Interviewing
Handling behavioral deep dives
Interruptions are interest. Prepare alternatives, one metric, one miss inside the success, and who else was involved.
Strong interviewers interrupt: "What alternatives did you consider?" "Why not X?" "What was the business impact?" "What would you do differently?" Treat interruptions as interest, not failure. If you only memorized a polished paragraph, follow-ups will break you.
Concept
A story bank entry is not a script. It is facts you can retrieve: dates, alternatives, system names, a metric, a miss. STAR is the spoken order, not a wall you hide behind.
Walking through prep
For each story, have ready:
- Two alternatives you rejected and why
- One metric and one qualitative outcome
- One mistake inside the success story (adds credibility)
- Who else was involved and how you coordinated
If you do not remember a number, say so and give a bounded estimate: "I don't remember the exact figure; order of magnitude was cutting the job from ~45 minutes to under 10." Invented precision is worse than honest approximation.
Example
Interviewer: "You said you added a cache. How did you avoid stampedes?"
Good: "We jittered TTLs between 300 and 360 seconds and used singleflight on the miss path so only one process rebuilt the key. I verified with a load test that origin QPS stayed flat when we flushed Redis in staging."
Weak: "Yeah, caching is a best practice so it was fine."
Practice methods that work: bullets first, two-minute recordings, mocks that dig, short and long versions, and a two-minute career arc for "tell me about yourself." End that arc on why this company or problem space so the interviewer can steer into relevant stories.
Amazon's own prep talks about situation/behavior/impact and preparing five or six experiences. A dedicated loop wants two rich stories with room to dig. A mixed round wants one tight three-minute STAR. Ask how much time they want if the agenda is mixed.
Common mistakes
- Memorized paragraph, empty attic. The first "why not X?" ends the story.
- Invented precision. "p99 dropped 87.3%" when you mean "about 800 ms to 120 ms."
- Ignoring the interruption. Answer the probe; do not finish the monologue first.
Check yourself
- What four backups should sit under every story?
Two rejected alternatives, one metric plus one qualitative outcome, one miss inside the success, and who else was involved.
- You cannot remember the exact timeout drop. What do you say?
That you do not remember the exact figure, then an order of magnitude.
Key takeaways
- Interruptions are interest; answer the probe.
- Deep stories need alternatives, numbers, a miss and collaborators.
- Honest approximation beats fake precision.
Further learning
Amazon's short video on Leadership Principles interviews:
A walkthrough of Amazon-style behavioral questions:
Back to the map: behavioral interviews.
Related
Behavioral interviews
Story banks, STAR(L), and signaling the level you want — ownership, conflict, ambiguity and impact with numbers.
Building a story bank
Eight to ten deep stories, each mapped to two or three themes. Rehearse out loud. Update after launches.
STAR answers
Situation, Task, Action, Result, plus Learning. Spend half the time on what you did.