%20scilla.studio.svg)
Your product grows for exactly two reasons: the users you already have stick around, and new users show up faster than the old ones leave. That's the whole game. A cohort-based growth model makes it concrete by splitting your user base into two pools: a stable base of long-tenured users who churn slowly, and the new cohorts arriving each month, retained at your monthly rate and amplified by whatever virality you have. Total active users next month = the base that survives + the new cohort that survives + the viral users that cohort brings. Get those two pools right and you can stop guessing whether a feature, a retention fix, or a referral loop actually moves the line.
A single MAU number tells you the line moved; it never tells you why. This model gives you the why: what the levers actually do, and where the benchmark bands sit.
Say your MAU was flat last month. Did nothing happen? Almost certainly not. What probably happened is that your stable base quietly leaked a few percent, and your new cohort almost exactly refilled the hole. Flat MAU can hide a base that's rotting and a top-of-funnel that's papering over it, or a healthy base with a top-of-funnel that's dried up. Same flat line, opposite diagnoses, opposite fixes.
That's the part most growth guides skip: net growth is the difference between two much larger flows. A 2% monthly change in MAU might be a 10% inflow fighting an 8% outflow. You can't manage a difference you can't see. So before you touch a single lever, you separate the pools.
Stable base. These are your long-tenured users, the ones who already survived the brutal first-month drop-off and turned into a habit. They churn, but slowly. In the model that powers the scilla.studio benchmark tool, mature users churn roughly ten times less than a brand-new cohort: their monthly retention is 1 − (1 − M1) × 0.10. That 0.10 is a deliberate modeling simplification in the benchmark engine, not a measured constant. It encodes the well-documented fact that tenured users churn far slower than fresh ones, without pretending the ratio is exactly ten everywhere. If new users retain at 55% month-one (M1), the mature base retains at about 95.5%. Survive the funnel once and you're a very different kind of user.
New cohorts. Each month's new arrivals. They retain at your raw M1 rate (the percentage of a cohort still active a month later), which is far lower than the mature rate, because month one is where most products bleed. The survivors of each new cohort then graduate into the stable base the following month. That graduation is the entire engine of durable growth: today's fragile new users are tomorrow's reliable base, but only the ones who make it past month one.
Here's the interesting part: these two pools behave so differently that averaging them together (which a single retention number does) destroys the signal. A cohort model keeps them separate on purpose.
Two inputs describe the world last month:
A0 − N0.Two levers describe how it evolves:
And the forward simulation, month by month:
stableBase[0] = A0 − N0
stableBase[t] = stableBase[t−1] × matureRetention + newUsers[t−1] × M1
newUsers[0] = N0
newUsers[t] = N0 + K × newUsers[t−1]
matureRetention = 1 − (1 − M1) × 0.10
Read that second line slowly, because it's the whole model. Each month the base is what's left of last month's base after mature churn, plus the survivors of last month's new cohort graduating in. The new-user line is your baseline acquisition N0 plus a viral kicker proportional to last month's new users. Total MAU is the two pools added back together.
Worked example. Start with A0 = 85,000 MAU and N0 = 5,000 new users last month, so the base is 80,000. With M1 = 55%, mature retention is 95.5%. Next month the base is 80,000 × 0.955 + 5,000 × 0.55 = 76,400 + 2,750 = 79,150, before you add the new cohort back on top. Notice the base alone dipped; whether your MAU grows depends entirely on what the new-user line does.
This is where the model earns its keep: small input changes produce qualitatively different futures.
| M1 retention | K-factor | What happens to your base |
|---|---|---|
| 100% | 0 | Nobody leaves, every new user graduates in. Base grows by exactly N0 every month. |
| 55% | 0 | Roughly flat. New-cohort survivors just offset mature churn. |
| 55% | 0.6 | Steady growth. Viral new users feed the base faster than it decays. |
| 30% | 0 | Slow decline. Too few new users survive to replace what the base loses. |
| 80% | 1.0 | Strong growth. High retention plus a near-exponential new-user line. |
(Source: the growth-model engine behind benchmark.scilla.studio, N0 = 5,000, A0 = 85,000.)
A few properties worth internalizing:
N0 / (1 − K). At K = 0.6 and N0 = 5,000, that ceiling is 12,500 new users a month: viral, but bounded (and yes, that is still the honest ceiling even if your growth deck drew an exponential).The honest takeaway: retention and acquisition are multiplied together, not separate problems. A great K-factor on top of a leaky M1 fills a base that can't hold water. That's why, when you read the benchmark bands below, you should look at retention before you celebrate any viral number.
Benchmarks here are context, not targets: directional bands to tell you whether your inputs are plausible, not goals to hit. They split hard between B2B SaaS and consumer, because the two follow different growth mechanics and comparing them directly is a category error.
The model's M1 is a monthly survival rate, closest in spirit to early-period retention. The tool's published retention bands (measured from first use) follow:
| Metric | B2B SaaS | Consumer apps |
|---|---|---|
| Day-1 retention | 50 to 70% | 20 to 30% |
| Day-7 retention | 40 to 60% | 8 to 15% |
| Day-14 retention | 35 to 55% | 4 to 8% |
| 90-day retention | 25 to 35% | 1 to 4% |
Sources: Pendo Product Benchmarks, Amplitude, Mixpanel (B2B); Adjust, AppsFlyer, UXCam (consumer).
The shape matters more than any single point. B2B retention drops, then flattens into a durable plateau, and that plateau is your stable base forming. Consumer curves drop far steeper and flatten much lower, which is why consumer products lean so much harder on acquisition and virality to grow. For the long-tail end of this, see 90-day retention and the fuller retention rate benchmarks.
| Metric | B2B SaaS | Consumer apps |
|---|---|---|
| K-factor (viral coefficient) | 0.1 to 0.3 | 0.3 to 0.7 (rarely >1) |
Sources: Reforge, Andrew Chen.
B2B virality is usually weak: anything above 0.3 is unusually strong unless you've got real collaboration loops built into the product. Consumer can reach 0.5 to 0.7, and a K above 1 (genuine self-sustaining viral growth) is extremely rare. So when you plug a K into the model, be honest: most products live well under 1, which means virality accelerates growth but doesn't replace the need for paid or organic acquisition (N0). The compounding ceiling of N0 / (1 − K) is doing the quiet, realistic work here, not some exponential fantasy.
If you want the full story on measuring K across multiple loops, and why cycle time matters as much as the coefficient itself, that belongs in a dedicated K-factor piece. The short version: a lower K with a faster loop can out-grow a higher K with a slow one, and this cohort model assumes a monthly cycle, so a product whose loop fires weekly will outperform what these monthly numbers suggest.
Run your real N0 and A0 through the model, then ask three questions:
The failure mode to avoid: optimizing the wrong pool. Here's what I actually think: pouring money into acquisition while your base rots is the most expensive mistake in growth, and it's the one this model exists to catch.
The free scilla.studio benchmark tool runs exactly this cohort model live. Enter your MAU and new-users-from-last-month, drag the M1-retention and K-factor sliders, and watch your 12-month growth curve redraw against the B2B and consumer benchmark bands, in a couple of minutes, no signup. It's the fastest way to find out whether your growth is base-led or cohort-led, and whether your inputs are even plausible against the industry ranges above.
What is a cohort-based growth model? A growth model that splits your users into a stable base (long-tenured users who churn slowly) and monthly new cohorts (recent arrivals retained at your monthly rate and amplified by K-factor). Total growth is the base that survives plus the new cohort that survives plus any viral users it brings, rather than tracking a single MAU number.
What's the difference between the stable base and new cohorts? The stable base has already survived the first-month drop-off and churns roughly 10× slower than fresh users. New cohorts retain at your raw monthly rate (M1), which is much lower, and the survivors graduate into the stable base the following month.
What is M1 retention in this model?
M1 is the fraction of each new cohort still active one month later. From it the model derives a mature retention rate of 1 − (1 − M1) × 0.10 for the long-tenured base. B2B SaaS early retention typically runs higher than consumer (Pendo, Amplitude), and consumer drops far steeper (Adjust, AppsFlyer).
Does K-factor have to be above 1 to grow?
No. With K below 1, your new-user line converges to a stable ceiling of N0 / (1 − K): still growth, just bounded. K ≥ 1 means true exponential viral growth and is extremely rare; typical B2B K-factor is 0.1 to 0.3 and consumer 0.3 to 0.7 (Reforge, Andrew Chen). Retention graduating cohorts into the base does most of the durable work even at K = 0.
Why can flat MAU be misleading? Because net growth is the small difference between two large flows: your base leaking and your new cohort refilling. Flat MAU can mean a healthy base with a dead funnel, or a rotting base propped up by acquisition. The cohort model separates the two so you fix the right one.
