- Define the impact
- Inspect evidence
- Propose a correction
- Verify recovery
Use this sequence to organize the investigation. Confirm the details in the affected platform and campaign.
An interviewer may describe a campaign with no delivery, two reports that disagree, or a video ad that fails on one device. These questions test whether you can turn an incomplete complaint into an investigation. The scenarios below are original practice examples, not a list of questions guaranteed to appear in an interview.
Adjust your answer to the role. Publisher operations emphasizes trafficking and delivery; programmatic operations adds buyer and auction evidence; video work requires player-level investigation; yield work requires revenue comparisons. Be clear about which systems you have used and where you would consult documentation.
Use a repeatable answer structure
Begin by confirming the objective, impact, and affected scope. Then move from symptoms to evidence. A short, well-supported answer is more useful than naming every setting you know.
- Clarify the campaign goal, platform, time window, and affected segment.
- Describe two or three plausible causes and the evidence that distinguishes them.
- Choose the smallest correction supported by the evidence.
- Explain the verification, recovery plan, and client update.
How would you investigate a campaign with no impressions?
Example answer: I would confirm the flight is live, then check whether matching ad requests exist. If requests are present, I would inspect line-item eligibility and creative status. If the line item is eligible but not selected, I would examine competition and delivery settings. If selected ads do not render, I would trace the creative and impression event.
Avoid promising that raising priority will solve the problem. Priority cannot restore missing requests or make an incompatible creative eligible.
What would you do when two impression reports disagree?
Example answer: I would align timezone, dates, placement, event definition, and filtering before calculating a discrepancy with a named denominator. I would then segment the gap and investigate tracking only where the evidence supports it.
Mention reporting delay and whether counts are finalized. Do not give a universal acceptable percentage without knowing the platform and contract.
How would you troubleshoot a failing video ad?
Example answer: I would obtain the affected player version, device, error code, and a failed request. I would trace redirects, then media loading and playback. A redirect timeout, wrapper limit, and media failure call for different checks.
A working test-player result is useful, but I would also reproduce the issue in the affected integration before closing it.
Why might CPM rise while revenue falls?
Example answer: CPM describes monetized impressions. A floor increase can remove enough winning opportunities that higher CPM does not offset lower fill. I would compare normalized revenue and demand coverage in comparable segments.
State the arithmetic and the experimental limitation.
How would you communicate an unresolved incident?
Give the observed impact, confirmed findings, current hypothesis, next check, and next update time. Separate facts from assumptions. Do not tell a client the issue is fixed because a configuration change was saved.
For practice, run a Flight Lab incident and explain your investigation aloud in two minutes. Then inspect whether your conclusion accounts for all the evidence, including facts that challenge your first hypothesis.
Test your reasoning: Answer without overclaiming
Original simulated exercise. This is not a real client case.
An interviewer asks why a campaign has stopped.
Evidence available
- You know it is active.
- You have not inspected a request.
- The client says targeting was unchanged.
What is a useful opening answer?
Read the answer and next check
I would establish when and where delivery stopped, then inspect a representative request to locate the failed stage. Active status and unchanged targeting do not establish eligibility.
Next check: Name the next evidence you would collect and what each possible result would mean.