Duolingo priced the growth its own monetisation was costing it

Almost every company that decelerates blames something outside the building. The category matured. A competitor got aggressive. Consumer spending softened. Duolingo did the opposite and said the slowdown was self-inflicted, which is a much harder sentence to get past a board.

The run it was giving up was genuinely rare. From Q2 2022 to Q2 2025, not one quarter came in below 40% year-over-year DAU growth. Thirteen quarters in a row. Q4 2025 closed at 52.7 million daily active users, up 30% from 40.5 million, with 12.2 million paid subscribers, up 28%. Over the full year the company crossed 50 million DAUs and passed a billion dollars in bookings, and it still put $42.0 million of net income on the board in the fourth quarter alone. Then it guided 2026 DAU growth to roughly 20%.

Cutting your headline growth rate from a 40%-plus run to 20% in guidance is the kind of thing that gets a CMO invited to an uncomfortable meeting. What makes this one worth studying is the reason attached to it. Duolingo pointed at monetisation friction it had deliberately added to lift bookings per user. It chose to make the free experience harder to sit inside indefinitely, knew that would cost it users at the top, and priced that cost into its own forecast rather than discovering it later.

So the question underneath this teardown is whether monetisation friction is a loan taken out against acquisition, repayable with interest, and whether you are allowed to describe it that way in your own reporting. Duolingo could, partly because it was sitting on $1.04 billion in cash and $360.4 million of free cash flow for the year. Honesty about a growth trade is a lot cheaper when the balance sheet absorbs the flinch.

Most people looking at Duolingo's Q2 would conclude the paywall change worked beautifully. Revenue up 18% to $298.5 million. That is a flattering number and it is also several months out of date. Subscription revenue recognised in the quarter was largely sold in prior periods, so deferred revenue keeps unwinding onto the income statement long after live demand has changed direction.

Bookings tell the version that happened this quarter. Up 8%, or 6% stripping out currency, to $289.1 million. Ten points of gap between the two, and that gap is pure arithmetic rather than noise. The full-year guide holds the same wedge in place, roughly 16.3% revenue growth against roughly 10.9% bookings growth. Anyone with a board deck built on revenue would have spent two quarters unaware anything had shifted.

The margin lines are just as slippery. Net income fell from $44.8 million to $33.2 million year over year, adjusted EBITDA margin slid from 31.2% to 25.9%, and in the same release the full-year margin outlook went up to about 26.5%. Both are true. Stock-based compensation running at almost 15% of revenue in 2026 against 13% the year before muddies it further, and gross margin at 72.6% barely moves regardless of what happens to demand, so it tells you nothing about whether the change landed.

My view is that bookings and DAU are the only honest pair here. One is cash committed now, the other is the population that will commit later, and neither can be flattered by timing.

Agree that with your CFO and your board before you ship the paywall change, in writing, naming the metric you expect to fall and the one you expect to hold. Doing it afterwards looks exactly like an excuse, because by then it is one.

58.7 million daily active users. 12.7 million paid subscribers. Round it however you like, the overwhelming majority of people opening the app in Q2 2026 pay nothing, and that is the entire reason this trade was available. The free tier is the acquisition channel. Every bit of paywall friction is friction applied to the top of the funnel, which is why it shows up as a DAU cost rather than a churn cost, and why the cost was worth naming out loud.

The second precondition is distribution that free users generate rather than consume. Duolingo's social accounts have cleared a billion organic impressions in each quarter. When your reach is manufactured by the same population you are about to squeeze, you can model the squeeze. If your growth arrives through performance media, you are not making this trade at all, you are just raising CAC and hoping the LTV catches up.

Then there is the balance sheet, which is the part nobody wants to talk about. $360.4 million of free cash flow in 2025, $1.04 billion of cash at year end, $78.6 million of free cash flow in Q2 2026 at a 26.3% margin, roughly $1.3 billion in cash and short-term investments, and a $400 million repurchase programme with about $71.9 million spent by the start of August. That is what lets a company publish a bill and survive the print.

The fourth condition has nothing to do with money. You need a board that will accept a stated cost rather than demand it be laundered into a mix explanation. You can buy cash. You can build organic reach over years. You cannot buy the willingness to say the number, and with over half of bookings coming from outside the US, the currency noise alone gives most CFOs an easier story to hide behind.

58.7 million daily active users in Q2 2026, up 23% year over year and two percentage points faster than the quarter before it. Against a full-year guide of roughly 20%, that is a company beating its own downgraded expectations two quarters after it published them. Anyone who shorted the slowdown story got a lesson in what happens when a management team guides to the number it can defend rather than the number it hopes for.

Retention did the quiet part. Current user retention hit an all-time high of 84%, up about a point on the prior year, which sounds small until you apply a point of retention to a base that size. And in June they ran a Streak Revival event that brought back 15.4 million learners, nearly 8 million of whom had no active streak at all when it started. That is a reactivation campaign pointed squarely at the cohort that paywall friction had pushed out the door. They broke something, then went and got the people back.

I cannot cleanly credit the friction pullback for the reacceleration and neither can you. Chess crossed seven million DAUs in under a year, new subjects were landing, and the reactivation push ran in the same window. Three things moved at once, which is how these windows always look in real companies rather than in case studies.

What I do trust is the shape of the trade. Paid subscribers grew 17%, a full six points behind the 23% DAU line, while engagement accelerated. A company drifting into weakness sees both lines sag together. Duolingo bought users with monetisation restraint, and the 100 million DAU target for 2028 tells you which currency it thinks is scarcer.

Most teams walk into a paywall test with one number on the wall, and it is revenue. That is the number that lies longest, because deferred bookings keep paying you for months after the acquisition damage is done. So before anyone ships extra friction, I would put four things on a single dashboard and keep them there permanently rather than for the length of a campaign.

The unmonetised share is the one I would argue about hardest in a leadership meeting. Duolingo carries 12.7 million paid subscribers inside 58.7 million daily actives, which means roughly four in five of the people using the product every day pay nothing. With a base shaped like that, friction has enormous headroom and enormous downside, and the ratio itself tells you which lever you are actually pulling. If your unmonetised share is 20% instead of 78%, squeezing harder is mostly just annoying your customers.

Then run a reactivation test sized to matter. Duolingo's June Streak Revival brought back 15.4 million learners, nearly 8 million of whom had no active streak at all when it started. That is a real experiment rather than a lifecycle email, and anything smaller cannot separate signal from noise.

Monetisation friction is a loan against acquisition. It draws instantly and repays slowly, and the only honest way to run it is to tell the board the deceleration number in advance rather than explain it in arrears. My view on Duolingo from here is that Chess and the other subjects, not the paywall, have to carry the run at 100 million DAUs in 2028, and I think they will.