Situation: On a scoping call the PM requested a new personalized recommendation feature to be shipped in two weeks. Given dataset size and labeling needs, that timeline seemed implausible.
Task: My goal was to clarify scope, surface constraints, and align on an achievable plan.
Action — clarifying questions I asked during the meeting:
- What is the primary objective of this feature? (engagement, revenue, retention — what metric will define success?)
- Who are the target users and which use-cases matter most initially?
- Do we need fully personalized recommendations or is a simple rule-based / popularity-based MVP acceptable?
- What data is available today (user events, item metadata)? How complete and recent is it?
- Is labeled data required, or can we start with implicit signals (clicks, views)?
- What accuracy or KPI thresholds are expected for launch?
- What infra and team capacity do we have for data processing, model training, and A/B testing in two weeks?
- Are there privacy/compliance constraints or external dependencies (analytics, backend APIs)?
- If two-week deadline is immovable, which parts of scope can be reduced (smaller cohort, shorter horizon, simpler model)?
Result — end-of-call summary I would read aloud to ensure alignment:
- Agreed goal: improve homepage click-through rate by personalized recommendations for logged-in users (PM metric).
- Scope for launch: deliver an MVP that uses popularity + simple collaborative filtering on last 30 days of implicit signals for a 10% user cohort.
- Data: Product and event logs available; labeling not required initially. Data engineering to validate data by Day 3 (owner: Data Eng).
- Deliverables & owners:
- Day 3: Data availability validated and sample dataset provided (Data Eng)
- Day 7: MVP model prototype and offline evaluation (me, AI Engineer)
- Day 10: Integration plan and API contract (Backend)
- Day 14: Internal rollout to 10% cohort + measurement dashboard (PM & Analytics)
- Risks: model quality may be lower without labels; monitor CTR and roll back if negative impact.
- Follow-ups: I’ll send the commit summary and JIRA tasks within 2 hours. If higher-quality personalization is required, we’ll plan a 6–8 week roadmap for labeled data and production-grade model.
This ensures we meet the business need with a pragmatic MVP and a clear path for iteration.
