The D5/D7 Ungameable North Star
One metric that fuses retention, engagement and growth — tracked as growth rate and cohort slope, never absolutes.
- Difficulty
- Easy
- Time to result
- ~weeks to results
- Steps
- 5
- Confidence
- 95%
The Browser Company tracks exactly one growth number: how many people use Arc on at least five of seven days. Because the bar is five days a week and requires real use (opening a tab), the single number simultaneously encodes retention, engagement and growth, and cannot be gamed by shallow activity. They then refuse to look at the absolute value, watching only week-over-week growth rate and whether retention is improving cohort by cohort.
Origin
Josh Miller's own framing, developed as the counterweight to the feelings-first philosophy — the honesty check. He notes other companies call the same shape L5/L7 (5-of-7 days), so the metric form is industry-known; the discipline of using it as the only number, and of ignoring absolutes in favour of rate and cohort slope, is his team's practice.
Core principles
- 01Pick the one metric that captures retention, engagement and growth at once, so there is nothing to trade off against it.
- 02Set the bar high enough that it cannot be gamed — five days a week is a bar very few pieces of software clear.
- 03Absolute numbers are meaningless if you are going to be big: any point in time will look inconsequential in hindsight.
- 04Track rate of change, not level: week-over-week growth rate, and retention improving cohort by cohort.
- 05Publicly refuse to extrapolate. Miller repeatedly says the current numbers will not hold.
How to run it
- 1
Define an active day that requires real use
Specify the event that counts. For Arc, merely launching the app does not count — you must open a tab on that day. The definition is what makes the metric ungameable.
Pro tip Test the definition adversarially: could a growth hacker inflate it with a notification or an autostart? If yes, tighten the event.
- 2
Set the frequency bar at habit level
Choose the days-per-week threshold that means the product has become part of life — five of seven for a browser. This is what folds engagement and retention into one count.
Watch out The right threshold is category-specific. A five-of-seven bar on a product that is legitimately weekly will just produce a flat, useless line.
- 3
Count humans, not sessions
Report the number of people clearing the bar. Because it is a count of people, growth is baked into the same number that measures quality of use.
- 4
Report growth rate, not level
Convert the count into week-over-week growth rate and make that the headline the whole company sees. Miller's team ran 10%+ weekly for about eight months on this number.
Pro tip Pair the growth rate with a stated expectation that it will decay, so the team does not treat a slowdown as failure.
- 5
Judge retention by cohort slope, not by today's value
Ignore the current retention figure. Ask instead whether each new cohort retains better than the last — even as you move further from your earliest, most passionate adopters.
Pro tip Improving cohort retention while moving away from early adopters is the strongest available evidence that the product, not the audience, is doing the work.
Watch out A rising absolute retention number can hide a deteriorating cohort curve if your mix is shifting.
In the wild
The Browser Company tracks D5/D7 — people using Arc at least five days out of seven, where a day only counts if a tab was opened. It is the only growth metric they obsess over; they do not look at DAU retention or WAU retention at all.
→ Roughly 10%+ week-over-week growth in that number for about eight straight months, with D5/D7 retention by cohort somewhere between the low-to-mid 30s and low 40s percent.
Rather than celebrating a strong retention curve from a year prior, Miller's point of pride is that twelve months later the curve has been inching up cohort by cohort, despite the company having moved well past its earliest and most passionate adopters.
→ A retention curve improving over time, which Miller treats as the meaningful signal and the absolute level as noise.
Common mistakes
Running a basket of metrics
The power of D5/D7 comes from there being one number that cannot be traded against another. The moment you also track DAU, WAU and session length, teams optimize whichever one is easiest to move.
Obsessing over the absolute number
If you succeed, every point in time will look inconsequential in retrospect. Fixating on today's total produces false pride and false panic; the rate and the cohort slope are the information.
Letting the metric become the goal
In Miller's own system D5/D7 exists to keep the team honest about whether the feelings-first bets worked. Promote it to the generator of ideas and you land right back in the Facebook OBPS trap he is arguing against.
Is it for you?
Best for
Early-stage consumer product teams who want one honest, ungameable growth scoreboard instead of a metrics dashboard nobody trusts.
Not ideal for
Products with inherently low natural frequency (annual filings, insurance, real estate) or enterprise tools measured on seats and contract value rather than daily habit.
From the transcript
“we really focus on one key metric as it relates to tracking our growth or how we're doing we call it D5 D7 a lot…”
“you zoom out and out and out any point in time will be inconsequential and small so what we really look is growth rate week…”
“we don't look at retention in this moment we look at is it improving cohort by cohort”
From the episode
Competing with giants: An inside look at how The Browser Company builds product
Josh Miller (CEO)