流失会自己变大，获客不会：Jason Cohen 诊断增长停滞的五问

Lenny Rachitsky frames the episode around the pain of a product that starts growing and then stops, quoting Will Smith's line that losing fame is miserable. Jason Cohen says many companies are in that spot, and the cause could be the economy, AI uncertainty, or simply size, since nobody grows 2x a year forever. More often than a sudden event, he sees growth slowly getting slower, like running through mud.

Cohen draws on four companies, two of them unicorns, and about 60 startup investments. He offers a series of questions rather than a checklist, asked in a specific order, because if an earlier problem isn't fixed, tuning a later one won't matter. It is like a funnel where one step falls apart: fixing the bottom of it does little.

The first question is whether customers are leaving, meaning logo churn. He calls it the worst problem because nothing can be done once a customer is gone, and it often correlates with negative reviews that hurt growth further. He also thinks of the gauntlet a customer survives, from discovering you to onboarding, and says that cancelling after all that means something is fundamentally broken.

Cohen says people throw around vague cancellation rates, calling 3% a month terrible while others call 5% to 7% fine. He prefers a metric that makes the damage visceral. His key point is that cancellations grow faster than marketing, since marketing channels don't care how many customers you have, while cancellations are a percentage of your base.

Imagine tripling the customer base overnight: AdWords and SEO would deliver the same number of leads, but the number of cancellations would triple. That is why cancellations always overtake marketing, whose growth depends on finding new channels. Lenny likens it to a leaky bucket, and Cohen adds that the leak automatically gets bigger.

The ceiling is simple to compute: new customers added per month divided by the monthly cancellation rate. At 100 new customers a month and 5% cancellation, the company will never exceed 2,000 customers, and growth slows dramatically as it approaches that number. The same math works for revenue churn, but Cohen uses customer counts because a stark customer cap motivates action.

Cohen explains that departing customers don't want to talk and that cancellation dropdowns are mostly noise. At SmartBear, one option was chosen more than others, and after randomizing the list's order all options were picked equally. His advice is to ask open-ended questions, and to ask "what made you cancel?" rather than "why did you cancel?", which he borrowed from a Groove case study where usable responses rose from 10% to 20%.

He says "too expensive" is never the real reason, because the customer already looked at the pricing page and decided to buy. He uses the analogy of a death certificate: stopping breathing is the proximate cause, but undiagnosed diabetes is the cause that matters. He dislikes the idea of a single root cause and prefers an array of interlocking causes, such as a missing Linear integration, or a market segment whose projects tend to end.

He also advises talking to customers who are in trouble but haven't yet cancelled, using signals like no data uploaded, too many or too few support calls, or no logins. When unsure what to do, he suggests working on onboarding, since most cancellation happens early and small changes there have large effects. He compares it to a YouTube retention curve, where moving the first-30-second drop-off from 50% to 55% could lift the end audience from 20% to 25%.

Lenny asks about using AI to summarize cancellation responses. Cohen says an LLM is an averaging machine, so it's good at themes and bad at surfacing interesting, non-average details, and the themes are rarely surprising. His workaround is to ask for themes and then have it pull every specific detail under each theme, with the customer who said it, so a human can spot what is actionable.

He recommends still reading the responses yourself, using AI at most to clean up bad grammar. Lenny adds that unless it is a large consumer app the volume is low, so reading is feasible. On catching customers early, Cohen says you don't need many customers to talk to those in trouble, but you need real data to know which behaviors correlate with canceling, and you must compare what good customers have that bad ones don't.

The second question is whether pricing is correct. Cohen cites Patrick Campbell, who has 4,200 data points about startups: prices are too low because you guessed and haven't changed them. People fear raising prices because of the textbook demand curve, but in practice, he says, signups often stay the same or even go up after a price rise, even for solopreneurs on Twitter.

He tells of a founder selling to enterprises and governments for $300 a year, who agreed to change it to $300 a month. He still got one or two signups a week, so Cohen told him to raise the price again rather than simply spend the extra money. The reason, Cohen says, is that pricing selects the market: a thousand-employee company sees a $2 or $100 a month product and assumes it isn't mature enough.

The demand curve is therefore more like a mesa than a line sloping down, and only the lowest end of the market has the slope the textbook worries about. Raising prices moves you into a different market, and retention and other metrics tend to improve with higher prices. Cohen adds that price is also structure and positioning, not just the number on the page.

Cohen illustrates positioning with a lightly fictionalized story of a company called Double Down that halves AdWords costs. For a customer spending $40,000 a month, a $5,000 price saves them $15,000 and everyone is happy. But framing it as cutting costs caps the price below the savings, and it tells the CEO the CMO saved money rather than grew the company.

If the same product is pitched as doubling leads per month at the same CAC, the customer is already willing to spend $40,000 for the current number of leads, so they'd pay about $40,000 to double it. That is eight times the price for the same product, and the CMO can tell the CEO that growth doubled. His lesson for product managers is to sell more of what the company already values, such as growth, retention or competitiveness, since the cap is perhaps an order of magnitude higher than for savings.

Lenny points to Jen Abel's insight that enterprise contracts commonly land between $75,000 and $150,000, where deals get easier. Cohen cautions that pricing isn't a standalone knob: moving upmarket means needs like SOC 2, integrations or professional services, and some companies, like Buffer, decide culturally they don't want to sell to big companies. Raising prices can mean changing the whole strategy.

The third question is whether existing customers are growing. Since cancellations can't be zero, expansion from upgrades or usage is the thing that can offset them in proportion to your size. The metric is net revenue retention: take today's customers and see what revenue remains a year later, with cancellations and downgrades pulling it down and upgrades pushing it up.

Cohen argues that NRR alone can mislead. If customer count falls too fast, too few customers remain to upgrade, and a 20% loss needs a 25% gain to recover, not 20%. Of more than a hundred public SaaS companies, only about two have NRR below 100%, and those have poor financials, while the median NRR at IPO is 119%.

To raise NRR, he says, the customer should agree they are getting more value when they pay more, ideally feeling that doubling the price brings five times the value. That means measuring the value customers receive, using proxy metrics or qualitative conversations if needed, rather than only usage metrics. His mantra is to create more value for the customer and then split it with them, remembering that it is not a charity.

Responding to Lenny's note about land-and-expand, Cohen warns that pricing moves driven by investors or internal pressure should pass one test: is it actually good for the customer, as well as for us? He adds that the same financial mechanics hold in consumer and small-business products, which lack easy upgrades. There the levers are a second product or existing customers bringing friends.

The fourth question is whether acquisition channels are saturated. Cohen says there are real limits, like the number of searches in your area or the inventory of magazine ads, and channels tend to decline over time. He calls this the elephant curve: an S-curve whose top begins to sag. Audiences have seen your message already, and channels like magazines or conferences can quietly decline.

His test is whether you know which channels are saturated. If not, then adding a feature and asking marketing to flog it isn't going to work. He suggests creative new channels, such as Constant Contact holding in-person workshops for small businesses, or HubSpot selling through agencies, which became about 50% of its revenue after four or five years, as WP Engine does too.

Where there are no more channels, he says, the answer may be a new product or market. His rule is to plant one foot in an existing strength or asset and move the other foot into the risky new thing. Lenny adds ecosystems as an emerging channel, citing Emily Kramer and a Wiz growth lead's idea of starting with 10,000 rather than zero, along with ChatGPT's app store.

The last question is whether you actually need to grow. Cohen says that depending on how you define the company, the answer might be a different product or market, or maximizing profit instead of revenue. He cites 37signals and bootstrapped companies where founders earn well and don't want a bigger organization.

Considering "if you're not growing, you're dying," he says stagnation for years is rarely fulfilling for founders or employees, though some people, like a CPA with good clients, are happy staying put. He suggests reading the phrase as applying to you personally: if you're stagnant and nothing incremental will change it, something dramatic, such as leaving or selling, may be needed.

Lenny notes that for bootstrapped founders, flat revenue can be acceptable, and Cohen adds that the question also sets boundaries, like refusing to grow a newsletter with low-quality content. He argues that early on you have a looser bar and tighten it as you can afford. On big decisions about quitting or moving, he says probability and expected value don't work, and the next book he wants to write is about making decisions under uncertainty.

Cohen describes his book, Hidden Multipliers, as being about small decisions with huge impact, like moving cancellation from 5% to 4% or improving onboarding. Asked for a common thread across the conversation, he says it is that customers actually get value: the product promises the right thing, delivers it, customers can onboard, they realize it, and you measure it. Then pricing, retention and growth have a good foundation, though he admits such a platitude isn't actionable alone.

In AI Corner, he says Gemini is good at converting a chart image into a table that pastes correctly into Google Sheets, letting him test theories on real data. In Contrarian Corner, he argues that A/B testing doesn't work on important things like strategy, and that for details most positive results are false positives, because real wins are rare. Lenny counters that it helps at huge scale, and Cohen recalls Shopify's CTO saying a third of effects disappear when rechecked with a holdout group.

Cohen recommends On Writing Well by William Zinsser for writing and Crossing the Chasm for product, noting that few people have actually read the latter. He and Lenny discuss how a market is defined by its members respecting each other's opinions, which is why case studies don't carry across markets. He is rewatching the TV show ER with his 16-year-old daughter, and his favorite recent products are Wispr Flow for dictation and anything made by Anker.

His life motto is "Be yourself, everyone else is taken," which he notes is attributed to Oscar Wilde without evidence. He explains the 2017 Ernst & Young Entrepreneur of the Year award behind him, which he co-won with WP Engine CEO Heather Brunner, whom he considers a late-joining co-founder. Listeners can find his free articles at asmartbear.com and pre-order his book at hiddenmultipliers.com.