Как делать аналитический post-mortem
Карьерник — квиз-тренажёр в Telegram с 1500+ вопросами для собесов аналитика. SQL, Python, A/B, метрики. Бесплатно.
Зачем это знать
Post-mortem incident — well-known. Но аналитический post-mortem — разбор feature launches, A/B results, major analyses — often skipped.
Ship → move on → repeat mistakes. С post-mortem — systematic learning.
Когда
Feature launch
Ship feature. 1-3 months later — assess.
A/B completion
Regardless outcome — analyze learning.
Campaign end
Marketing campaign done — ROI analysis + learning.
Major analysis
After big deep-dive — what worked methodology-wise.
Bad decision
Launched que flopped. Analyze.
Цель
- What expected vs happened
- What worked / didn't
- What learn для future
- Improve process
Not blame. Learning.
Шаблон
1. Summary
What shipped / analyzed. Dates. Outcome short.
2. Context
Why did this. Business goal. Hypothesis.
3. Results
Actual metrics. Compare к expected.
4. What went well
Positives. Reinforce practices.
5. What went poorly
Problems. Root causes.
6. Insights
Unexpected learnings.
7. Actions
Process improvements. Next time differently.
Пример: feature launch
Feature
Redesigned checkout в e-commerce app.
Expected
Primary: CR +5%. Timeline: 2 weeks build + 4 weeks A/B.
Actual
Primary: CR +2% (less than expected). Secondary: time-to-purchase -20% (unexpectedly). Mobile: +8%, Web: -3%.
Analysis
- CR moderate positive overall
- Mobile wins (primary target)
- Web slight loss (needs investigation)
- Time dropped — good
- Expected 5% but 2% — novelty settled
Insights
- Mobile checkout improvements more impactful
- Web users different flow preferences
- Novelty effect real (initial 7% lift)
Actions
- Ship mobile. Keep web old для now.
- Investigate web discrepancy
- Standardize post-launch analysis template
Для A/B tests
Even не significant
Learning valuable:
- Was hypothesis reasonable?
- Did we underestimate effect?
- Methodology problem?
- Different user behavior than predicted?
Significant
- Replicable?
- Side effects?
- Long-term holdout says?
Document
Post-experiment report — permanent record.
Для campaign / марatingвaye
Ad campaign
Spend, reach, conversions. Compared к forecast. ROI.
Email campaign
Open, click, convert rates. Benchmarks.
Referral program
K-factor achieved? Quality referrals?
Structure
Facts
- What happened
- Dates
- Numbers
- Timeline
Analysis
- Why success / failure
- Contributing factors
- Counterfactuals («what if не»)
Learning
- General principles
- Specific к product / domain
- Applicable future work
Actions
- Changes processes
- New documentation
- Follow-up analyses
Blameless
Same как incident post-mortem:
- Systems / processes — blame
- People — support
- Learning — goal
Don't point fingers.
Sharing
Team meeting
Present post-mortem. Discuss.
Written
Doc в wiki / Notion.
Broader audience
Major learnings — share company-wide.
Examples industry
Amazon
Extensive post-mortems. 6-page narrative style.
Netflix
Chaos engineering — learn от failures.
Big tech
Post-mortems culture.
Russian startups
Varies. Most formal companies do.
Challenges
Takes time
Busy team — skip post-mortem.
Emotions
«Failed project» hard discuss objectively.
Actions не implemented
Learning shallow.
Not shared
Knowledge silos.
Tools
Templates
Company templates standardize.
Meetings
30-60 min dedicated.
Wiki
Searchable archive.
Tagging
Easy find past post-mortems.
Specific analyst post-mortems
Metric definitions
«We defined retention — now find problem. Fix definition. Document correct».
Dashboard issues
«Dashboard wrong for week. Root cause. Prevent».
A/B mistakes
«Experiment bad setup. How avoid next time».
Analysis errors
«Analysis wrong conclusion. Methodology flaw».
Learning from success
Often skipped. «It worked, why analyze?».
Learn:
- Why worked?
- Replicable?
- Scaling вопрос?
Success post-mortem not less valuable.
Mental models
Pre-mortem
Before shipping: «Imagine this fails. Why?»
Proactive risk identification.
5 Whys
Recursive причины.
Ishikawa (fishbone)
Categories причин.
Frequency
Regular rhythm
End-of-quarter post-mortem review.
Per project
Each major project dedicated post-mortem.
Per week (small teams)
Quick retros (Agile teams).
Mistakes post-mortem
Blame
People defensive. Culture suffers.
Vague
«Communication could improve» — non-actionable.
No follow-up
Actions not tracked. Repeat issues.
Only big failures
Small wins / failures also learning material.
Action tracking
Assign owner
- Action item
- Owner
- Deadline
- Status
Review
Next post-mortem — did we do previous actions?
Escalate
If actions не happen — address at management.
Culture
Leadership support
Exec models: «I want learn failures».
Psychological safety
People open share без fear.
Continuous
Process ongoing. Not one-off.
Visible
Learnings shared broadly.
Для собеса
«Post-mortem experience?»
Specific:
- What analyzed
- Findings
- Actions
- Outcomes
Shows maturity.
«Failed project?»
Post-mortem mindset: what learn. Don't blame.
Personal application
Your project failures
Own post-mortems даже если не team.
Career decisions
«Chose wrong company — why? Learn».
Learning projects
«Didn't finish course — why?»
Self-improvement.
Small vs big
Small
15-min team discussion. Informal.
Medium
1-hour meeting + doc.
Big
Multi-week formal analysis. Executive share.
Match scale к impact.
Privacy
Sensitive issues
Some cases confidential. Limited distribution.
Customer impact
Legal implications. Legal reviews.
Tools для post-mortem
- Notion / Confluence templates
- Google Docs collaborative
- Miro / FigJam (visual)
- Jira (action tracking)
На собесе
«Last failed project?»
Story:
- Situation
- What went wrong
- Learnings
- Actions taken
Negative story framed positively.
«Process improvement?»
Post-mortem-driven changes.
Связанные темы
- Как писать post-mortem
- Как проводить root cause analysis
- Как провести performance review
- Почему A/B проваливаются
FAQ
Always write?
Major — yes. Small — ok skip.
Share с exec?
Big learnings — yes. Small — team.
Who owns?
Usually project lead. Analyst often contributes.
Тренируйте — откройте тренажёр с 1500+ вопросами для собесов.