How To Better Understand Your Users

Why does it matter? Because the graph going up doesn't mean your product is working.

A tool Google Photos used with over a billion users fits on a single printed page and takes 10 minutes to build. Most founders have never made one — they're watching DAU graphs tick upward while users quietly churn underneath the line. This episode makes the case that aggregate metrics are structurally incapable of the diagnosis founders actually need.

• DAU/MAU trends upward even as users bail — new signups mask churn, and growth conceals the damage • The "dot plot" — a 2D grid of users vs. days — surfaces behavioral patterns no BI dashboard can reproduce • The event you choose to plot determines whether you get real insight or false confidence • A B2B account with 3 of 10 seats ever activated is a churn event already in motion

DAU/MAU is structurally incapable of telling you if your product is broken

The line is going up. That tells you almost nothing.

DAUs and MAUs lump every user together, so new signups mask churned ones. If you have any growth at all, the graph trends positive — even as users try your product once and abandon it. You can spend months building on a broken foundation while the dashboard reassures you.

The fix is a two-dimensional grid: users on rows, days on columns, a dot placed wherever a user completes a value-creating action. That's a dot plot. Each row is one person. Each column is one day. At a glance, you see who showed up Monday and never returned, who uses the product exclusively on weekends, which feature correlates with streaks of consecutive days. Your visual cortex detects patterns no pre-specified query could surface.

Build it before you build anything else. Modern AI coding tools can generate one in 10 minutes, and until you hit hundreds of users, it can be your only dashboard.

Chart 'app opens' and you get noise dressed as data — the event you pick is everything

An app-open dot plot creates the feeling of data-informed decisions. Dense grids, good vibes, zero signal.

The temptation when building your first dot plot is to log the easiest event — app opens, logins, page views. They fire regardless of whether anyone got value from the product. The result is a visualization that feels informative and isn't.

Pick something real: a photo shared, a song played, an invoice processed. That event — not a proxy — is what goes in the cell. Force yourself to answer one question before you start: what action proves a user got value today? Time granularity matters too. Weekly buckets hide the gaps where users stop showing up. Day-level resolution is where the patterns live.

PayPal didn't know what fraud looked like — so they stopped trying to specify it, and won

Nobody at PayPal could describe what fraud looked like — including PayPal. So instead of writing detection rules, they built a visualization of all transactions and put humans in front of it.

The humans couldn't articulate the pattern in advance. But staring at the grid, they'd point: "that thing there — that's different." Then they'd dig.

Dot plots run on the same principle. You're not confirming a hypothesis when you review them — you're letting your visual cortex find the anomaly you never thought to query for. Schedule a standing session where your team stares at the grid. Something will always stand out. That's where you go next.

3 of 10 seats activated, nobody used it twice a week — that $80K contract was already lost

Three of ten seats ever activated. Usage never exceeded two days per week, and it was sporadic at that. A recent YC company had this data — they just weren't looking.

They'd signed an $80K annual contract with a name-brand customer. Ten seats purchased. The champion who'd pushed for the deal left the company mid-year. The replacement asked why they were paying for the software. They opted out at renewal.

A dot plot of that account would have shown thin, declining usage from week one. For B2B products, seat activation rate and weekly usage density are the leading indicators. A sparse grid on any account isn't a data quality problem — it's an early warning to re-engage before the champion walks.

The tool you build at 10 users is the one Google Photos ran at a billion

The tool you build at 10 users is the same one Google Photos ran with well over a billion. At that scale, the team printed out dozens of pages — iOS users in France on one sheet, US web users earning above $80K on another — and sat in a room reading them together.

That's also the essential pairing: dot plots and cohort retention curves answer different questions and need each other. Retention curves tell you in aggregate whether cohorts stick over time. Dot plots show how individual users actually move through the product — which features drive streaks of consecutive days, where drop-off lives at the person level, what usage looks like on the accounts you're about to lose. The curve tells you there's a problem. The dot plot tells you what to fix.

The metrics founders default to were designed for decks, not diagnosis

The deeper shift here: DAUs, MAUs, weekly actives — these were optimized for investor updates. They're smooth, shareable, and built for confidence. Dot plots are the opposite: granular, unglamorous, impossible to spin. They show you exactly which users are slipping away and exactly which features are holding others. The founders who make them a weekly ritual — at ten users, at ten million — will always know something their competitors don't.

Build the grid. The patterns will find you.

---

Topics: user research, product analytics, startup metrics, retention, B2B SaaS, data visualization, YC advice, product-market fit, cohort analysis