skip to content

How do you quantify a result in an interview story when the original dashboards are gone?

level: seniorimportance: must knowfreq 66%

answer

  1. Something else survived the dashboard
  2. Estimate out loud, never silently
  3. Commit to a range, not a point
  4. Proxy, counterfactual, confidence band
  5. Name the source of the estimate first

basics

~20 s

Reconstruct from something that survived — ticket counts, rota sheets, release notes — state the figure as a range rather than a point, and say out loud that it is an estimate and where it came from. A labelled estimate outscores an invented precise number.

solid answer

~50 s

I rebuild it on three columns: a **proxy** that survived and moves with the thing I care about, a **counterfactual** for what the number would have been without the work, and a **confidence band** I would defend under pushback. For the triage automation the dashboards were gone, but the ticket tracker still held the triage queue, so I say: 'I no longer have access to those dashboards, so this is an estimate from the ticket volume.' Then I give the range — roughly 30 to 40 hours a month of rota time before the change — and the surviving hard edges: 214 open findings down to 38. Hedging the estimate and being exact about the parts I can still evidence reads as rigour; producing one clean percentage for something nobody measured is the thing that gets probed apart.

go deeper

for a junior

Know that 'I estimated this, and here is where the estimate came from' is a legitimate and expected answer. Never fill a gap in your memory with a confident invented figure.

for a middle

Be able to name the surviving artefact your estimate rests on and how it converts into the unit you quoted — tickets into hours, releases into days — without fumbling the arithmetic aloud.

for a senior

Demonstrate judgment in choosing the proxy and honesty in the counterfactual, including the part that shrinks your own claim, and commit to a band you would still defend when pushed from both ends.

for a principal

Own the limits of the evidence itself: say what the reconstruction cannot establish, what a sceptical reviewer would attack first, and why the conclusion still holds despite that.

## The situation Most strong stories are one or two roles old. The dashboards that measured them are behind an account you no longer have, the wiki page is gone, and what remains is memory. This is the moment where candidates quietly invent — a clean percentage arrives, delivered confidently, and it is the single most common integrity failure in behavioral answers. It is also unnecessary, because the honest reconstruction usually scores better. ## The reconstruction worksheet Before the loop, take each story whose metric is gone and fill three columns. **Column 1 — the proxy.** What survived that moves with the thing you care about? For the triage automation, the dashboards were gone but the ticket tracker still held every triage ticket ever opened and closed. Ticket volume is not rota hours, but it moves with them, and you can say exactly how: each ticket took a human roughly a quarter of an hour, so a monthly count converts into a defensible hours figure. A credible proxy has three properties: it survived, it moves with the target for a reason you can state in one sentence, and it is not something you chose because it flatters. A proxy you cannot justify mechanically is a guess wearing evidence's clothes. **Column 2 — the counterfactual.** What would the number have been if the work had never shipped? Alert volume was growing as services were added, so the honest counterfactual is not 'the rota would have stayed at roughly 30 to 40 hours a month' but 'it would have kept climbing with the estate'. Counterfactuals cut both ways and stating one that reduces your claim is exactly what buys credibility for the parts that increase it. **Column 3 — the confidence band.** The range you would still defend if pushed from both ends. Not a comfortable spread — an honest one. 'Roughly 30 to 40 hours a month' says the bottom of that range is a figure you are confident about and the top is one you would not exceed. A band so wide it excludes nothing is as useless as a fabricated point. ## Saying it out loud The worksheet is preparation; delivery is one sentence in front of the number: 'I no longer have access to those dashboards, so this is an estimate from the ticket volume.' Then the range, then whatever hard edges survived — 214 open findings down to 38, the median age of an unresolved high-severity finding moving from 19 days to 6 — because a story usually holds a mix of remembered and evidenced facts, and the interviewer wants to know which is which. Candidates worry that hedging weakens the answer. It does the opposite for anyone experienced across the table. A hedge tells them you can distinguish what you know from what you believe, which is the same discipline they will need from you in an incident review. What weakens an answer is an unhedged number that dissolves under one follow-up question. ## Under probing Expect three probes and prepare all three: - *How did you measure that?* Answer with the proxy and the conversion, in that order. - *Why that proxy?* Answer with the mechanism, not the convenience. - *What would have happened anyway?* Answer with the counterfactual, including the part that shrinks your claim. The reconstruction survives all three because it was built to. An invented figure survives none of them, and the failure is visible in the pause before the answer. ## What not to do Do not present a reconstruction in the voice of a measurement. Do not convert an estimate into a percentage — proportional form implies a precision the estimate does not have. Do not hedge every sentence until no claim remains; the point of the band is to still commit to something. And do not reach for someone else's remembered figure, because you cannot defend a number whose derivation you never saw. ## Why this is worth the preparation An interviewer cannot verify your numbers, so what they are really assessing is your relationship with evidence. A candidate who says 'roughly 30 to 40 hours a month, estimated from ticket volume, and here is why that proxy holds' has demonstrated more than a result — they have demonstrated how they will report a number to their next team when the real answer is uncomfortable.

  • Doesn't saying 'this is an estimate' make your result sound weaker?
    It sounds weaker for about two seconds and stronger for the rest of the loop. Labelling the estimate shows I can separate what I measured from what I inferred, which is the same discipline an incident review needs. The version that actually sounds weak is a confident precise figure that dissolves the moment someone asks how it was measured.
  • What makes a proxy credible rather than merely convenient?
    Three things: it survived independently of me, it moves with the target for a mechanical reason I can state in one sentence, and I would still have chosen it if it made my result look smaller. Triage ticket counts qualify because each ticket cost a human a roughly known slice of time. A chat-message count would not.
  • How would you defend the top end of that range if I pushed on it?
    The top comes from the busiest months in the ticket export, converted at the same per-ticket time as the rest. I would not go above it, and I would drop to the bottom of the range if you assumed the rota was faster per ticket than I remember. That is why I give the band instead of a single figure.
  • What would the number have been if you had never shipped the automation?
    Worse than the baseline, not equal to it. The alert volume was climbing as services were added, so the rota cost would have kept rising rather than sitting at roughly 30 to 40 hours a month. I say that explicitly because the counterfactual is the part candidates usually skip, and skipping it makes the whole claim look untested.

Reconstructing a lost metric is like dating a felled tree by its rings when the planting record has been destroyed. You are not guessing; you are reading a surviving record that moves with the thing you actually want to know, and you state your answer as a range because the ring count justifies exactly that much precision and no more.

saying these in an interview costs you the question

  • Producing a precise percentage for something never actually measured
  • Delivering a reconstruction in the voice of a real measurement
  • Choosing a proxy that moves for reasons unrelated to the work
  • Hedging so heavily that no claim survives the sentence
  • Widening the range until it excludes nothing at all
  • Repeating a teammate's remembered figure you cannot derive

context