“It was all me”: finding what a candidate really did

When a candidate says "I built that whole thing," you have learned almost nothing. You have heard a claim. The work of the interview is turning it into facts, and most interviewers stop one question too early.

Why "I did all of it" tells you so little

Because it is a claim with nothing attached. It may well be true. In the interview room, nothing supports it yet.

One rule works for companies and people alike: sort what you hear into claims and facts. A number a company reports about itself is a claim until an independent source backs it. That rule is in my glossary. Applied to interviews, every "I led," "I owned," "I drove" goes into the claims pile first.

Most people who say this aren't lying. Memory does it for them. The longer you spend on a project, the more of the team's work drifts over to your side of the ledger. People from big companies have an extra blind spot. The brand, the budget and the existing customers were doing part of the work, and from the inside that is hard to see.

In Japan there is a twist. Many candidates do the opposite and say "we" for everything, even when they made the call. The habit is modest and polite, and it hides contribution just as well. So the task runs both ways. Separate "we" from "I," whichever one the candidate leans on.

Split "we" into "I"

Ask about things that happened, and keep going until you reach a personal action. That is the whole method. The behavioral event interview developed by the psychologist David McClelland and the familiar STAR format share the same root: ask about a real past event, never "what would you do if." The name of the technique matters less than the direction of the follow-ups.

Push in five directions.

Personal actions. "You said conversion went up. That week, what did you yourself do?" If the subject stays "we," ask once more: "Within that we, which part was yours?"

Decision points. "What other options were on the table? Who made the final call?" Someone who can tell you which cards they held and which one they threw away was usually sitting in that seat. Someone who can't may have been executing another person's decision. Execution is a real skill. Score it separately from leadership.

Numbers. "From what to what? Was the market growing at the same time? Where was your predecessor when you took over?" A good growth curve is usually three contributions stacked on top of each other: the market, the predecessor and the candidate. Ask about all three and you can see which slice is theirs.

Failure. "Where did it nearly go wrong? What did you change?" A story with no trouble in it is the least useful kind. How someone judges and adjusts under pressure shows you their ceiling. I wrote about this layer in Telling failure as evidence.

Other people's roles. "Who pushed back on you? If you had left, who could have taken over?" A person who really ran the work can tell you what each teammate did, sometimes including who was better than them. A candidate who casts everyone else as an extra deserves a closer look.

Is this worth the interview time?

Yes. Interviews with fixed questions and fixed scoring predict job performance better than free conversation, and the research has said so for decades.

The classic reference is a 1998 meta-analysis by Frank Schmidt and John Hunter in Psychological Bulletin. It put the validity of structured interviews at 0.51 and unstructured interviews at 0.38. In 2022, Paul Sackett and colleagues redid the numbers in the Journal of Applied Psychology. They argued that earlier work had overcorrected for range restriction. With their correction, both figures fell: 0.42 for structured interviews, 0.19 for unstructured ones.

Look at what moved. In the new estimates, structured interviews came out on top of the methods they reviewed. Unstructured interviews lost about half their validity. The comfortable feeling that a conversation went well is worth even less than we thought.

The revision is itself disputed. In 2023, Oh and colleagues challenged the correction method, and Sackett's group replied in the same journal. Most of the argument is about cognitive ability tests. The exact numbers will probably keep moving. The direction, structure over no structure, holds in both versions.

Pushing in five directions is the cheapest way I know to add structure. No assessment vendor required. You change how you ask and what you write down.

Put what you dig out back into the four layers

The events you collect go back into the four-layer model for reading people: core, patterns, equipment and context.

Personal actions and numbers mostly tell you about equipment, meaning what the person can actually do. How they judged things when it went wrong is the patterns layer, which sets their ceiling. Why they took the project on, and why they later left, touches the core.

Context is the layer people skip. A lot of "I did all of it" cannot be repeated somewhere else. The old employer's brand opened doors. The old boss absorbed the hits. When you ask the fifth question, about other people's roles, listen for how much is left once that environment is gone. If your company can't offer the same setting, that needs deciding before the offer, and the fit check looks at the weakest layer for exactly this reason.

In After the resume screen: what I look for in an interview, I described following one event through its starting conditions, trade-offs, actions and costs. This piece looks at a single seam inside that: the line between "we" and "I." Used together, every row on your scorecard should end up with an event attached.

One problem I haven't solved. People who explain badly haven't necessarily done less. Candidates interviewing in a second language, quiet people, people bound by confidentiality all lose points on these five questions. And a well-coached candidate can answer all five without a stumble. Follow-ups separate claims from facts. They don't separate "really did it" from "rehearsed it well." The best available move is to come back to the same event later from another angle, note where the details drift, and leave those for the reference check.

At your next interview, add one column to the scorecard called "event." If a row stays empty, don't score it yet.