Case library
Real products, dated outcomes.
Each case is mapped to the stage the product was in, with the outcome’s number, its year and its source. Two grades keep it honest: how solid the number is, and how well the mechanism explains it.
How to read the grades
Case grade for the number: B. Evidence grade for the mechanism link: D.Number is graded by its source: a filing or study (A), the company (B), the press (C) or an anecdote (D). Link grades how well the book mechanism explains the outcome; our own inference is always D.
Stage
Product
Evidence
Technique
46 cases, strongest evidence first
24 Hour Fitness (megastudy) · 2021
RoundThe gym megastudy: most effects didn't last
54 interventions, 61,293 members of 24 Hour Fitness: 45% significantly raised weekly gym visits (by 9-27%) during the four-week programs; only 8% produced measurable change after they ended; impartial forecasters failed to pick the winners (Milkman et al., Nature, Dec 8, 2021).
ReturnStayCase grade for the number: A. Evidence grade for the mechanism link: A.Penn Medicine (STEP UP trial) · 2019
RoundPenn STEP UP: competition beat support and collaboration
During the 24-week intervention: competition +920 steps a day vs control, support +689, collaboration +637. In the 12-week follow-up only competition stayed significantly ahead (+569). Patel et al., JAMA Internal Medicine, Sept 2019.
ReturnCase grade for the number: A. Evidence grade for the mechanism link: A.Airbnb · 2017
RoundAirbnb: reputation beats similarity
In an experiment with 8,906 Airbnb users, high reputation significantly increased trust between dissimilar users, offsetting the pull toward similar others; analyses of about 1M real Airbnb interactions agreed (Abrahao et al., PNAS, Sept 2017).
M3 ×DiscoverStayCase grade for the number: A. Evidence grade for the mechanism link: A.Pokemon GO (Niantic) · 2016
RoundPokemon GO: the extra steps faded
+955 steps a day in the first week after installing, falling over the following weeks and back to pre-install levels by week 6 (Howe et al., BMJ, Dec 13, 2016; about 1,182 US adults aged 18-35).
M5 ×ReturnCase grade for the number: A. Evidence grade for the mechanism link: A.Opower · 2014
RoundOpower: your energy use vs your neighbors'
Average energy reduction of about 2.0% (1.4-3.3% across 12 utilities) in a randomized evaluation of 588,446 households; homes dropped after about two years still saved about 2% (Allcott & Rogers, American Economic Review, Oct 2014; via J-PAL). Oracle bought Opower in 2016 (~$532M).
M3 ×StayCase grade for the number: A. Evidence grade for the mechanism link: A.Hotel field experiment (Goldstein, Cialdini & Griskevicius) · 2008
RoundThe hotel towel study: “guests in this room”
Experiment 1: 44.1% towel reuse with the descriptive norm vs 35.1% with the environmental appeal. Experiment 2: the same-room norm reached 49.3% vs 42.8% for the other norm messages combined, and the standard message 37.2%. Peer-reviewed field experiments.
StartCase grade for the number: A. Evidence grade for the mechanism link: A.Duolingo · 2020
RoundDuolingo: smarter reminders, not more reminders
The bandit algorithm lifted total DAU 0.5% and new-user retention 2% over a strong baseline, without sending more notifications (Yancey & Settles, KDD 2020; company authors, peer-reviewed).
M4 ×ReturnCase grade for the number: A. Evidence grade for the mechanism link: B.Fashion retailer (study) vs the Amazon claim · 2018
RoundProduct recommendations: the 35% claim vs a controlled test
Recommendations raised sales of recommended items 9% while sales of the originally viewed items fell 1.9%; overall product sales rose 11% vs a no-recommendation control, across 2M+ product page views (Kumar & Hosanagar, 2018). The 35% Amazon figure is a disputed estimate.
M4 ×PayCase grade for the number: A. Evidence grade for the mechanism link: B.Google · 2015
RoundGoogle: more ads taught users to ignore ads
Long-running experiments showed users' tendency to click ads shifts with the quality of ads they have seen (ad blindness or sightedness). The results led to a 50% cut in ad load on Google's mobile search and an auction change favoring ad quality (Hohnhold, O'Brien & Tang, KDD 2015).
M4 ×StayPayCase grade for the number: A. Evidence grade for the mechanism link: B.Amazon · 2025
RoundAmazon Prime: easy to join, hard to leave
Sept 25, 2025: Amazon agreed to pay $2.5B ($1B civil penalty plus $1.5B in refunds to about 35M customers), offer clear ways to decline Prime, disclose terms at enrollment and simplify cancellation. Amazon said it has always followed the law.
M4 ×StayPayCase grade for the number: A. Evidence grade for the mechanism link: C.Booking.com · 2019
RoundBooking.com: urgency claims meet regulators
Feb 6, 2019: Booking.com, Expedia, Hotels.com, Agoda, ebookers and trivago committed to the UK CMA to stop false availability or popularity claims and pressure selling by Sept 1, 2019. EU-wide commitments followed (2019-2020). In Sept 2019, Which? found 5 of 10 checked claims misleading.
M4 ×DiscoverPayCase grade for the number: A. Evidence grade for the mechanism link: C.LinkedIn · 2015
RoundLinkedIn: invitations sent in your name
Perkins v. LinkedIn: a $13M settlement announced in Oct 2015 (final approval in Feb 2016) for Add Connections users from Sept 2011 to Oct 2014. The suit alleged members agreed to the first invitation but not the two reminder emails. LinkedIn denied wrongdoing.
M3 ×SpreadCase grade for the number: A. Evidence grade for the mechanism link: C.Duolingo · 2026
RoundDuolingo: the Streak Revival event
15.4M learners revived their streaks, including nearly 8M who had no active streak when the event began (Q2 2026 shareholder letter, Aug 5, 2026). Q2 2026 DAU were 58.7M, up 23% year over year; many causes.
M4 ×ReturnCase grade for the number: A. Evidence grade for the mechanism link: D.Analyst inferenceCorrelationalMatch Group (Tinder, Hinge) · 2026
RoundTinder vs Hinge: variety vs investment
Q2 2026 (8-K, Aug 4, 2026): Match Group payers 13.25M, down 6% YoY; Hinge revenue up 22% YoY with monthly users up 13%; Tinder's YoY decline in daily users narrowed to 4%. Many causes; the mechanism link is our inference.
M4 ×ReturnPayCase grade for the number: A. Evidence grade for the mechanism link: D.Analyst inferenceCorrelationalMercado Libre · 2026
Mercado Libre: seller reputation as currency
Q2 2026 (8-K, Aug 5, 2026): 89M unique active buyers (71M a year earlier, +26%), fintech monthly users 88M (+30%), MELI+ subscribers up 72% YoY (no absolute number), and users active in both commerce and fintech up 37%. Many causes; the mechanism link is our inference.
M4 ×StayPayCase grade for the number: A. Evidence grade for the mechanism link: D.Analyst inferenceCorrelationalPeloton · 2026
RoundPeloton: loyal members, shrinking top of the funnel
Paid Connected Fitness subscriptions fell from 2.800M to 2.553M (-8.8%) in fiscal 2026, while average net monthly churn stayed at 1.7% (1.6% in fiscal 2025), per the 10-K for the year ended June 30, 2026.
M4-M5 ×DiscoverStayCase grade for the number: A. Evidence grade for the mechanism link: D.Analyst inferenceCorrelationalPinterest · 2026
RoundPinterest: Hooked's textbook example, then and now
640M global monthly active users in Q2 2026, an all-time high, up 11% YoY and the 11th straight quarter of double-digit user growth (Pinterest release, Aug 4, 2026). About 50M in the book era. Many causes; the Hook is not claimed to have caused the growth.
M3-M4 ×ReturnCase grade for the number: A. Evidence grade for the mechanism link: D.Analyst inferenceCorrelationalFigma · 2025
Figma Community: remix culture
250,000+ Community resources, including 10,000+ plugins and widgets, and about 450k paid customers as of March 31, 2025 (Figma S-1, 2025). The link from Community to growth is our inference [Analyst; many causes].
M4 ×StaySpreadCase grade for the number: A. Evidence grade for the mechanism link: D.Analyst inferenceCorrelationalSnapchat (Snap Inc.) · 2025
RoundSnapchat: Snapstreaks and a state lawsuit
Utah sued Snap on June 30, 2025; the complaint names Snapstreaks among allegedly gambling-like features aimed at minors. These are allegations, not findings, and Snap disputes them. Snap reported 493M DAU in Q2 2026 (SEC filing).
M4 ×ReturnCase grade for the number: A. Evidence grade for the mechanism link: D.Analyst inferenceRobinhood · 2024
RoundRobinhood: confetti on trades
Confetti removed March 31, 2021. Jan 18, 2024: a $7.5M settlement with Massachusetts securities regulators over gamified features (confetti, digital scratch tickets, free stock, push notifications, popularity lists), barring celebratory imagery tied to trading for Massachusetts accounts. Robinhood neither admitted nor denied those allegations.
M3-M4 ×StartReturnCase grade for the number: A. Evidence grade for the mechanism link: D.Analyst inferenceTikTok · 2024
RoundTikTok Lite: paid to watch, then withdrawn
Aug 2024: the European Commission made binding TikTok's commitment to permanently withdraw TikTok Lite Rewards from the EU and not to launch any programme that circumvents the withdrawal, the first DSA case closed with commitments. No admission.
M4 ×ReturnCase grade for the number: A. Evidence grade for the mechanism link: D.Analyst inferenceDuolingo · 2023
RoundDuolingo: weekly leagues
Leaderboards raised overall learning time 17% and tripled highly engaged learners (1h+ a day, 5 days a week), per Duolingo's Jorge Mazal (Feb 2023). Company-reported, and part of a broader retention push that grew DAU 4.5x over four years.
M4 ×ReturnCase grade for the number: B. Evidence grade for the mechanism link: B.CorrelationalMany causesDuolingo · 2022
RoundDuolingo: a streak with a safety valve
Letting learners equip up to two Streak Freezes raised daily active learners by 0.38% (company A/B result, Jan 2022). Learners who reached a 7-day streak were 3.6x more likely to finish their course, a correlational figure.
M4 ×ReturnStayCase grade for the number: B. Evidence grade for the mechanism link: B.CorrelationalCompany-reportedSuperhuman · 2018
RoundSuperhuman: 22% to 33% without shipping a feature
“Very disappointed” went from 22% (summer 2017) to 33% by segmenting to the right personas, with no product change, and to 58% within about three quarters, per CEO Rahul Vohra (First Round Review, Nov 13, 2018). Company-described method.
M1 ×ValueCase grade for the number: B. Evidence grade for the mechanism link: B.Company-reportedTwitter (now X) · 2017
RoundTwitter: follow about 30 accounts
Users who followed about 30 accounts became long-term active users, per Twitter's Josh Elman (Hacking Growth excerpt, Apr 25, 2017; also in Hooked, Ch8). Correlational; a variant of 5-10 follows on day one circulates secondhand. Twitter was renamed X in 2023.
M2 ×StartCase grade for the number: B. Evidence grade for the mechanism link: B.CorrelationalCompany-reportedSlack · 2015
RoundSlack: the 2,000-message threshold
About 93% of teams past 2,000 messages were still using Slack, per co-founder Stewart Butterfield (First Round Review interview, Feb 27, 2015). No underlying data was published; correlational. It is not a paid-conversion figure.
M2 ×StartCase grade for the number: B. Evidence grade for the mechanism link: B.CorrelationalCompany-reportedInstagram (Meta) · 2021
RoundInstagram: hiding likes, then letting people choose
May 26, 2021: hiding like counts became optional on Instagram and Facebook. Meta said hiding counts helped some people and frustrated others, partly because people use likes to spot what is trending. No quantitative results were published.
M4 ×ReturnCase grade for the number: B. Evidence grade for the mechanism link: C.Company-reportedSpotify · 2025
RoundSpotify Wrapped: identity as a reward
200M+ engaged users in the first 24 hours of Wrapped 2025, 19% more than 2024 (which took 62 hours to reach 200M), and 500M+ shares, up 41% (Spotify, via TechCrunch, Dec 4, 2025). Company-reported.
M4 ×ReturnSpreadCase grade for the number: B. Evidence grade for the mechanism link: D.Analyst inferenceCompany-reportedStarbucks · 2025
RoundStarbucks Rewards: stars, tiers and devaluation risk
34.2M US 90-day active members, and member spend at 58% of US company-operated tender (data as of Sept 28, 2025; 60% a year earlier). The three-tier relaunch on March 10, 2026 drew mostly negative reaction from casual members (press).
M4-M5 ×StayPayCase grade for the number: B. Evidence grade for the mechanism link: D.Analyst inferenceCorrelationalStrava · 2025
RoundStrava: one-tap kudos
14B kudos given (up 20% on 2024) by a community of about 180M, for data from Sept 1, 2024 to Aug 30, 2025 (Strava Year in Sport 2025, Dec 2025). Company-reported. In Dec 2025 the Year in Sport recap moved behind the subscription, and users pushed back.
M4 ×ReturnCase grade for the number: B. Evidence grade for the mechanism link: D.Analyst inferenceCorrelationalNotion · 2024
Notion: templates as investment and community
51M template duplications by nearly 11M people in one year; the gallery grew from about 600 to 5,000+ templates, with 2,000+ creator profiles (Notion blog, 2024). Company-reported; effects on activation or retention were not published.
M4 ×StartStayCase grade for the number: B. Evidence grade for the mechanism link: D.Analyst inferenceCorrelationalFacebook (Meta) · 2016
RoundFacebook: 7 friends in 10 days
“7 friends in 10 days” as Facebook's activation milestone, per growth leader Chamath Palihapitiya's public talk and repeated by Index Ventures (Nov 28, 2016). Correlational; the exact derivation was never published. Reforge frames it as a setup threshold whose aha moment is an interesting feed.
M2-M3 ×StartCase grade for the number: C. Evidence grade for the mechanism link: B.CorrelationalPress-reportedEvernote · 2011
Evernote: paying grew with tenure
Paid share about 0.5% after month 1, 11% by month 33 and 26% at month 42, per Phil Libin (2011), as reported in Hooked (Ch1). A book-era figure; Evernote was later acquired by Bending Spoons (not re-verified).
M3 ×PayCase grade for the number: C. Evidence grade for the mechanism link: C.CorrelationalPress-reportedCanva · 2025
Canva: instant competence, templates first
About 265M monthly active users at the end of 2025 (up from 180M a year earlier) and 31M+ paid users, per analyst firm Sacra; Canva's own 2025 recap reported about $3.5B in revenue. Many causes; the link to templates-first onboarding is our inference.
M3 ×StartCase grade for the number: C. Evidence grade for the mechanism link: D.Analyst inferenceCorrelationalStack Overflow · 2025
RoundStack Overflow: when instant answers beat points
Monthly new questions fell to about the level of the site's 2009 launch by May 2025, after declining since about 2014 and sharply after Nov 2022 (The Pragmatic Engineer, May 15, 2025, citing public data). Prosus acquired the company for about $1.8B (June 2, 2021).
M5 ×StayCase grade for the number: C. Evidence grade for the mechanism link: D.Analyst inferenceCorrelationalYouVersion (Bible App) · 2025
RoundYouVersion: an ethical habit at a billion installs
1 billion downloads across YouVersion's apps (announced in late Oct 2025, reported Nov 17, 2025), with installs up 12% YoY and daily usage up 18% (press). Book era: 100M downloads by July 2013 (Hooked, Ch7). Many causes.
M4 ×ReturnCase grade for the number: C. Evidence grade for the mechanism link: D.Analyst inferenceCorrelationalApple (Apple Watch) · 2024
RoundApple Watch: permission to rest
watchOS 11, announced in June 2024 and released Sept 16, 2024, lets people pause their rings for up to 90 days and resume their streak afterwards (9to5Mac, July 16, 2024). No public engagement metrics.
M4 ×ReturnCase grade for the number: C. Evidence grade for the mechanism link: D.Analyst inferencePress-reportedMint (Intuit) · 2024
RoundMint: stored value didn't save it
Mint shut down on March 23, 2024; Intuit pointed users to Credit Karma (NerdWallet). Many causes, chiefly the owner's strategy. The closure doesn't refute the mechanics; it shows they can't outweigh a business decision.
M5 ×StartReturnCase grade for the number: C. Evidence grade for the mechanism link: D.Analyst inferenceMany causesTOMS · 2019
RoundTOMS: One for One, retired
Nov 2019: TOMS retired One for One for a pledge of one third of net profits to grassroots causes (Retail Brew). Dec 2019: creditors took ownership of the company (CNBC). Many causes; a purpose mechanic is not a business model.
M4-M5 ×PaySpreadCase grade for the number: C. Evidence grade for the mechanism link: D.Analyst inferenceMany causesWaze · 2013
Waze: drivers fighting the traffic monster
Google bought Waze on June 11, 2013 (about $1.1B reported; later disclosed as about $966M). Waze's teams merged into Google's Geo organization in Dec 2022 and the app continues standalone. The contributor effects are qualitative.
M2-M3 ×DiscoverStayCase grade for the number: C. Evidence grade for the mechanism link: D.Many causesPress-reportedeBay · 2015
eBay: auction countdowns and seller stars
Qualitative, from the books (Actionable Gamification Ch6; Hooked Ch5). No outcome numbers are attached. Power Seller has since been replaced by Top Rated Seller (not re-verified).
M4 ×StayPayCase grade for the number: D. Evidence grade for the mechanism link: D.AnecdoteGoogle · 2014
Google Search: a habit rivals couldn't break
Qualitative, from the books: Hooked (Ch1) on search habit and switching costs, and Ch3 on the radically simple 1998 homepage; Actionable Gamification (Ch6) on making users feel smart, versus Google+. Consumer Google+ was shut down in 2019 (widely reported; not re-verified). No outcome number.
M4 ×StartReturnCase grade for the number: D. Evidence grade for the mechanism link: D.AnecdoteLinkedIn · 2012
LinkedIn: the profile strength meter
No public controlled data on the meter's effect on profile completion; the percentages that circulate online are unsourced and should not be cited. In Feb 2012 LinkedIn reweighted completeness toward fields members control, such as Skills (LinkedIn blog).
M3 ×StartCase grade for the number: D. Evidence grade for the mechanism link: D.AnecdoteZynga · 2012
Zynga and FarmVille: the Black Hat rebound
Book-era analyses: FarmVille reached 83.8M monthly users and Zynga a valuation above $10B, then lost more than 80% of its value (Hooked, Ch4). Take-Two acquired Zynga in May 2022 (not re-verified here). Many causes; grade D.
M5 ×StayPayCase grade for the number: D. Evidence grade for the mechanism link: D.Many causesAnecdoteDropbox · 2010
Dropbox: a referral reward that is the product
Sign-ups reportedly about 60% higher with two-sided storage rewards, and about 2.8M referral invites in one month of 2010 (secondary sources; grade D). Retellings disagree on the endpoints, so avoid the “3,900% in 15 months” headline.
M3 ×SpreadCase grade for the number: D. Evidence grade for the mechanism link: D.AnecdoteFacebook (Meta) · 2004
Facebook: exclusivity that became identity
Qualitative, from the books (Actionable Gamification Ch16; Hooked Ch8). No outcome number is attached to the staged rollout itself, and many causes drove Facebook's growth.
M1-M3 ×DiscoverCase grade for the number: D. Evidence grade for the mechanism link: D.Many causesAnecdote
Company names are used editorially; no endorsement is implied. No outcome is attributed to a single technique: real results have many causes. Every number carries its year and links to its source.
Your product
Find the stage your product is stuck in
Your product has its own version of these stories. Find the stage it’s in and the move that fits it. Connecting is free; a Pro subscription starts it.