Why Product-Led Growth Is Harder Than Everyone Pretends

San Francisco · May 7, 2026

The pitch deck had a slide called "The PLG Flywheel." It was beautiful. Arrows going in a circle, each one labeled with something aspirational: Acquire, Activate, Engage, Monetize, Expand. In the center, in bold Helvetica: "The Product Sells Itself."

I was sitting in a board room in 2022, at a company called Corral — the one that would be dead within eighteen months — watching our CEO present this slide to investors. He believed it. They believed it. I think even I believed it, for about six weeks, until we launched the free tier and learned what product-led growth actually looks like when nobody's reading the case study about it afterward. (I later wrote about what happened when the freemium model broke at my current company — different company, same lesson.)

Here's what it looked like: 14,000 free signups in the first month. A activation rate of 11%. A free-to-paid conversion rate of 0.8%. And a support ticket volume that tripled overnight because free users have the same number of questions as paid users but zero tolerance for waiting.

"The funnel will optimize," the CEO said. "Give it a quarter."

We gave it two. The activation rate climbed to 19%. The conversion rate crept up to 1.4%. The support team was drowning. And the board, which had funded us based on the flywheel slide, was asking questions that started with "When" instead of "If."

The problem with PLG — and I say this as someone who has now attempted it at three companies — is that the theory is elegant and the execution is brutal. The theory says: build a product so good that people adopt it themselves, tell their colleagues, and eventually pay you. Slack did it. Dropbox did it. Figma did it. Therefore, the logic goes, you can do it too.

But here's what gets left out of the case studies. Slack had a product that replaced email for teams. The value was obvious within five minutes. Dropbox synced your files — something people needed every single day. Figma replaced a $600/year desktop application with a free browser-based tool. These products didn't just sell themselves. They solved problems that were so painful and so universal that adoption was almost gravitational.

Most B2B SaaS products are not like this. Most B2B SaaS products are useful, not essential. They make a workflow 30% faster or a report 20% more accurate. That's enough to justify a purchase if a salesperson walks you through it and your boss approves the budget. It's not enough to make someone sign up on a Saturday afternoon and start inviting their team.

At Corral, we built a product analytics tool. Good product. Clean UI. Accurate data. But product analytics is a considered purchase. You don't wake up one morning and decide to implement event tracking because it sounds fun. You implement it because someone at your company made a business case and got buy-in from engineering. That's a sales motion, not a product motion, and no amount of flywheel diagrams changes the underlying buyer behavior.

The theory says the product sells itself. In practice, the product sits there while your support team explains it to people who signed up because it was free.

When I joined my current company, the CEO wanted PLG. She'd read the OpenView reports. She'd seen the multiples that PLG companies commanded in 2021 (and she wasn't alone — the decision to kill the sales team came from the same PLG conviction) (before the market corrected and those multiples evaporated, but that's a different essay). She wanted a free tier, a self-serve motion, and a "land and expand" strategy.

I told her what I'd learned at Corral. I told her about the 0.8% conversion rate and the support ticket avalanche and the eighteen months we spent optimizing a funnel that was never going to work because the product wasn't built for self-serve discovery.

"Ours is different," she said.

"Every CEO says that," I said.

"Am I wrong?"

I thought about it. Actually thought about it, instead of just reacting from scar tissue. Our product was a workflow automation platform. Users could set up automations — connect apps, trigger actions, build sequences — without code. The time to first value was about twelve minutes. You could sign up, connect two apps, build an automation, and see it run. Twelve minutes. That's PLG territory.

"You might not be wrong," I said. "But we need to do it differently than Corral did."

The first thing we did differently was define what "activation" actually meant. At Corral, activation was "completed onboarding." That's a vanity metric. It measures whether someone clicked through five screens, not whether they received value.

For our product, we defined activation as: created at least one automation that ran successfully at least three times. That's a behavior that correlates with retention — our data showed that users who hit this threshold retained at 74% after ninety days, versus 12% for users who signed up but never built a working automation.

Our initial activation rate, by this definition, was 23%. That means 77% of free signups never built a working automation. They signed up, poked around, maybe started creating something, and left. Seventy-seven percent.

"That's normal for PLG," our advisor said. He'd been VP of Growth at a company that's held up as a PLG success story. "Our activation rate was 18% when we started. We got it to 34% over two years."

Two years. That's what nobody tells you. PLG is a multi-year optimization project. You're not launching a free tier and watching the flywheel spin. You're launching a free tier and then spending 24 months grinding on onboarding flows, activation emails, in-app nudges, and tooltip placement. It's the least glamorous work in growth.

We rewrote the onboarding four times in six months. I'm not exaggerating. Four complete rebuilds.

Version one: a five-step wizard that walked users through connecting their first app. Activation rate: 23%. No improvement. Users completed the wizard and then stared at a blank screen, not knowing what to build.

Version two: we added templates. Pre-built automations that users could install with one click and then customize. Activation rate: 29%. Better. But the templates were generic, and users who installed them didn't understand what they did well enough to build their own.

Version three: we replaced the wizard with a "goal selector." Instead of "Connect your first app," the first screen asked "What do you want to automate?" with options like "Send Slack notifications from new form submissions" or "Sync new CRM contacts to my email tool." Each goal led to a guided setup flow. Activation rate: 36%.

Version four — the one that's still live — combined the goal selector with an AI assistant that watched what users were doing and offered contextual suggestions. "It looks like you're trying to connect Salesforce. Would you like to see the three most popular Salesforce automations?" Activation rate: 41%.

From 23% to 41% took eight months and approximately $340,000 in engineering time. It required a dedicated team of two engineers and one designer working on nothing but onboarding. The head of product hated it because those engineers could have been building features. The CEO loved it because the activation curve was going up. I was somewhere in between, staring at the numbers and wondering if 41% was good enough or if we needed 50% to make the unit economics work.

From 23% to 41% activation took eight months and $340,000 in engineering time. The blog posts about PLG don't mention numbers like these.

Activation is only half the problem. The other half is conversion — getting activated free users to pay you money.

Our free tier allowed five automations and 1,000 tasks per month. The paid tier started at $29/month for unlimited automations and 10,000 tasks. We set the free limits based on what we thought was generous enough to demonstrate value but restrictive enough to create upgrade pressure.

We were wrong about the limits. Sixty-two percent of activated free users never hit the task ceiling. They were happily running three or four automations within the free limits, getting real value, and never paying us anything. They were the best kind of user — engaged, active, retained — and they were generating zero revenue.

"Lower the free limits," the CFO said.

"If we lower the limits, activation will drop," I said. "Users won't hit the aha moment fast enough."

"If they don't convert, who cares about the aha moment?"

She had a point. But so did I. This is the fundamental tension of PLG: the free tier needs to be good enough to create engagement and bad enough to create upgrade motivation. Too generous and you're running a charity. Too restrictive and you're running a trial, not a freemium product, and you should just call it a trial.

We compromised. We kept the automation limit at five but dropped the task ceiling from 1,000 to 500. We also added a feature gate: advanced conditions (if/then logic within automations) required a paid plan. The theory was that users would build simple automations for free, realize they needed conditional logic, and upgrade.

Free-to-paid conversion went from 3.1% to 4.8%. The feature gate was the biggest driver — conditional logic turned out to be the thing that separated casual users from serious ones. Users who tried to add a condition and hit the paywall converted at 22%.

Here's where we are now, two years into the PLG motion. The numbers I would have put on a pitch deck slide:

Free signups: 8,200/month. Activation rate: 41%. Free-to-paid conversion: 4.8%. Average time from signup to first payment: 34 days. Paid customer retention at twelve months: 88%. Net revenue retention: 112%.

Those are good numbers. PLG is working. The flywheel is — to use a word I hate — spinning.

Here are the numbers I wouldn't put on the slide:

Total investment in PLG over two years: approximately $1.4 million in engineering, design, and analytics time. Onboarding rewrites: four. A/B tests run on the free-to-paid conversion flow: 67. Tests that produced statistically significant results: 11. Tests where the "winning" variant turned out to be a local maximum that we later had to unwind: 3.

We employ two full-time people whose entire job is optimizing the free-to-paid funnel. Their Slack channel, #plg-experiments, has 4,300 messages. If you scroll through it, you'll find more failed hypotheses than successful ones. You'll find arguments about statistical significance thresholds. You'll find a thread from October 2024 where I wrote, at 11 p.m., "I genuinely don't know if PLG is working or if we're just getting better at measuring our own delusion."

Product-led growth works. I believe that now, with evidence, in a way I didn't when we started. But it works the way losing weight works: slowly, painfully, with constant effort, and with no single moment where you can point and say "that's when it happened." It's a thousand small optimizations compounding over years.

If someone shows you a flywheel diagram and tells you the product sells itself, ask them three questions: What's your activation rate? How many times did you rewrite your onboarding? And how much did it cost?

If they can answer all three, they've done the work. If they can't, they're selling you a slide deck.