A Metric That Overstates Growth Three Days Before the Board
Three days before a board meeting you learn the customer-growth dashboard overstates growth. A junior analyst found that 4,100 of the 31,000 "active" accounts are duplicates created by a CRM migration in October; deduplicating would move reported growth from 12.4% to roughly 9.1%. The CFO used 12.4% on an investor call last week and the board deck was circulated last night. Data engineering says a corrected dataset is feasible in about 18 hours if they pause the Q1 pipeline work due Friday. The analyst is worried about being blamed and has asked you not to name her in the writeup. The CEO is traveling and reachable by phone.
1. What is the first thing you do, and what is the sequence over the next 72 hours?
Read the worked answer
Verify the 9.1% before you escalate it. The analyst is probably right, but a three-point revision to a number already spoken publicly must be reproduced independently - a second analyst, the same definition, the same window. You do not want to walk into the CEO's call with a number you cannot defend line by line.
Sequence:
- Reproduce the deduplication and document the rule (which record survives, on what key, what the definition of "active" is). Half a day.
- Call the CEO - by phone, before the deck is discussed by anyone else. Lead with "here is what we know, here is what we're confirming, here is my recommendation," not with chronology.
- Brief the CFO before the CEO calls the CFO. A CFO who learns second that their public number is wrong becomes a problem you cannot fix with analysis.
- Correct the deck with a clearly marked revision and a one-line note on cause and fix.
- Decide on the investor call with Legal and the CFO - this is their call, not yours; your job is to give them the facts early enough to have it.
The failure mode is spending 72 hours getting the number perfect and delivering it after the meeting.
2. Data engineering can produce the corrected dataset in 18 hours only by pausing the Q1 pipeline due Friday. How do you decide?
Read the worked answer
Ask what the 18 hours actually buys. Two distinct uses:
- A defensible headline number for the board. This needs the deduplication rule and a reconciled count, not a rebuilt pipeline. That is often a day of analyst work, not a day of engineering.
- A permanently corrected dashboard. This is the real fix and it is not due Friday or Monday; rushing it is how you get duplicate record #4,101.
Separate them. If analysts can produce a reconciled, defensible figure from a query over existing tables, the pipeline does not need to stop and the conflict evaporates. Only stop the pipeline if the corrected number is genuinely unreachable without it - and if so, that is a decision for whoever owns the Friday commitment, escalated explicitly with the trade named, not absorbed silently by the data team.
The principle: an emergency justifies consuming someone else's capacity only when there is no smaller intervention that resolves the emergency. Here there almost certainly is, because the board needs a correct number, not correct infrastructure.
3. The analyst asks not to be named. How do you handle attribution, and what do you do about the underlying control failure?
Read the worked answer
Grant the request on attribution, refuse it on substance.
The writeup says "an analyst identified the discrepancy through routine validation" - that is true, it protects her, and it costs nothing. Attribution is not what the board needs; the finding and the fix are. Retaliation risk is real and the cost of avoiding it is a few words. Do not, however, let anonymity become a reason the finding is softened or the timeline is vague.
On the control failure, be specific and own it upward: a migration in October introduced duplicates, and the growth metric was reported for roughly five months without a reconciliation check that would have caught a 13% inflation. The analyst did not create the problem; the absence of a validation control did. Name the control - a duplicate-rate check on the active-account count, with a threshold and an owner - and give it a date.
One thing to say plainly to the analyst, and to the team afterward: finding this was the correct outcome, and the person who finds the error is never the problem. If your team learns otherwise, the next discrepancy surfaces at the board meeting instead of three days before it.