Why product-led growth fails: You are probably fixing the wrong one

“Product-led growth isn’t working” is not a diagnosis. It’s three different failures wearing the same sentence, and they have opposite fixes. One is solved by shortening the path to value. One is solved by raising prices and letting the smallest customers go. One is solved by hiring salespeople, which is the exact move the first two teams must not make.

The most common version: you launch a free tier, signups look healthy, conversion stays flat, and you conclude the model doesn’t work for your product. That conclusion is usually half right: something is broken, but it isn’t what you think.

Why does product-led growth fail?

Product-led growth is a go-to-market model where the product itself does the work a sales team would otherwise do: people find it, try it, and decide to pay without a conversation, and the product carries acquisition, conversion and expansion. It fails in three ways: activation, unit economics, or a sales ceiling. Activation: users never form a habit. Unit economics: the price can’t cover what serving each account costs. Ceiling: self-serve carries you until deals need a human. They are confused for each other because all three produce the same chart, flat revenue against healthy signups, and because activation is the one with a name everybody already uses.

The three PLG failures, and why their fixes are incompatible

In failure one, users never come back, so the price was never tested. In failure two they do come back, and serving them costs more than they pay. In failure three nothing is broken at all: growth stalls where deals need a human, and the product cannot reach that.

Pick the wrong failure and you fix something that was never broken.

How do you tell which PLG failure you have?

Before the three, one question costs nothing and settles whether the rest apply. Put a stranger in front of your product with nobody to help them. If they get somewhere real, product-led growth can carry you; if they stall and email support, it cannot, and no amount of execution changes that. That answer is available before you build anything, which makes it the only one of these you can settle today and the cheapest to get wrong. What to do when the answer is no comes later in this chapter.

Then test the three in order. Each one only means something if the previous answer was yes.

1. Users who sign up do not come back. Chart weekly return rate through Week 8 for a cohort from 90 days ago. A large first-session drop is normal (the widely repeated industry rule of thumb puts it at 40-60% for free trials, though nobody has ever sourced it),¹ so what matters is the shape after that. A curve still falling toward zero at Week 4 is this failure, and nothing downstream is worth touching yet. A curve that flattens anywhere above zero means some cohort found real value; that cohort is your instruction manual.
2. The survivors do not cover what they cost. Take customer acquisition cost (CAC) plus support and infrastructure for your median paying account, and compare it to what that account earns you after costs. If the payback period is long enough to make you uncomfortable, this is your failure, and more volume makes it worse. Twelve months is the figure most often quoted as the line, but whose line matters. It is a venture convention, and it assumes a balance sheet that can wait a year to be repaid. Rob Walling, who has funded and advised bootstrapped software companies for two decades, puts the workable range for a company without investors at two to six months, and says that at the peak of Drip, the email tool he founded, he ran seven or eight.² Those are one practitioner’s numbers, not a measured threshold. The useful part is the question underneath them: a payback period is a claim about how long your own cash can wait, so use your own model’s history if you have any, and if you have none, ask who could survive the answer you are about to accept.
3. Your largest deals stall where no product change reaches. Find the last step every opportunity above your top self-serve tier completed. Procurement, security review, and contract terms mean you’ve hit the ceiling, and the answer is human.

A company that declares PLG dead without asking question one has not diagnosed anything. The model wasn’t wrong. It was unfinished.

Failure one: the product never earned the habit

In B2B this failure is quiet. A team adopts your tool for one project. Two people use it daily for three weeks, the project ships, and the workspace goes cold. Seats stay paid until renewal because nobody notices, then churn arrives as a surprise. Your monthly actives look stable the whole time, because new projects keep replacing dead ones at roughly the same rate. The product was never wrong. It just never became the place work happens, and a tool people only use for launches gets opened only when they launch.

Clubhouse: scarcity fills a room once, habit fills it on a Tuesday

Clubhouse is the same failure at a much larger scale. The live-audio app’s valuation moved by more than an order of magnitude in under a year, reaching $4 billion by April 2021 on invite-only scarcity, backed by Andreessen Horowitz, Tiger Global, and DST Global.³ Two years later its founders announced they were “scaling back our org by over 50%” to reset the company.⁴ Growth had flattened before a revenue model ever existed.³ Nothing about the waitlist and scarcity mechanics failed; they delivered millions of people to the front door. What failed was everything after the door. Scarcity is a reason to show up once. It isn’t a reason to come back on a Tuesday.

The fix: find the action that predicts retention, then rebuild the first session around reaching it. The action is your aha moment, and the distance to it is time to value. The commonly cited target is value inside five minutes, which is a working convention, not a measured benchmark, so treat it as a direction rather than a bar. It is not a pricing problem, a sales problem, or a market problem. Clubhouse didn’t need enterprise account executives. It needed a reason to open the app in month three.

Failure two: the price never covered the cost

Airtable’s failure was unit economics, and it is the more painful kind, because the product worked.

It sells a spreadsheet that behaves like a database, which lets non-engineers build small internal tools without writing code. People understood it, adopted it, and built real workflows in it. Activation was not the problem. What follows is an interpretation, not a disclosure, because Airtable is private and publishes none of these numbers: the company appears to have been acquiring and supporting a large number of small accounts whose revenue could not pay for the attention they took.

Why Airtable cut staff twice

The correction was brutal and public. Airtable cut 254 people in December 2022, a fifth of the company, then another 237 nine months later in September 2023, reported as 27% of what remained.⁵⁶ The September round cut hardest into the teams serving small clients, and CEO Howie Liu’s new frame was explicit: consistently win customers with “million-dollar-plus spend rates, versus supporting lots of little ten-thousand-dollar customers from a sales touch standpoint.”⁵ Airtable later retired Plus, its cheapest paid tier at $10 per seat, and folded the old Pro tier into a more expensive Team plan. That left Free, Team, Business, and Enterprise Scale.⁶

Liu’s own stated reason was different: he described a company swept up in the post-COVID hiring frenzy that needed to operate more maturely.⁵ Every SaaS company cut staff in that window, so overhiring and the rate shock explain the headcount on their own. What supports the unit-economics reading is the shape of the cuts, not their size. A pure cost correction trims evenly and keeps selling to everyone. Airtable cut hardest into the teams serving small accounts, then removed the cheapest paid tier those accounts lived on. That is how a company behaves when it has decided the small-account business itself is the problem.

Standard PLG advice actively makes this worse. Shorten time to value, widen the free tier, reduce friction, drive more signups: every one of those moves grows the population of accounts you can’t profitably serve.

The fix: raise prices and serve fewer, larger accounts. It means moving the free tier’s ceiling down until it stops absorbing real workloads. It means repricing the middle so that accounts which consume support pay for it. It means accepting that some customers will leave, and that this is the point, not a side effect. And it means telling a board that has been celebrating signup growth for six quarters that signup growth was the problem.

That last part is why teams reach for an onboarding project instead. Activation is the one problem everybody knows how to work on, and there is a tooling industry ready to sell you the work. There is no tooling industry for “stop serving your smallest segment.” Getting pricing and packaging right is the fix here. Onboarding is not.

Failure three: PLG worked and still was not enough

The sales ceiling is the failure nobody calls a failure. That is why it catches teams unprepared. The model works. It just stops working alone, at a ceiling you cannot see from below.

How much of Dropbox’s revenue came from self-serve

Dropbox is the clearest case, and it made the argument itself. Its IPO filing reported that more than 90% of revenue came through self-serve channels, with essentially no traditional sales motion required to produce it.⁷ That is close to the theoretical ideal of product-led growth. The same filing describes complementing that with “a focused outbound sales effort targeted at organizations with existing organic adoption of Dropbox,” noting that nearly all of its largest outbound deals began as smaller self-serve deployments.⁷

A company getting nine-tenths of its revenue without salespeople still concluded it needed salespeople, and pointed them at accounts the product had already won. That’s not a company abandoning PLG. That’s a company discovering that self-serve adoption and enterprise expansion are two different jobs. It’s why land and expand has to be run deliberately. It doesn’t happen just because people like your product.

Does Atlassian have a sales team?

Atlassian shows how long the wrong lesson can survive. Atlassian makes Jira and Confluence, the issue tracker and internal wiki used by a large share of software teams. It reached $102 million in revenue in 2011 with no salespeople at all, and its 2015 F-1 showed R&D running at around 44% of revenue: the product was the sales team, and the spending said so.⁸ Five months after the December 2015 IPO, Bloomberg was still running the headline “This $5 Billion Software Company Has No Sales Staff.”⁹ Atlassian now runs a conventional enterprise sales organization, and the transition was never announced. Founders benchmarked themselves against a version of Atlassian that had already stopped existing.

The fix: add sales on top of proven adoption, aimed by usage signals. None of this means PLG was a mistake. Dropbox and Atlassian both built enormous businesses on self-serve first, and adding sales on top of proven bottom-up adoption is a different and far cheaper motion than selling cold. The mistake is reaching the ceiling with no signal for which accounts to call. A product-qualified lead is an account whose own usage says it is ready to buy. A marketing-qualified lead only says somebody downloaded something. That difference is what aims the sales team at accounts the product has already warmed up. Without PQLs your adoption advantage disappears, and your sales team ends up cold-calling like everyone else’s.

Every common PLG mistake is one of the three in disguise

The commonly cited PLG mistakes are real. Each one is a symptom of a specific failure, and naming which is how you avoid fixing the wrong thing.

Every one of those is a failure of product-led growth, fixable inside the model. The next section is about the other kind.

Product-led vs sales-led: when the product can’t do the selling

Aaron Kassover founded AgentMethods, a marketing tool for insurance agents, and wanted the low-touch motion most founders want. He puts the reason plainly: founders prefer to buy that way themselves, while “most markets prefer high-touch.” Switching to a high-touch motion was what got him to product-market fit.¹⁰ The test below asks what your buyers can do alone, not what you would prefer them to do, and answering it kindly is the expensive mistake.

The test is the stranger from the top of this chapter, run for real: a person outside your company, a fresh account, nobody helping. Watch whether they reach something they would miss if you took it away. Two things decide whether they can. The product has to be comprehensible without explanation, because a product that needs a person to make sense of it has already hired one. Self-serve earns less per customer, so it needs volume to make up the difference. Where the total number of companies who could ever buy is small, no amount of product quality closes the arithmetic of failure two, and the market itself has decided you are running a sales motion.

Five conditions override a product that otherwise looks like a good fit.

Negotiation is the override that gets misread, because it looks identical to failure three and points the opposite way. In failure three, procurement arrives on top of adoption the product already won, which is why the answer is to add sales at the ceiling. Here it arrives in front of every deal, before anyone has used anything, which is why there is no ceiling to reach. The tell is whether you have accounts that adopted without talking to you at all.

Two more cases belong with these: launching before product-market fit, and running the model on the side of a sales organization with no dedicated team and no retired competing process. Those are not failures of product-led growth. They are failures to attempt it.

A middling answer is not a verdict either. Where some of this holds and some does not, the move is to test the model somewhere contained, in one product line or under a separate brand, rather than betting the company on it or walking away from it.

When should you abandon product-led growth?

Abandon it when the math stops working, not when growth slows. Pre-commit to these tests before you are emotionally invested in the answer, because the moment to decide is never the moment you are deciding.

Airtable’s response to its own crisis was to reprice and narrow rather than to abandon self-serve, and that is usually the right order: exhaust the pricing fix before concluding the model is wrong. What the math actually says is a question the metrics chapter answers properly. None of these tests means anything, though, until you know which of the three failures you are in.

Find out which failure you are in

1. Run the stranger test with an actual stranger. Not a colleague, not a friendly customer. A fresh account, nobody helping, and a note of where they stop. It takes an afternoon and it settles whether any of the three failures above is the one you have.
2. Chart the cohort. Plot weekly return rate through Week 8 for signups from 90 days ago. You’re looking for whether the curve flattens or keeps falling, not for the Day 1 drop. If it trends to zero, you’re in failure one and every other item here is premature.
3. Calculate payback on your median account, not your best one. CAC plus support plus infrastructure, against what the account actually earns you, for the middle of your paying customers. It is tempting to model this on your largest customers, which flatters the answer.
4. Find the last step your biggest deals completed. Every opportunity above your top self-serve tier, last two quarters. If they cluster at procurement, security, or legal, no product change reaches them and you should be staffing, not designing.
5. Write down which of the three you’re in, and the date. Then set a review 90 days out. The failure mode isn’t usually picking wrong. It’s never picking at all, and spending a year applying all three fixes at once.

Once you know which failure you’re in, the PLG flywheel breaks growth into five stages (Evaluate, Activate, Adopt, Expand, Advocate) and shows which stage to fix first. That’s the next chapter.

Updated August 2026