Business

Know the Business

Duolingo is a consumer mobile-software business dressed up as EdTech: a freemium subscription engine where word-of-mouth distribution, a gamified daily habit, and ~9% paid conversion produce 72% gross margins, ~30% Adjusted EBITDA margins, $360M of free cash flow, and zero debt. Daily active engagement is the lever that compounds into paid subscribers; the two most commonly mis-weighted variables are the non-discretionary toll paid to Apple and Google inside cost of revenue, and franchise durability against zero-cost generative-AI substitutes at the lower-skill end of the funnel.

1. How This Business Actually Works

This is a freemium consumer subscription. Anyone can use the app for as long as they like at no cost; about 9% of monthly users pay roughly $84 a year for an ad-free experience and extra features. Subscription is 84% of revenue; the remaining 16% is advertising on the free tier, the Duolingo English Test, and small in-app virtual-good sales. Cost of revenue is overwhelmingly third-party payment processing fees charged by Apple and Google (15–30% of in-app subscription proceeds) plus cloud hosting and a rising AI-inference line. Everything else on the income statement is people — and the largest people line is R&D, not sales and marketing.

The economics tilt unusually far toward the product and the platform tax. FY2025 S&M was only 12% of revenue (including stock comp) because the brand grows by going viral, not by buying impressions; R&D ran 30% of revenue and is best read as the cost of supply, since the "factory" producing supply is engineers, designers, and the AI tooling that now generates 20,500 course units a quarter — more than 10× the 2024 baseline. Pricing power is shallow at the high end: management does not push price aggressively because the freemium funnel works only as long as the paywall feels optional. The two parties with real bargaining power are the consumer (instant churn, free LLM alternatives) and the duopoly distribution platforms.

2. The Playing Field

Duolingo has no clean public peer. Its closest economic substitutes — Babbel, Memrise, Busuu, Rosetta Stone, Preply — are private. The listed comparable set therefore blends three different things: direct consumer EdTech that is shrinking (COUR, CHGG), profitable but structurally different K-12 EdTech (LRN), and large freemium-engagement platforms used as templates for what scaled monetization can look like (SPOT, RBLX). Read the table accordingly: Duolingo is the only name combining 30%+ growth, 70%+ gross margins, positive GAAP operating income, and material FCF generation.

What the peer set reveals: the base rate for "EdTech consumer subscription" is bad (COUR decelerating to single digits, CHGG losing two-thirds of subscribers to free LLMs), and the only listed name with Duolingo-like growth-plus-margin is Roblox — but Roblox burns cash on stock comp at 23% of revenue and runs negative GAAP operating income. Stride is profitable but answers to school-district procurement cycles, not consumer App Store rankings. Spotify is the relevant template for what scaled freemium subscription "looks like" at maturity (~10% growth, low-teens operating margin, high-teens FCF margin); the bull case for Duolingo is that better gross margins and a more defensible niche let it stay above Spotify's terminal economics. The bear case sits one row up — Chegg shows what happens when a thin-moat consumer EdTech name meets free generative AI.

3. Is This Business Cyclical?

Duolingo is secular, not cyclical. There is no inventory cycle, no rate-sensitive customer, no capex pulse, no commodity input. Revenue compounds at 35–45% off a flywheel — more users → more data → better product → more users → higher paid conversion. The thing that looks cyclical in the history is the post-IPO consumer-tech reset (2022) and the COVID demand pull (2020–21), both driven by the equity-market environment rather than the underlying business.

What is sensitive to outside conditions is mix and timing, not direction. EM ARPU moves with FX and local consumer discretionary income (the average revenue per user has trended down as the install base shifts toward lower-ARPU geographies). Bookings, which are collected upfront, show up before revenue does — that is why the stock sold off in February 2026 on a soft FY26 bookings guide even though revenue growth was intact. App-store policy can shift gross margin by hundreds of basis points overnight if EU DMA-style web-billing relief expands. None of these is a classical macro cycle; they are mostly platform and regulatory exposures inside an otherwise secular trajectory.

The structural risk that would create a real cycle for this business is AI substitution. The Chegg base rate — subscriber count collapsing from ~8M to ~3M as ChatGPT/Gemini replaced paid homework help — is the cautionary template, even if Duolingo's defenses (gamification, brand, social streaks, parent–child trust signals) are materially stronger.

4. The Metrics That Actually Matter

Read this business through five numbers. Revenue growth is a lagging blend of all of them and tells you less than people think.

Two further notes belong here. S&M intensity (12% of revenue including SBC) is the canary for the word-of-mouth flywheel — if Duolingo had to ramp S&M to keep DAUs growing, the operating leverage story would unwind quickly. And SBC at 13% of revenue, guided to ~15% in FY26, is the largest non-cash cost in the P&L; treating Adjusted EBITDA at face value (29.5% margin) without subtracting SBC overstates true cash earnings by half. The cash-comp-adjusted FCF margin is closer to 22%, not 35%.

5. What Is This Business Worth?

Duolingo is best valued as one economic engine, not a sum of parts. The Duolingo English Test is real (6,100+ accepting institutions) but is bundled inside "Other" revenue, ran roughly flat at $11M in Q1 FY26, and is not a separately listed asset. Math, Music, and Chess are early-stage product extensions that share the same app, the same data flywheel, and the same paying user. There is no listed subsidiary, no holdco discount, no regulated/unregulated mix to disaggregate. Force-fitting SOTP here would manufacture false precision.

The right lens is price-to-free-cash-flow with a growth-runway overlay — the same lens you would apply to Spotify or any scaled consumer-software cash compounder — adjusted for stock-based compensation because SBC is a real cost.

On headline EV/FCF, Duolingo screens as cheap relative to comparable engagement platforms. On a stock-comp-honest basis (subtracting ~$140M of annual SBC from cash flow), the multiple is roughly in line with where Spotify and Roblox trade. A premium versus those names is supportable only if (a) DAU growth stays above 20% into 2027–28, (b) per-unit AI compute costs continue to fall, and (c) the AI-substitution narrative does not start showing up in retention curves. A discount is warranted if any one of those breaks.

6. What I'd Tell a Young Analyst

Don't be impressed by GAAP net income. The FY2025 figure of $414M includes a $256.7M one-time tax benefit from releasing the valuation allowance on deferred tax assets. Underlying GAAP operating income was $136M (13% margin). Adjusted EBITDA was $306M (29.5% margin). Free cash flow was $360M (35% margin). After subtracting the $137M of real SBC cost, owner earnings are closer to $220M — still excellent, but only about half of what the headline net income line suggests.

Watch bookings before you watch revenue. Annual subs collect cash upfront and are recognized monthly, so revenue lags reality by 6–12 months. The February 2026 stock drop happened on a soft FY26 bookings guide while the revenue print was clean. If you wait for revenue, the multiple has already moved.

Watch DAU growth and S&M intensity together. DAU growth above 20% with S&M at 12% of revenue (roughly today) is the flywheel working. DAU growth at 12% and S&M at 18% would mean the flywheel has stopped compounding and management is buying users to cover the gap — that is the moment the multiple compresses.

The gross-margin path is a policy choice, not an outcome. Management is choosing to let gross margin drift from 72% to 69% to fund AI-feature differentiation. If gross margin breaks below 69% for non-AI reasons (mix shift to advertising, app-store fee changes), that is a real signal. Inside the guidance, it is not.

The real moat is gamification and brand, not curriculum. Anyone with a credit card and an LLM API key can teach Spanish. What is hard to replicate is a daily streak that 15M people have held for over a year, a brand that has become an internet meme, and a 2-billion-exercise-per-day data flywheel. The day to worry is the day those streaks stop showing up in App Annie engagement data — that is the leading indicator of AI substitution, and it leads everything else by a quarter or two.

Three things would change the thesis. (1) DAU growth dropping below 15% for two consecutive quarters. (2) Paid % of MAU flattening or falling year-over-year. (3) Gross margin slipping below 68% with management blaming "mix" rather than "AI investment." Any one of those, in isolation, is noise. Two in the same quarter is when the de-rating playbook typically starts to run.