinely useful content from the fourth email into the second, and we ran the shortened version for six weeks against a held-back control group. Activation went up about nine percent over that six week period compared with the previous cohort. It confirmed what we had suspected and gave us the evidence to make the change properly, rather than going on instinct and hoping nobody asked us to justify it later.

Where it breaks

Look at the sequence of events. The suspicion came first, the query was written to examine that suspicion, the numbers agreed, the change was made. At no point in this account did anything surprise anybody. A person who only ever asks questions they already half-know the answer to will produce exactly this story every time, and they will believe, honestly, that they are data-driven. The final sentence gives it away completely: it confirmed what we had suspected. That is the candidate telling you the role the analysis played, which was to authorise a decision that had already been made. Somebody describing a time they used data to solve a problem should be able to name the moment the numbers contradicted them, and if no such moment exists anywhere in their examples, that is itself the finding.

The follow-up

What did the data say that your gut did not?

That the drop was on the third email specifically, not gradual. I had assumed people were getting bored over time. It was one email doing the damage, and when I looked at it, it was the one asking them to invite their team before they had done anything themselves. So, yeah, my instinct was that the sequence was too long. The sequence was not really too long. One email was in the wrong place.

That is a genuinely different story and it arrived on the second question. His hypothesis was length. The data said sequencing. That means his original fix, cutting five emails down to three, was partly solving a problem that did not exist, and he shipped it anyway because it was already drafted. The activation improvement might have come entirely from moving one message rather than from removing two. He does not know which of the two moved it, because the change went out as a single bundle and there was no way afterwards to separate them. That is the honest version and it is considerably more interesting than the one he prepared, because it contains a thing he got wrong and a thing he would now do differently.

An answer that holds up

I thought our onboarding sequence was too long because people were dropping off, so I pulled open rates across all five emails expecting a gradual decline. It was not gradual. It fell off a cliff at email three, from fifty percent to twelve, and email three was the one asking people to invite their team before they had done anything in the product themselves. My hypothesis was length and the answer was sequencing. I cut the sequence to three and moved the invite ask to the end, and activation rose about nine percent. What I would do differently is ship those separately, because I bundled a change I had evidence for with a change I had a hunch about, and I still cannot tell you which one moved the number. My instinct was wrong in a specific and useful way, and I would have missed it if I had only looked at the total drop-off.

The interviewer asks what he would have decided without the data.

I would have cut it from five emails to four and left the invite ask sitting at position three, which was the actual problem. The change would have looked reasonable, the number would have moved slightly, and I would have concluded I was right.

Practice this question

Say your answer to this question out loud, then get asked what the numbers told you that your instinct did not. Answer it now, free, no account.