The dashboard looks good. Signups up, revenue past a milestone, a chart worth screenshotting. Four months later a chunk of that cohort has gone quiet: cancelled, downgraded, or simply stopped logging in. The instinct is to find a new channel to replace what was lost.

Here’s the uncomfortable part: the number that explained it was in the data the whole time. Nobody looked, because the number founders do look at was still going up.

That’s the traction trap. Not the absence of growth. Real, measurable growth that isn’t evidence of what everyone assumes it’s evidence of.

Why the topline chart cannot show you the leak

Start with the arithmetic, because it’s the part most founders have never actually run.

If you add a constant amount of new revenue each month and lose a constant percentage of your base, you have a hard ceiling. New revenue is linear. Churn is proportional to size. They converge, and growth stops. Jason Cohen calls the resulting number Max MRR:

Max MRR = New MRR ÷ monthly cancellation rate

Cohen’s worked example runs at $1,000 of new MRR a month and 5% cancellation, plateauing at $20,000. Scale it to something closer to a post-PMF business: at £10,000 of new MRR a month and 5% monthly cancellation, you plateau at £200,000 MRR. At 2% cancellation, the same acquisition engine reaches £500,000. Same marketing, same sales team, 2.5x the company.

Two things follow, and both explain why the trap is so hard to see from the inside.

One: the ceiling arrives long before the market runs out. Founders reach for market-size or channel explanations when growth stalls. Usually it’s churn, and it was churn all along, it just wasn’t the dominant term in the equation until the base got big enough.

Two: revenue is a lagging indicator of the thing that matters. Cohen’s analysis of Buffer’s public metrics shows this cleanly. Buffer ran roughly 6% monthly cancellation from 2014 to 2020, accelerating new MRR the whole time and growing linearly to about $140k MRR.

Cancellation never improved. Then it ticked up another point to 7%, and revenue fell, from $140k to $90k, even though new MRR was still large. The ceiling had dropped below the business. In 2023 Joel Gascoigne reset the company around a narrower ideal customer and a pricing model that fitted them; new MRR rose, cancellation eased slightly, and growth restarted. Had Buffer instead cut cancellation from 8% to 3% over 2014 to 2019, Cohen’s model puts revenue at roughly double where it landed.

Nothing about Buffer’s monthly growth chart, in the good years, said “this ends at $140k.” Max MRR said it years ahead.

The 2026 version of the trap: expansion is doing the hiding

The traditional framing is that new acquisition masks churn. That’s still true. But there’s a second mask, and it’s the one doing most of the work right now.

Net revenue retention includes upsells, cross-sells and price increases. Gross revenue retention doesn’t. So an NRR of 105% can be 90% GRR plus 15% expansion, a healthy business, or 70% GRR plus 35% expansion, a business bleeding customers and covering it with the ones that remain. From the NRR number alone, these are indistinguishable.

And the mask is getting heavier. Benchmarkit’s 2025 report found:

  • Median NRR at 101%, down from 105% in 2021 and 103% in 2022 (n=228)
  • Median GRR sliding from 90% to 88% across three years (n=225), though Benchmarkit notes this could partly reflect selection bias in who responded
  • Expansion ARR rising to 40% of all new ARR, up five percentage points in a single year, and above 50% for companies over $50M ARR (n=81)

Both retention numbers have softened by roughly the same amount. What’s changed underneath is the mix: a growing share of the revenue holding NRR near 100% is coming from selling more to existing accounts rather than from those accounts staying. Benchmarkit’s own read is that companies are increasing their dependence on expansion ARR. A board deck showing NRR alone in 2026 is therefore showing less than the same slide did in 2021.

There’s a cost angle too. Benchmarkit puts the median new-customer CAC ratio at $2.00 of sales and marketing spend per $1 of new ARR (bottom quartile: $2.82, n=73), against $1.00 for expansion ARR (n=21, and very few companies measure this at all). Buying a dollar of revenue from a new logo costs roughly twice what buying it from an existing account does. Which makes replacing churned customers the most expensive growth available, and it’s where most early-stage budgets go first.

Where your numbers should actually sit

Benchmarks are only useful segmented by contract value; a $20/month product and a $250k/year product have almost nothing in common. Two reference points.

Annual revenue retention by ACV, from SaaS Capital’s 2025 survey of private B2B SaaS:

ACV bandMedian NRRTop quartileBottom quartile
$25k–$50k102%111%97%

Higher ACV correlates with higher NRR and, at the top end, higher GRR: longer sales cycles, real implementation, dedicated account management. The practical implication is that if you sell below $25k ACV, you should judge yourself against the low end of the gradient. An NRR near 100% down there is a different achievement to an NRR near 100% at enterprise pricing.

New-customer cohort retention, from ChartMogul’s SaaS Retention Report across 2,100+ SaaS businesses:

Cohort ageTop quartile, any ARRTop quartile, ASP >$500/moTop quartile, ASP <$10/mo
3 months90%98%87%
12 months70%——

Two details in that data are worth more than the headline numbers.

First, ChartMogul’s own read: “A low retention rate in the first 3 months signals issues with onboarding or acquisition (i.e. acquisition of bad-fit customers).” Ninety-day churn is rarely a product problem. It’s a targeting problem or an activation problem, and those get fixed in different places.

Second, the shape of the curve. Retention decays exponentially through month 11, then drops sharply in months 11 and 12 as annual plans come up for renewal. If you sell annual contracts and only look at monthly logo churn, your first eleven months look reassuring and your renewal cliff is invisible until it arrives.

ChartMogul’s 2023 report draws on cohorts from January 2021 to December 2022. Since industry GRR has softened since, treat these levels as a generous benchmark rather than a current one. The shape of the curve is the durable part.

The Traction Trap Test: four numbers, run together

None of these work alone. That’s the point: the trap survives precisely because each number, viewed in isolation, has an innocent explanation.

#Pull thisPassingFailing
1Cohort retention at 30/60/90 days for your last 4–6 monthly cohortsCurves flat or improving cohort over cohortAggregate revenue rising while cohort curves flatten or decline
2GRR alongside NRR, same periodGap under ~10 points; GRR carrying the numberNRR healthy only because expansion is wide; GRR falling
3Retention split by acquisition sourceRetention broadly consistent across channelsOne high-volume channel materially below the rest
4Max MRR (new MRR ÷ monthly cancellation)Ceiling comfortably above your 24-month planCurrent MRR approaching the ceiling

1. Cohort retention, not aggregate growth. Take your last four to six monthly cohorts and track what share of each is still active and paying at 30, 60 and 90 days. If total revenue is growing while those curves are flat or declining, growth is coming entirely from new acquisition covering a leak, not from a product getting stickier.

2. GRR next to NRR, every time. If you only report one retention number, report gross. Net without gross tells you the business is holding together; it doesn’t tell you whether it’s holding together because customers stay or because a handful of accounts keep buying more. Those need completely different responses.

3. Retention by acquisition source. A channel producing high volume and disproportionately low retention isn’t a growth engine, it’s a treadmill: it needs constant spend to replace what it churns, and at $2.00 of CAC per $1 of new ARR, scaling it makes the hole bigger, not smaller. This is also the single fastest diagnostic most teams have never run. The data is usually already sitting in the CRM, just never joined up.

4. Max MRR. Run the division. If current MRR is within striking distance of the ceiling, no acquisition plan will fix it, because you’re optimising the numerator of a fraction whose denominator is the problem.

Then talk to five people who left

The numbers tell you that. They don’t tell you why, and the why is where the fix lives.

Not an exit survey with checkboxes, an actual conversation with five customers who churned last quarter. You’re listening for the job the product failed to do, not a list of missing features. Feature requests are easy to act on and rarely address the real reason someone left.

Two reasons this is worth the discomfort.

Bain surveyed 362 companies and found 80% believed they delivered a superior experience to customers. Asked directly, customers said 8% of companies actually did. That’s the delivery gap, and it isn’t a gap in effort. Over 95% of the management teams Bain surveyed described themselves as customer-focused. It’s a gap in knowing, and it doesn’t close on its own.

And Fred Reichheld’s HBR work on defection makes the structural point: companies fail to learn from churn not because the lessons aren’t there, but because defection is unpleasant to study, hard to define, and career-hazardous to analyse honestly. The information exists. The process for extracting it usually doesn’t.

Which is why “nobody’s complaining” means nothing. Most churned customers never complain on the way out. Absence of complaint is absence of a listening mechanism, not evidence of satisfaction.

Common mistakes founders make with the traction trap

  • Reading growth as product-market fit. Growth is necessary evidence, not sufficient evidence. It has to be checked against cohort retention before it means what you want it to mean.
  • Reporting NRR without GRR. With expansion now supplying 40% of new ARR industry-wide, NRR alone is a materially weaker signal than it was five years ago.
  • Treating churn as a support ticket. Elevated churn is data about who you’re attracting and what they hired you for. Routed to a support backlog, it buries the one number most likely to explain what’s wrong.
  • Scaling a channel before checking its retention. More volume from a low-retention channel doesn’t fix the leak. At a $2.00 CAC ratio, it accelerates the cash burn required to replace people who were never going to stay.
  • Assuming retention transfers to a new market. The habits, trust and workflows that made a product sticky in its first context don’t travel with the signup form. Buffer’s 2023 strategy reset went the other way, narrowing back to the customer the product actually fitted rather than reaching for new markets. Treat every expansion as a fresh retention hypothesis.

The takeaway

Early traction is real information. It’s just only answering the acquisition question.

Before it becomes the story you tell investors, your team or yourself, put it next to four numbers, cohort curves, gross retention alongside net, retention by source, and your Max MRR ceiling, and five honest conversations with people who left. That’s a week of work. It’s considerably cheaper than a year of acquisition spend layered on top of a leak nobody measured.

This gets sharper once you’ve fixed who’s coming in the door: a tightly defined ICP means you’re checking whether the right people stay, not whether people in general do. And it connects directly to activation: if buyers never reach real time-to-value, the 90-day cohort number was decided before they ever opened the product.

If your growth numbers look good and you’re not certain retention underneath them would survive a cohort-level check, start with a growth diagnosis before the next acquisition push papers over a leak that’s still there.

Sources