Explore & Exploit: The Growth Playbook of Duolingo, Grammarly & Chess.com
Albert Cheng
Head of Growth & Monetization Chess.com · ex-Duolingo · ex-Grammarly
CONSUMER GROWTH
Framework
Explore & Exploit: Find the Mountain, Then Climb It
"When you're in exploratory mode, think of it as finding the right mountain to climb. When you're in exploitation mode, it's focusing your resources on climbing that mountain effectively."
Too much explore = scattershot teams, no through-line
Too much exploit = local maximum, stagnation
The magic: oscillate between the two deliberately
Losing experiments are as valuable as wins — surface both
The Wins
Three Breakthrough Growth Experiments (Real Results)
Grammarly — Sample Paid Features Free
Free users only saw spelling/grammar fixes. The team interspersed paid suggestions into the free experience so users could feel the product's real power. Result: dramatically higher conversion and a fundamentally different perception of Grammarly's value.
Chess.com — Flip Loss Reviews Positive
80% of game reviews happened after wins — not losses as designed. After a loss, the product now surfaces your best moves and encouragement instead of blunders. Result: game reviews +25%, subscriptions +20%, retention up significantly.
"We flip it on its head — show you your brilliant moves, your best moves, and have the coach say something encouraging. That change alone was pretty dramatic."
+25%
game reviews after positive framing
+20%
subscriptions from same change
1,000
experiments/year target
30–50%
typical experiment win rate
Duolingo — Brand + Experiments = Rocket Fuel
Duo the owl's TikTok personality drove 20–30% of new user registrations on peak days. Albert's lesson: product experimentation is the steady engine; brand moments are the waves. Don't ignore the waves — they can 4× registrations overnight.
Playbook
How to Run a World-Class Growth Operation
Set a cadence goal: Chess.com runs ~250 experiments/year; Duolingo's product reviews are 10–15 min each. Consistent velocity beats periodic big bets.
Broadcast insights company-wide: When a PM finds a human psychology insight (e.g. people prefer wins), adjacent PMs (puzzles, lessons) should immediately audit their own cold patterns.
Use AI for data velocity: Text-to-SQL Slack bots answer ad hoc data questions instantly — freeing analysts for higher-order work.
Build experiment explorer tools: Scan patterns across all running experiments. Declining statistical significance = time to shift from exploit back to explore.
Retention is gold: At consumer subscription products, if you don't retain users, all monetization pressure falls on day one.
Team Building Counterintuition
Albert's highest performers weren't domain experts. They had high agency, clock speed, and energy. In a world where AI is shifting the ground, learned habits can become a crutch.
"Sometimes experience could be a crutch. A lot of your learned habits actually need to be intentionally discarded."
Music → Growth Transfer
Both rely on consistent repetition, tight feedback loops, and resilience to mistakes. Both need structural underpinning (theory / growth model) plus in-the-moment creativity.
Tactic
The Insight Cascade: 1 Win → 10× Impact
The original PM doesn't need to solve it for everyone — just clearly articulate the hypothesis and finding
As growth leader, encourage swarming: more minds, same insight, higher success rate
Human psychology insights are the most portable — they cross product boundaries
When experiment wins cluster, you're in exploitation mode — ride it
Albert's goal at every company
Run 1,000 experiments per year. Even at Chess.com's current 250/year, systematic sharing multiplies real impact far beyond the raw experiment count.
Contrarian
Albert's Counterintuitive Takes on Consumer Growth
✗Show users their mistakes so they can fix themINSTEAD →✓ 80% of people review games after wins. Flip the experience positive — show best moves after losses. Human psychology beats product logic every time.
✗Hire domain experts for growth rolesINSTEAD →✓ High agency, clock speed, and energy matter more than deep expertise. In a fast-shifting AI world, experience can actually be a crutch — learned habits need to be discarded.
✗Brand and product growth are separate machinesINSTEAD →✓ Duo the owl's TikTok personality drove 20–30% of new users on peak days. Brand moments are the waves; experiments are the engine. You need both — and they amplify each other.
✗Growth is just metrics hackingINSTEAD →✓ "Growth as a job is to connect users to the value of your product." Retention is the real metric. If you don't retain users, all monetization pressure collapses onto day one.