"Don't take people away from the core — you leave yourself ripe for disruption and end up distracted."
Mixpanel hit 40% revenue churn after splitting eng across 3 products
Customers didn't stop needing analytics — they churned to competitors who out-invested the core
Expand with profits (or VC), never by moving engineers off the core
Framework
Mixpanel's 3-Phase Product Turnaround
40%
churn before pivot
100+
features in year one
50
NPS (was 16)
Phase 1 — Feature Sprint (2018–2019)
Sorted all churn reasons by ARR lost. Made the top 10 the entire roadmap. Gave every engineer direct access to the customers with those problems. No top-down feature specs — engineers invented the solutions. Shipped 100+ features in year one.
Phase 2 — Design System (2019–2021)
Paused tactical feature work for 3 months so design could rethink system architecture from scratch. Identified fewest possible building blocks. Result: every new feature now inherits consistent UX automatically — multiplying reach with zero extra work.
The Sequencing Principle
You can't mow the lawn while the house is on fire. Stop the bleeding with speed (Phase 1), then invest in quality (Phase 2). Don't skip the sequence — they each have a context where they're right.
Playbook
How Mixpanel Plans & Prioritizes
6-Month Planning Cycle — The "Bets" System
Leadership writes a strategy memo: company direction for the next 12 months
Teams develop bets — not OKRs, not features. Anatomy: problem + hypothesis + plan to win + success metric
CPO and Head of Design join every team's FigJam jam session — actively contribute ideas, don't just review
Conclude with company-wide roadshow for alignment before locking plans
Team Structure: Paired Problems
Teams own long-lived problem pairs in tension. Example: Power & Simplicity is one team — they're forced to make that trade-off themselves, not escalate it.
Other pairs in play: data trust + onboarding; collaboration + price-performance.
Fix the RICE Trap
Problem: C (confidence) + E (effort) in RICE systematically kill your highest-reach ideas — because innovative work is always murky to estimate.
Fix: Drop C and E for a week. Bring engineers + designers. Sincerely try to make the idea work. You'll find a higher-confidence, lower-effort approach. Then add C+E back and RICE normally.
Goal: End up with a mix of innovative bets, incremental improvements, and tech debt — not just what was easiest to scope.
Appetite over Estimates (Shape Up)
Set a time box (appetite) as the input, not an estimate as the output. Pick 6 weeks. Then ask: what would we do differently in 4 weeks? 8 weeks? Find the efficient frontier, then ship something complete — not milestone one of an unfinished idea.
Tactics
Engineer-to-Customer Direct Line
Pipe all CS/sales customer gaps into a Slack channel — raw, ungated, no PM filter
Engineers react with an email emoji to claim a thread, then email that customer directly
Enrich the feed: tag with account ARR, CSM name, tier — engineers have context before reaching out
Expand the signal set: NPS, Twitter mentions, win/loss notes — all pipe into the same feed via data warehouse
Use reverse-ETL (Census) to push from BigQuery → Slack/Notion so it's enriched and real-time
Biggest Analytics Mistake
Client-side SDK tracking drops 20–30% of events (ad blockers), duplicates events across iOS/Android, and is frozen for old app versions. Track from your servers instead — it's just structured logs with a user ID. Instant cross-platform, always current, zero drops.
Contrarian
Product Myths Vijay Debunks
✗Expand products to grow revenueINSTEAD →✓ Adjacent bolt-on products rarely become best-in-class. They contribute 5–10% revenue while starving your core of the engineers who could have made it unbeatable. Cut mild successes early — it's 10x more painful than you think.
✗PMs should own all customer contextINSTEAD →✓ Give every engineer and designer a raw, ungated feed of real customer feedback. Engineers who email customers directly build better products — and it takes cognitive load off PMs being the only bottleneck to customer truth.
✗RICE is the gold standard for prioritizationINSTEAD →✓ RICE systematically kills your best ideas because C and E are murky for high-reach, innovative work. Drop them first, explore the idea for a week, then bring them back. Your most innovative bets will survive.
✗Client-side SDK tracking is the defaultINSTEAD →✓ Server-side tracking is 100% cross-platform, always up-to-date, and drops zero events. It's just structured logs with a user ID — engineers have been doing it for decades. Make it the default; supplement with client-side only when needed.