Explore & Exploit at the Insight Level
Oscillate between finding the right mountain and climbing it — but do it insight by insight, not just at strategy level.
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 4
- Confidence
- 95%
A growth operating rhythm: exploration mode finds the right opportunity ('the right mountain to climb'), exploitation mode concentrates resources to climb it. Cheng's twist is running this loop at the micro/insight level, not just as a macro portfolio choice. Use it to avoid both scattershot experimentation and local-maxima stagnation.
Origin
Cheng heard the concept from his Grammarly engineering partner Nurmal, who took Reforge classes; he credits Brian Balfour as the likely originator. Cheng adapted it to operate at the individual-insight level with his teams at Chess.com.
Core principles
- 01Too much exploration makes a team feel scattershot with no through-line; too much exploitation leads to saturation and stagnation (local maxima).
- 02A single experiment insight — even a losing one — is the unit to explore and then exploit across the org.
- 03The original experimenter need not solve the whole product; their job is to articulate hypothesis and finding so others can swarm.
- 04Falling statistical significance across experiments is the signal you've exploited too far and should return to divergent brainstorming.
How to run it
- 1
Explore for a breakthrough insight
Run experiments in an area to find something that breaks through the noise — a counterintuitive behavioral or psychological insight, not just a metric bump.
Pro tip Typical experiment win rate is only 30-50%; expect most hypotheses to fail.
- 2
Articulate the hypothesis and finding clearly
The PM who ran the experiment writes up exactly what they hypothesized and what they found, so the insight is portable to other teams.
Pro tip Losing experiments are also super valuable — surface those too.
- 3
Swarm adjacent surfaces (exploit)
As a leader, share the insight broadly and encourage adjacent PMs to apply it to their features — copy tweaks, colors, success ratings — expanding one win 10x across the org.
- 4
Watch for saturation, then re-explore
Use experiment-explorer tooling to spot when results stop being statistically significant; that's the cue to stop exploiting and get divergent again.
Pro tip Invest in explorer tools that surface patterns across many experiments at once.
Watch out Don't keep squeezing a zone once the 'juice' is gone — you'll waste cycles.
In the wild
PM Dylan found 80% of people review games after a WIN, not a loss. Chess.com flipped the post-loss experience to surface brilliant moves and encouragement instead of blunders. Adjacent PMs (e.g. puzzles) then audited their own 'cold' negative patterns and made them positive.
→ Game reviews +25%, subscriptions +20%, and large retention gains — then the insight was expanded across other product surfaces.
Common mistakes
Living permanently in exploit mode
The default MO of growth teams is endless exploitation, which locally maximizes and stagnates. You must deliberately return to exploration.
Treating it only as a macro decision
Applying explore/exploit only at the strategy level misses the compounding value of expanding individual insights across teams.
Is it for you?
Best for
Growth leaders at consumer products with enough scale to run a steady volume of experiments and multiple adjacent product teams.
Not ideal for
Pre-PMF startups with too little traffic to reach statistical significance, or B2B products that don't experiment.
From the transcript
“when you're in exploratory mode, think of it as like finding the right mountain to climb. And then when you're in exploitation mode, it's like…”
“you can take this like experiment win and expand it out 10x across your organization”
“If I'm starting to see like more and more experiments that are not statistically significant, that may be a signal to me to say, okay,…”
From the episode
How to find hidden growth opportunities in your product
Albert Cheng (Duolingo, Grammarly, Chess.com)