LLenny's Podcast
← All episodes
Albert Cheng (Duolingo, Grammarly, Chess.com)05 October 2025

How to find hidden growth opportunities in your product

8Frameworks
15Insights

Frameworks in this episode

Leadership4 steps

Connect-Users-to-Value Journey Staffing

Growth's job is to connect users to your product's value — so staff teams around each stage of the user journey.

Read →
Strategy3 steps

Existing-User Retention & Resurrection Priority

As a consumer subscription matures, the biggest lever isn't new users — it's current-user retention and resurrecting the dormant.

Read →
Strategy4 steps

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.

Read →
Marketing3 steps

Full-Value Free Sampling (Reverse Free Trial)

Make your free product a live, rationed taste of everything paid can do — not a walled-off basic tier.

Read →
Innovation3 steps

Gamification's Three Pillars: Core Loop, Metagame, Profile

Durable habit-forming products stand on three legs: a tight core loop, a long-horizon metagame, and an accumulating profile.

Read →
Marketing3 steps

Lean-Into-Organic-Sharing Virality

Don't manufacture virality — instrument where users already share, then make those exact moments 5-10x more delightful.

Read →
Leadership4 steps

The 1,000-Experiments 'What Would Need To Be True' Goal

Set an audacious experiment-volume goal — not to hit it, but to force the conversations that unlock a whole experimentation culture.

Read →
Innovation3 steps

The Cold-Pattern Positivity Flip

Find the moments where users feel bad, and flip the product to reinforce progress instead of failure.

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Hot Take· 2

Hot Take41:30

Brand vs Growth Is Wrong. It's Rocket Fuel.

Albert admits he used to see viral brand marketing (Duo the Owl, TikTok, mascots) as opposed to data-driven growth experimentation. Watching Duolingo, he changed his mind: the personality built through product and push notifications feeds the marketing team's memes, and on some days those channels drove 20-30% of new users. Chess.com reinforced it, with pandemic and Queen's Gambit waves quadrupling registrations overnight.

  • Brand marketing and growth experimentation combine rather than compete
  • Duo the Owl's personality, built in-product, fueled TikTok/YouTube marketing
  • Attribution showed those channels driving 20-30% of new users on some days
  • Steady experimentation plus occasional big cultural waves compound
  • Chess.com saw registrations quadruple overnight during big waves

“I used to think it was versus, but now I realize that they combine really well. It could be rocket fuel for for your growth.”

Albert Chen · 42:00
#brand#marketing#growth#virality
Hot Take1:08:00

Hire for Agency, Not Experience (Experience Can Be a Crutch)

The default hiring path fills a JD by poaching from similar companies. But Albert's highest performers were high-agency people with clock speed and energy who didn't necessarily have deep domain experience. In a world shifting fast with AI, learned habits often need to be intentionally discarded, so speed of learning beats experience, and much of that signal shows up outside the formal interview.

  • Top performers had high agency, clock speed, and energy over deep experience
  • Experience can be a crutch, especially as AI shifts the ground fast
  • Learned habits often need to be intentionally discarded; keep a beginner's mind
  • Bet on the fastest speed of learning
  • Agency signals appear outside the interview: questions asked, references, energy, whether they went deep on your product

“sometimes that experience could be a crutch in certain ways.”

Albert Chen · 1:08:30

“a lot of your like learn habits actually need to be intentionally discarded.”

Albert Chen · 1:08:30
#hiring#teams#agency#leadership

Explainer· 4

Explainer28:30

Why Retention Is Gold (And the D1 Benchmark to Aim For)

For consumer subscription products, retention is what makes the model work; without it, all the pressure falls on getting users to pay on day one, which is brutal. Albert gives a rough benchmark: a D1 retention around 30-40% is solid for a consumer app. Below that, he questions whether you can mathematically acquire enough users to build a large daily active base.

  • Weak retention forces you into aggressive day-one monetization
  • D1 retention around 30-40% is a solid mark for a consumer app
  • Much lower D1 raises doubts about user intent and DAU-base math
  • Retention is the margin that lets scrappy consumer products survive

“user retention is gold for consumer subscription companies.”

Albert Chen · 28:30

“I think when you have your D1 retention somewhere around like the 30 or 40% mark, like that's quite solid, I think, for for a…”

Albert Chen · 29:30
#retention#consumer-subscription#metrics#benchmarks
Explainer34:30

How Duolingo, Grammarly, and Chess.com Actually Differ

Albert contrasts the three products he grew. Duolingo runs on a rigid, template-driven 'green machine' where the experience changes multiple times a day per user. Grammarly evolved into a product-led-sales/B2B motion where the core product team (not growth) most drives repeat usage. Chess.com is defined by fanatical, chess-obsessed staff who dogfood the product constantly.

  • Duolingo: highly structured process, product changes multiple times/day per user
  • Grammarly: product-led sales, core product team drives current-user retention
  • Chess.com: staff are fanatical chess players who constantly dogfood
  • There's no single right way; all three succeed very differently
  • At Grammarly, Albert realized growth was 'getting in the way' of the core product team and handed the metric over

“The product experience of Dualingo actually like changes multiple times per day for each user which is pretty shocking.”

Albert Chen · 35:30
#growth#company-culture#duolingo#grammarly
Explainer47:00

Why Chess Engines Crush LLMs (and Grandmasters)

Albert explains that chess engines like Stockfish are dramatically stronger than any human, calculating tens of millions of positions per second. On the ELO scale, a top grandmaster like Magnus Carlsen is ~2,800 while engines sit around 3,600, so strong an engine could give up a rook and still compete. LLMs, by contrast, are bad at chess because they hallucinate moves and can't go deep on a specific line.

  • Chess engines are dramatically better than the top grandmasters in the world
  • Magnus Carlsen ~2,800 ELO vs Stockfish-class engines ~3,600
  • An engine could play down a rook and still be competitive with the best
  • LLMs are bad at chess: they hallucinate moves and can't reason deeply on lines
  • Chess and AI have been intertwined since Deep Blue beat Kasparov in 1997

“A top grandmaster like Magnus Carlson is like a 2,800 and then Stockfish and similar engines are like 3600.”

Albert Chen · 47:30

“interestingly, LM themselves are quite bad at playing chess. Like, they hallucinate moves.”

Albert Chen · 49:30
#ai#chess#llms#engines
Explainer1:05:30

Why New Chess Players Are So Hard to Retain

Over 75% of chess.com's new users self-identify as complete beginners, and beginners have a rough first experience: fewer than a third win their first live game. Since losing a game retains 10% worse than winning, and you can't simplify chess's rules, the team crafts gentler onboarding, like a delightful learn-to-play flow and hiding your rating for the first five games.

  • Over 75% of new users classify themselves as new or beginner
  • Fewer than a third of those users win their first game
  • Losing a game retains 10% worse than winning
  • You can't simplify chess's rules like a casual mobile game
  • Fixes: a gentler learn-to-play flow and hiding rating for the first five games

“over 75% of our new users, they classify themselves as like, I'm completely new to chess or I'm a beginner.”

Albert Chen · 1:06:00

“when you lose a game, user retention is 10% worse than when you win a game.”

Albert Chen · 1:06:00
#onboarding#retention#chess#product

Story· 6

Story11:00

The Chess.com Positivity Flip That Grew Reviews 25%

Chess.com found that 80% of players review a game after a win, not a loss, which contradicted the feature's original design assumption. So they flipped the post-loss experience to surface brilliant moves and encouraging coach messages instead of blunders. That single change grew game reviews 25%, subscriptions 20%, and lifted retention significantly.

  • 80% of people who review their games do so after a win, not a loss
  • The team originally assumed players wanted to study their mistakes
  • After a loss, they now show best moves and encouraging coach copy
  • Result: +25% game reviews, +20% subscriptions, large retention lift
  • The insight was then spread to adjacent teams (puzzles, etc.) to compound

“what he observes is that 80% of people that review their games actually do so after a win”

Albert Chen · 11:30

“That change alone was pretty dramatic for us. It grew game reviews by 25%, subscriptions by 20%, user retention by a lot as well.”

Albert Chen · 12:00
#growth#retention#product#chess
Story23:00

The Grammarly Move That Nearly Doubled Upgrades

Most free Grammarly users experienced it as a spelling-and-grammar fixer because those were the only free suggestions shown. The team interspersed a limited taste of paid suggestions (tone, clarity, rewrites) directly into free users' writing. Despite fears of giving too much away, upgrade rates nearly doubled because users suddenly saw Grammarly as far more powerful.

  • Free users only saw correctness suggestions, so they undervalued the product
  • They sampled paid suggestions intermingled into free users' writing in real time
  • Concern that giving features away would kill conversion proved false
  • Upgrade rates nearly doubled from the change
  • Lesson: make your free product a reflection of everything the product can do

“all of a sudden people were seeing Gramly as a much more powerful tool than they were before and our upgrade rates like nearly doubled…”

Albert Chen · 24:00

“It's basically like a reverse free trial but in real time like while you're writing as opposed to a time based one.”

Albert Chen · 24:30
#monetization#freemium#grammarly#conversion
Story57:00

The 1,000-Experiments Goal He Openly Made Up

Chess.com went from practically zero experiments before 2023 to ~50, then ~250, with a target of 1,000 next year. Albert cheerfully admits he invented the number. The point isn't hitting it; a stretch goal forces the conversation about what would have to be true to get there, which surfaces the need to enable experimentation across lifecycle marketing, app store assets, and no-code screens.

  • Chess.com barely experimented before 2023; ~50 last year, ~250 this year, 1,000 target
  • Albert openly admits he made the 1,000 number up
  • A goal's value is forcing the 'what would need to be true' conversation
  • Hitting the number matters less than the capabilities it unlocks
  • Enables experiments in lifecycle marketing, app store assets, and no-code screens

“Did I make it up? Yes, absolutely. I made it up.”

Albert Chen · 58:00

“The whole point of setting a goal is that you can have conversations about what would need to be true to actually hit that goal.”

Albert Chen · 58:00
#experimentation#goals#growth#chess
Story1:01:30

The Experiment Tool That Had Retention Backwards

Albert warns that the experimentation system matters as much as any single experiment, and instrumentation is where it breaks. At one company, an in-house tool had user retention configured backwards for three months, so every positive result was actually negative. It's a cautionary tale for why you must instrument your product thoroughly before trusting results.

  • The system and instrumentation matter as much as any single experiment
  • One in-house tool had retention configured backwards for ~3 months
  • All 'positive' results were actually negative during that window
  • Bad instrumentation produces wonky, misleading experiment results

“we realized that user retention was actually configured backwards. So all positive results were negative results.”

Albert Chen · 1:02:00
#experimentation#instrumentation#data-quality#growth
Story1:02:30

How Duolingo Found Its Viral Moments With Screenshot Tracking

Rather than forcing virality, Duolingo added temporary screenshot tracking to discover where users were already organically sharing. Hotspots included streak milestones and funny challenges (not, say, advancing in the leaderboard top three). They then staffed those exact moments with illustrators and animators to make them far more delightful, driving 5-10x more sharing.

  • Virality is hard to manufacture; find where users already share
  • Temporary screenshot tracking revealed organic sharing hotspots
  • Streak milestones and funny challenges were highly shared; leaderboard climbs were not
  • They staffed those moments with illustrators/animators to amplify them
  • Leaning into existing behavior drove roughly 5-10x more sharing

“we invested actually in some time to um essentially add screenshot tracking for like a brief period of time in the app”

Albert Chen · 1:03:00

“grab the moments where users are already organically screenshotting and make those much much much better”

Albert Chen · 1:04:00
#virality#duolingo#growth#sharing
Story1:13:30

The Chariot Failure: A Solution Searching for a Problem

At the commuter-shuttle startup Chariot, Albert's team built 'Chariot Direct' to add Uber/Lyft-style dynamic routes onto fixed shuttle lines. It failed, teaching three lessons: it was a solution searching for a problem rather than a real user need; in a marketplace they over-focused on the rider app and neglected drivers and operations; and they ran PR before validating demand, which created sunk-cost pressure to see a doomed idea through.

  • 'Chariot Direct' added dynamic Uber/Lyft-style routing to fixed shuttle lines and failed
  • Lesson 1: don't chase 'wouldn't it be nice' ideas; start from a real user and problem
  • Lesson 2: in marketplaces, there's more than one end user; they neglected drivers and ops
  • Lesson 3: heavy PR before validation created costly sunk-cost momentum
  • The opposite failure mode of validating everything before telling anyone

“this was kind of a solution searching for a problem.”

Albert Chen · 1:15:00

“doing it before you have validation that customers definitely want the thing is quite risky.”

Albert Chen · 1:16:00
#failure#product#marketplace#validation

Tool· 1

Tool16:30

The Text-to-SQL Slack Bot That Answers Data Questions

Chess.com is building a Slack bot that takes ad-hoc data questions (e.g. subscribers in South Africa) and actually runs the analysis, not just generating a query. It replaces the slow loop of a data analyst prioritizing and running one-off requests. A side effect: because people are less embarrassed asking a bot, question volume explodes and the company becomes more data-informed.

  • A Slack bot performs the actual analysis via text-to-SQL, not just query generation
  • Removes the bottleneck of analysts triaging one-off ad-hoc questions
  • Lowers social friction so people ask questions they'd otherwise avoid
  • Result is a large increase in questions asked and a more data-informed company

“if you have a question that you feel like, you know, you might be a bit embarrassed to ask or you don't want to bother…”

Albert Chen · 17:30
#ai#analytics#data#tools

Takeaway· 2

Takeaway30:30

Mature Companies Grow by Resurrecting Dormant Users

As a company matures, the biggest retention lever shifts from new users to existing habitual users, whose retention compounds into a daily habit. Mature products also stack up hundreds of millions of dormant and sporadic users, so reactivated and resurrected users become as large a component as brand-new ones. It's worth designing a deliberate experience for resurrected users, as Duolingo does with social notifications and placement re-tests.

  • Existing-user retention compounds and builds the daily habit that matters most
  • For mature products, reactivated/resurrected users rival new users in volume
  • Dormant and sporadic users accumulate into hundreds of millions
  • Design a deliberate 'resurrected user' experience for strong ROI
  • Duolingo uses social push notifications and placement re-tests to bring users back

“it's that retention rate that really compounds and build that builds that daily habit.”

Albert Chen · 31:00

“it's actually like pretty interesting that the components of your active user base are actually not heavily weighed in the new user set after you…”

Albert Chen · 32:30
#retention#reactivation#growth#duolingo
Takeaway1:10:30

Everyone Has a Company Stage Where They Shine

Albert believes everyone has a company stage that fits them best. Having done big tech, tiny startups, and the middle, he landed in a medium-sized 'goldilocks zone': companies of ~500-1,000 people, typically 10-20 years old, durable and ideally profitable but still at key inflection points. It gives him both scale and the ability to execute on a daily/weekly rather than monthly/quarterly cadence.

  • Everyone has a company stage where they perform best
  • Big tech offers scale but moves slowly; tiny startups move fast but are grueling and unknown
  • His goldilocks zone: ~500-1,000 people, 10-20 years old, durable, ideally profitable
  • These companies are at inflection points, not stagnant
  • The sweet spot balances scale with a daily/weekly execution pace

“I I definitely believe that everyone has a company stage that they shine best at.”

Albert Chen · 1:11:00

“these companies that we've talked about in the podcast are about 500 to a,000 people.”

Albert Chen · 1:12:30
#career#company-stage#startups#growth