Read the chart, read the business

You never need to read every cell. Three sweeps of the eyes extract everything the triangle knows:

• Across a row - one class of customers aging. The January row is the January joiners' whole life story: 100% at month 0, then the fade. Where the fade stops mattering more than where it starts - that is the curve shape, and Part 2 is entirely about it.
• Down a column - are newer classes better? Down the month-1 column, every cohort is compared at the exact same age, so this is the cleanest "is the product improving" read that exists. 45, 47, 50, 54 going down means every new class of customers sticks better than the last - whatever changed since January is working.
• Along a diagonal - calendar shocks. Cells on one diagonal share a calendar month. A dark or light stripe along a diagonal means something hit everyone at once regardless of age: an outage, a price change, a holiday season. (Its cousin: a vertical stripe means something hits every cohort at the same age - like a paywall that appears in month 2.)

The order matters. Across tells you what kind of business you have, down tells you whether it is getting better, diagonal tells you what the calendar did to it. Run all three before anyone starts explaining the chart to you, and you will usually be done before they are.

The shapes come with decision language. Practice these sentences; they are the difference between reading a chart and using one.

• Slide to zero: "Pause the acquisition scale-up - every cohort we buy is fully evaporating. The money belongs in onboarding and core product until this curve grows a floor." A sliding curve means no product-market fit for that audience yet; traffic cannot fix it, only feed it.
• Flatten above zero: "We have a durable base - each cohort leaves permanent residents. Now acquisition spend buys compounding assets, so this is the moment to scale it. And the floor height is the KPI: what moves it?" This is the classic PMF read (Sequoia's retention essays): a flattened curve of the key action at its natural frequency, plus month-over-month new-customer growth, is what product-market fit actually looks like in data (Casey Winters' formulation).
• The smile: "Something is resurrecting churned customers - find out what, and bottle it." Smiles are rare and precious: seasonal products, winback campaigns that land, or a feature that pulls dormant users back. Verify it is real resurrection and not a definition quirk, then fund whatever causes it.

One discipline ties all three together: acquisition spend is gated on the flatten. It is the boardroom version of "retention precedes monetization" from s1 - retention also precedes scale.

When Unilever paid $1B for Dollar Shave Club in 2016, the headline was the viral video. The diligence story was shape 2. Second Measure's transaction data showed DSC's cohort curves flattening at roughly 50% retention at six months - subscriber classes that fell and then stayed, cohort after cohort, leading its whole category.

• Why that shape is worth a billion dollars: a flattening curve means each acquired cohort leaves a permanent layer of recurring revenue. Stack the layers and revenue compounds even if marketing never improves - the "durable base" verdict, priced.
• Why the triangle beat the topline: topline revenue growth can be bought with spend (s1's treadmill). Flattening cohort rows cannot be faked that way - they show what customers do after the coupon is spent. This is why sophisticated acquirers and investors read cohort charts, not just revenue charts.
• Your takeaway as a leader: when you want to know if a business - yours or one you are buying - has durable revenue, ask for the cohort curves and look for the flatten. It is the same read at every scale.

A benchmark without a year and a source population is a rumor with an axis. Before any number becomes a target, make it answer three questions:

• What year is it from? 40/20/10 dates to the early free-to-play era (its exact attribution is murky even in the industry). Markets saturate, install quality changes, and a 2013 top-quartile bar quoted as a 2026 average will set your team an impossible target - then teach them to hit it by redefining the metric.
• Who is in the population? Puzzle games and hyper-casual games have wildly different profiles even within one 2025 study - and a games benchmark means nothing for a shop. For ecommerce, use ecommerce yardsticks: repeat purchase rate near the ~28% cross-industry average, and second orders concentrated inside 90 days.
• Which definition of retention? Session s1's homework question returns: a "day-7" computed on all installs vs activated users, or on a rolling instead of fixed-window basis, moves the number enormously with zero change in the business.

The sentence to keep: "What is the vintage and the population on that benchmark?" Asking it once, publicly, permanently upgrades the quality of numbers people bring you.