Original practice question. The scoring rubric is our teaching guide. How we research and score answers
TL;DR: Pick a story where you stopped something based on evidence, ideally against criteria set in advance. Explain the signals, how you checked you were not stopping too early, how you communicated with the team and stakeholders, what happened to the people and learnings, and what you did with the freed capacity. The best stories show the decision came earlier than it would have without your criteria.
How to approach it
STAR-L with emphasis on decision quality: criteria, evidence, timing, communication and what happened next.
A strong answer
What to show. Decisions based on evidence; awareness of sunk-cost bias, the tendency to keep investing because of what has already been spent (Arkes and Blumer, 1985); respect for the team's work; clear communication; and redirection of effort. AI work makes this question common in practice: Gartner forecast in June 2025 that over 40% of agentic AI projects will be cancelled by the end of 2027, a forecast rather than an outcome.
An illustration of the structure (adapt to your own experience):
Situation. A team was building an AI feature to auto-generate product descriptions for a marketplace's long-tail sellers.
Task. The PM owned the bet and had set kill criteria at kickoff: 85% of generated descriptions accepted without major edits on a 200-listing pilot, and a measurable conversion lift within eight weeks.
Action. At week six, acceptance plateaued around 60%, mostly because sellers' product data was too sparse for good descriptions. Before recommending a stop, the PM ruled out the cheapest explanation, a broken test: the team re-read 30 rejected descriptions with sellers to confirm the rejections were about content, not a grading or formatting bug. Anthropic's eval guidance warns that a near-zero pass rate across many trials "is most often a signal of a broken task, not an incapable agent" (January 2026), and a stalled one deserves the same check. Then the PM tested whether better prompts or a different model would close the gap with a quick comparison; improvements were small. The PM wrote a one-page summary: criteria, evidence, what was tried, and a recommendation to stop. They met the team first, recognised the work and what was learned, then presented to leadership.
Result. The project stopped two weeks before its planned end. The finding (sparse seller data was the bottleneck) led to a smaller project improving listing data capture, which later made AI descriptions viable.
Learning. "Writing kill criteria at the start made the decision about evidence, not about people's effort. I now set them for every exploratory bet and review them on a fixed date."
Communication principles. Team first, then stakeholders; acknowledge the work; share what was learned; be clear about what happens to people's roles next.
What interviewers probe next
"How did you know you weren't stopping too early?" I ruled out a broken evaluation first, tested the most likely fixes quickly and checked whether the gap was closing; it was not. I also asked whether the decision was reversible. Stopping a pilot is a two-way door, in Amazon's phrase: the prompts, eval set and findings stay on the shelf if better seller data arrives.
"Was eight weeks long enough to see a conversion lift?" Probably not on long-tail listings with little traffic, which is why the acceptance criterion carried the decision and the conversion criterion was treated as a secondary signal. Say which criterion actually decided it.
"How did the team react?" Mixed at first; the clear criteria and the follow-up project helped.
How it's scored
| Signal | Weak | Hire | Strong Hire |
|---|---|---|---|
| Decision basis | Gut feel or pressure | Evidence | Pre-set criteria, evidence, quick checks against stopping too early |
| Communication | Announced abruptly | Explained to stakeholders | Team first, recognition, written summary, clear next steps |
| Learning | None | Personal | Kill criteria adopted as practice, learnings reused in a follow-on project |
Common mistakes
Stopping only when forced by leadership.
No criteria, so the decision looks arbitrary.
Ignoring the team's morale and next steps.
See kill criteria.