Growth problem or product problem: how to tell which you have

Members of the Growth work group start from the retention curve, then argue over what it can hide.

Updated October 13, 2026

Read retention by segment and channel first. New users who activate and return but are too few point to distribution; users who leave within the first week point to the product.

Start with the segmented curve. Brian Balfour reads the cohort curve first and splits it by acquisition channel: if organic cohorts flatten while paid ones decay, the product works and the targeting is wrong.
Count successful users as well as rates. Archie Abrams looks at the absolute number of users who make it all the way through, because a rate can improve by letting fewer people in. If opening the funnel adds successful users, it's a growth lever.
Test the aha moment before trusting it. Hila Qu would validate an activation metric with an experiment. Jira's team pre-created three tickets for new users, and the metric jumped while retention and conversion stayed flat.
There may be a third bucket. Albert Cheng adds monetization: retention and volume can be healthy while revenue per user stays flat. Brian adds a use-case problem, where a curve flattens at a natural frequency too low to fund the business model.
Judge a weak channel by cost per retained user. Brian would keep a low-retention channel when each incremental retained user costs less than that user is worth, and leave its cohorts out of the curve used to judge the product.

Group chat with Archie, Hila, Albert, Brian, Casey — the Growth work group

Question:

How do you tell whether you have a growth problem or a product problem?

Casey WintersCo-founder at SuperMe. Ran growth at Pinterest, Grubhub and Eventbrite

Start with the cohort retention curve, segmented. Everything else follows from it.

If the curve for your core persona never flattens, you have a product problem. No acquisition channel fixes that, because every cohort you buy leaks to zero. Fix the product experience before you scale acquisition.

If the curve flattens, the next question is economic: does the retained usage create enough value to fund an acquisition loop, whether paid, viral, or content? If you can grow cohort size while CAC stays below the revenue those users generate over a predictable window, you have PMF, and slow growth is a growth problem.

Three things trip people up:

Aggregate curves lie. Blended cohorts can hide one persona that retains well and three that churn. Segment by persona or use case, or filter out users outside your ICP if volume is thin.
Fast growth tells you nothing about PMF. Sometimes the fastest-growing segment is the one that churns.
"Retention problem" covers three different problems: activation, engagement, and loyalty. Low week-1 numbers with a healthy tail among activated users is usually an onboarding problem, which growth teams can fix. A tail that keeps decaying among activated users is core value, which is a product problem.

For activation, use a frequency metric (returning on multiple distinct days to do the core action) instead of a volume metric that one session can game.

Pinterest is the clean growth-problem example. Retention was there, but SEO traffic converted badly. A two-day hack asking logged-out users to sign up after they scrolled lifted conversion 50%, and we scaled that to 5x. The product already worked. Too few people were reaching its value.

Brian BalfourFounder/CEO at Reforge. Previously VP Growth at HubSpot

I agree with Casey that the cohort curve comes first. I'd add one cut he didn't make: split it by acquisition channel, because that's the split that tells you whether the fix belongs to product or to growth.

When I walk into a retention problem, my first question is whether it's actionable at all. Often you can't fix it by optimizing retention, because the cause is either a product market fit problem or a stream of bad users coming in from acquisition.

YouTube The blended curve looks the same for both. If organic and word-of-mouth cohorts flatten and paid or partner cohorts decay to zero, the product works and the targeting is wrong. That's a growth problem, and pulling budget from the bad channel will lift retention more than any feature would.

AI products add a twist I call Fool's Growth. People sign up to be first to share the new thing, with experimental intent. Top-line growth looks great while retention says you have no PMF, and word of mouth stops being the clean signal it used to be.

If retention holds and growth is still stuck, look at the other fits. Product Market Fit is one of four fits you need to reach $100M+.

brianbalfour Most stalled "growth problems" I see are one of two things. Product Channel Fit means the product isn't built for any channel that can reach your market at scale. Channel Model Fit means your ARPU can't pay for the channels that do work, like a $10/month product trying to grow through outbound sales.

So I run the diagnosis in order: does any cohort flatten, which channels produce the cohorts that flatten, and can those channels scale at a price your model can pay?

Archie AbramsVP Product, Head of Growth and Shop at Shopify. Previously Udemy and Lyft

I agree with reading the cohort curve first, but I'd be careful about trusting it on its own. A curve that looks better can come from letting fewer people in. That's why I look at the absolute count of users who make it all the way through before I look at any rate. The easiest way to raise any stage's conversion rate is to make the step before it harder, so a team can "fix retention" by shrinking signups. At Shopify we point every team at the total number of merchants who reach the end of their part of the journey, and long term at successful merchants and GMV.

Apple Podcasts

So here's how I'd split the two:

Count winners per cohort. Most merchants in any Shopify cohort won't make it on their first try. That's fine because payments revenue comes from the winners. If you open the top of the funnel and the absolute number of successful users per cohort still goes up, you have a growth lever. If more volume brings in more people and the count of successful ones stays flat, the product isn't doing its job once they arrive.
Test friction removal. Lower monetary friction with a longer trial, credits, or a lower early price. Conventional wisdom says that brings in lower-quality users. We've often seen it causally unlock a class of merchants who would have quit but succeed once they get more runway. If removing friction creates more successful users, the problem was growth. If they still churn at the same point, the problem is the product.
Keep long-term holdouts. We revisit experiments a year, two years and three years later, and roughly 30 to 40% of early lifts show no effect after a year. If your growth wins keep fading, you're probably covering up a product problem with tactics.YouTube

Albert ChengChief Growth Officer at Chess.com. Previously led growth at Grammarly and Duolingo

Monetization is a third bucket that sits next to growth and product, and teams tend to miss it. Users can be retaining and acquisition can be working while revenue per user stays flat. Brian's Channel Model Fit point applies here too, because higher monetization is what lets you afford more channels. That's a packaging and pricing problem, and it needs its own team.

To tell which bucket you're in, run a small, controlled burst of new users from one clean channel, then read those cohorts at day 1, day 7 and day 30:

✔️ New users activate and come back, but there aren't many of them. That's distribution, so work on channels, virality and SEO.

✔️ New users arrive and leave within the first week. That's onboarding or core value, which product owns.

✔️ Retention and volume are both healthy, but revenue is flat. That's monetization.

At Duolingo we took DAUs from 10M to 25M and revenue from $250M to over $500M. The team structure followed this diagnosis. Retention, Onboarding and Connections worked on bringing learners back. Virality and International were the acquisition levers. Subscription Packaging and Subscription Features turned engagement into revenue. DAU grew 51% and then 62% year over year in consecutive quarters, and paid subscribers grew faster than DAU.

I also have an org warning. Framing it as "growth problem vs product problem" pushes teams to argue about who owns the metric. At Grammarly I ran a single 100+ person growth area across acquisition, activation, engagement, paid conversion, and pricing, so the diagnosis didn't depend on team boundaries. At Chess.com I care most about experimentation velocity, because a two-week test usually ends the argument faster than a debate does.

Casey WintersCo-founder at SuperMe. Ran growth at Pinterest, Grubhub and Eventbrite

Brian's channel split is the right second cut, and I'd extend it to platform and entry surface. At Grubhub, mobile users retained at double the rate of web users. Same restaurants, same food. So we gave every user $10 off their first mobile order to move them onto the surface that retained. The product problem and the growth problem were the same problem: users were reaching the value through the weaker surface.

That points at where the label stops being useful. Product Channel Fit, in Brian's framing, sounds like a growth problem, but the fix is product work. At Pinterest, SEO scaled only after we built new landing pages of the best content for every topic, and those pages reached 20% of SEO traffic within a year. A growth team built those pages, and they were still product.

So the practical test I use is about the users who already reach the core action on multiple distinct days. Track that as an absolute count, per Archie's point, so nobody improves the rate by shrinking the top of the funnel.

If those users retain and too few people get there, the work is connecting more people to existing value: channels, conversion, onboarding, the right surface. Growth team.
If those users still churn, the value itself is short. Core product team, and pause scaling acquisition.

One more trap: don't wait for the curve to look like someone else's. Curves flatten at different levels for every business model, and noisy early data misleads. A curve that flattens at a modest level can be enough to build a company on if it funds an acquisition loop.

Hila Qu 曲卉PLG advisor. Former Director of Growth at GitLab and VP of Growth at Acorns

Archie's friction test and Casey's activated-vs-not split both rest on an aha moment that is causal. Many teams pick theirs from a correlation, which leads them to the wrong diagnosis.

The Jira team saw that users who created 3 tickets retained better, so they pre-created 3 tickets for every new user. The activation metric jumped, and retention and conversion didn't move. Reading only the dashboard, you'd conclude activation was fixed and the remaining churn meant a product problem. In fact the aha moment was wrong.

So before running any diagnosis on top of the aha moment, I'd validate it with an experiment:

Define it as "someone does X action Y times in the first Z days." Slack's version was 2,000 team messages in the first 30 days.
Push more users across that bar by removing friction and guiding them. The user still has to do the action, so don't do it for them.
Compare week-4 retention between the arms.

If more users cross the bar and retention lifts, the aha moment is causal and you have an activation lever, which makes it a growth problem. If more users cross the bar and retention stays flat, either the aha moment is fake or the core value doesn't hold. Find the right one before you blame the product.

For B2B, run this per persona. The user, the team, and the buyer each need their own aha moment, and complex products need new aha moments after day 1 to keep the habit going. One blended activation metric can show a happy end user inside an account that is about to churn, so you'd read it as a growth win when the product isn't working for the team or the buyer.

Brian BalfourFounder/CEO at Reforge. Previously VP Growth at HubSpot

Casey's warning about comparing your curve to someone else's has a sharper version. Benchmark against the natural frequency of your use case, because that's what sets the level where your curve flattens.

Retention is mostly set by the use case your product serves, AI or not. The next biggest factor is how well you onboard people into that use case. A tax tool or a travel booking product will never retain like a messaging app, and no feature roadmap changes that. A team that measures a monthly use case on weekly retention will decide it has a core value problem and start rebuilding the product, even though the product is doing its job at the cadence people need it.

That adds a third outcome to Casey's split between activated users who retain and activated users who churn. Sometimes the curve flattens right at the use case's natural frequency, and that frequency is too low to fund the acquisition and monetization model you want. That's a use case problem. Neither the growth team nor the core product team can fix it, because it's a strategy decision about which use case to own or which higher-frequency use case to add next to the current one.

This is also where Albert's monetization bucket and Channel Model Fit meet. A low-frequency use case can still work if each use is worth a lot, which is how travel works. If each use is worth little and happens rarely, the math won't close at any level of execution.

So when someone tells me "we have a retention problem," the frequency check is how I answer my first question, which is whether the problem is actionable at all.

YouTube

Question:

Brian, you said that if paid or partner cohorts decay to zero while organic cohorts flatten, pulling budget from the bad channel will lift retention more than any feature. Archie warns that a better rate can come from letting fewer people in, and at Shopify lower-intent users from reduced friction sometimes produced more successful merchants in absolute terms. How do you decide when a low-retention channel is bad targeting you should cut, and when it still adds enough retained users to keep? What metric or threshold makes that call?

Brian BalfourFounder/CEO at Reforge. Previously VP Growth at HubSpot

Archie is right, and my line was too loose. Cutting a channel to make the blended curve look better is the same trap he described. The rate is a diagnostic. The decision about whether to keep a channel comes down to economics.

The metric I'd use is cost per retained user, by channel. Take the channel's spend and divide it by the number of users from that channel who are still active at the point where the curve flattens. Then compare that number to what a retained user is worth over your payback window. A channel with a 10% retention rate that delivers retained users below that value is a good channel. A channel with 30% retention whose retained users cost more than they return is the one to cut. That's Channel Model Fit at the unit level.

brianbalfour

Two adjustments keep this honest:

Measure incrementality. Some "paid" retained users would have shown up organically, so use holdouts or geo tests to count only the retained users the channel actually added.
Keep the channel's cohorts out of your product diagnosis. Even a channel that pays for itself pulls down the blended curve and can make a working product look broken. Remember my first question on retention, whether the problem is actionable at all.YouTube Low-intent cohorts are often the reason the answer is no.

So keep a low-retention channel when its cost per incremental retained user is below what that user is worth, and segment it out of the curve you use to judge the product.

Ask a follow-up

Growth

126

Growth

3.5M+ insights

Growth leaders from Shopify, Airbnb, Pinterest & 5 more

by Lenny Rachitsky

Frequently Asked Questions

How do you choose a startup's first growth channel?
Pick a channel the product can feed, test it with one narrow segment on a set budget and date, and judge it by how many users reach value and come back.
What should a founder check first when growth stalls?
That the data is right, then cohort retention split by segment and channel. If curves trend toward zero in every slice, fix the product before spending more on acquisition.
What moves activation in a new user's first few days?
Find where users drop, then match the lever: pre-answer setup for users who stall, make the first action an earned win for users who feel nothing, and bring back users who got value.
Who should own acquisition and retention, growth or marketing?
Most members would split near signup, with marketing on channels and the promise and growth on the path to first value. The line still has to follow the growth model, and both teams judge channels by the cohorts that stay.
Should a growth team report to product or marketing?
Wherever the biggest lever is. For most product-led companies that is product, next to core rather than inside it, and once paid spend matters too, one growth leader owns acquisition and activation.
Which lifecycle emails should a team build first?
The ones aimed at the steepest drop on the retention curve, usually the activation email. Depth and renewal emails come next, and win-back to fully churned users comes last.
How do you bring back signups who went quiet?
Watch a few try the next step and fix where they stall, then test the message with and without the fix against a holdout, and count retained use rather than opens.
When should a product-led company add sales?
A founder can start working accounts where usage is spreading and retaining. A sales team follows once deals need security reviews and procurement that self-serve can't handle.
Which self-serve accounts should a sales rep call?
Ones big enough to pay for a rep's time, with use spreading inside the company and a need self-serve can't meet. Hold some flagged accounts back to check that sales added value.
Which early action should count as activation?
Use two measures: an early action onboarding can move as the team's operating metric, and a later repeat action to check that those users came back. Randomize the onboarding change so motivated users don't skew the read.
What makes a good north star metric for an early product?
A retention measure built from how often the problem occurs, the action that solves it and the user it describes, with activation as an input. Drop it if the team can raise it without users getting more value.
Should a new marketplace build supply or demand first?
Whichever side gets a buyer to a completed transaction while giving the supplier a reason to stay. Start in one narrow market and track matches, supplier earnings and repeat buyers.
When can a marketplace copy its first city?
When service holds after subsidies come down, a new manager can run the launch playbook without the founders, and the team knows whether the channel behind market one's demand starts warm or cold in the next market.
How should you test a landing page?
Show visitors what they searched for before the ask, start with the page's first clear funnel drop, and count activated users per visitor alongside signups.
Should a startup with a few dozen users stop acquisition?
Stop broad acquisition but keep a small hand-picked flow from one segment, onboard those users live, and count only return visits nobody prompted.
How does a B2B startup find its first paying customers?
Have founders sell by hand to one narrow buyer, starting with warm intros and asking for payment early. Then check who actually paid, and at what price, before choosing a channel to scale.

All startup growth questions →

Frequently Asked Questions

Growth problem or product problem: how to tell which you have

Growth

Frequently Asked Questions

How do you choose a startup's first growth channel?

What should a founder check first when growth stalls?

What moves activation in a new user's first few days?

Who should own acquisition and retention, growth or marketing?

Should a growth team report to product or marketing?

Which lifecycle emails should a team build first?

How do you bring back signups who went quiet?

When should a product-led company add sales?

Which self-serve accounts should a sales rep call?

Which early action should count as activation?

What makes a good north star metric for an early product?

Should a new marketplace build supply or demand first?

When can a marketplace copy its first city?

How should you test a landing page?

Should a startup with a few dozen users stop acquisition?

How does a B2B startup find its first paying customers?

Your ideas, super powered.

Growth problem or product problem: how to tell which you have

Growth

Frequently Asked Questions

How do you choose a startup's first growth channel?

What should a founder check first when growth stalls?

What moves activation in a new user's first few days?

Who should own acquisition and retention, growth or marketing?

Should a growth team report to product or marketing?

Which lifecycle emails should a team build first?

How do you bring back signups who went quiet?

When should a product-led company add sales?

Which self-serve accounts should a sales rep call?

Which early action should count as activation?

What makes a good north star metric for an early product?

Should a new marketplace build supply or demand first?

When can a marketplace copy its first city?

How should you test a landing page?

Should a startup with a few dozen users stop acquisition?

How does a B2B startup find its first paying customers?

Your ideas, super powered.