Business

How to Read Duolingo

Duolingo is a consumer subscription app dressed as an edtech company: a freemium mobile product that converts about 9 of every 100 monthly active users into paying subscribers, then monetizes the rest with programmatic ads. The economic engine is operating leverage on a fixed content and engineering base — not content, brand, or AI per se — and the metric that matters above all others is the DAU/MAU ratio, because engagement is what converts to subs and compounds word-of-mouth growth. The market is likely overestimating near-term AI threat and underestimating how rare this cost structure is: 72% gross margins, 35% FCF margins, over $1B in net cash, and zero reliance on paid acquisition to grow users.

1. How This Business Actually Works

The revenue engine runs on two rails: a paid subscription (Super Duolingo / Duolingo Max) and ads on free-user sessions, with a small third rail from the Duolingo English Test and in-app purchases. In FY2025, subscriptions were 84% of revenue and growing 44%; "Other" (ads + DET + IAP) was 16% and growing 17%.

The cost structure is what makes this work. Cost of revenue is about 28% of revenue and consists almost entirely of app-store take-rates, hosting, and a rising sliver of generative-AI inference cost. R&D is the biggest expense at 30% of revenue — this is a software shop, not a content shop. Sales and marketing is just 12% of revenue, which is the single most telling number on the page: consumer-subscription peers routinely spend 40 to 60 percent.

Where incremental profit actually comes from: every marginal MAU costs almost nothing to serve — hosting is cheap, content is a fixed investment — and roughly 9% of them eventually pay ~$70–85/year gross. The remaining 91% see ads that monetize at a small but non-zero CPM times billions of daily lessons. This is why revenue grew 39% while operating income more than doubled — classic fixed-cost leverage on a viral distribution model. Bargaining power is asymmetric: Apple and Google are the only meaningful counterparties, and they extract roughly 15 to 30 percent of gross subscription dollars depending on tier and tenure. That is the single largest structural margin drag, and it will not change on its own.

2. The Playing Field

Duolingo is the only profitable, growing, cash-generating public edtech at scale. The peer set is instructive by contrast: Chegg is being rerouted out of business by generative AI, Coursera is stuck in enterprise/consumer limbo with no operating profit, Nerdy shrinks every year, and Udemy barely scrapes breakeven. The "competitive landscape" narrative you will hear — Babbel, Busuu, Rosetta Stone — is mostly private companies at a fraction of the scale. Real competition for user time is YouTube, TikTok, and mobile games.

What the peer set reveals: scale is necessary but not sufficient in this industry. Udemy has roughly Duolingo's revenue and is barely profitable because its unit economics are built on third-party instructors (rev share), paid marketing, and enterprise sales. Coursera has the same revenue as Duolingo and loses money because degrees-and-certificates require partner institutions (more rev share) and enterprise sales cycles. Duolingo is the only operator in the cohort that has found the consumer-subscription model that works in education: own the content, gamify the habit, let virality do the distribution, and monetize with app-store subs. The peer set also reveals what is not a moat in this industry: content libraries, university partnerships, and AI models.

3. Is This Business Cyclical?

In a normal recession, Duolingo is fairly defensive: learners stay on the free tier, sub churn may tick up, but the app is a $7–13/month habit that competes with streaming subscriptions, not discretionary luxury. The real cycle risk is elsewhere: app-store fee regimes, FX, ad-market weakness, and AI platform shifts. Subscription revenue is reported in USD but collected globally; a stronger dollar can drag reported growth several points. Ad revenue (about 12% of total) cycles with digital ad markets. And the Duolingo English Test is tied to international student mobility — it softens in a visa crackdown and strengthens on reopened borders.

The 2022–2023 period is the only real stress test we have. Duolingo kept compounding MAUs through the 2022 ad downturn, grew subs through the 2023 rate shock, and accelerated through the 2024–25 mobile-ads weakness that crushed most consumer internet companies. The sharpest reminder that the market can reprice this name quickly was Q3 FY2025: the stock fell roughly 25% on lighter-than-expected guidance as management telegraphed a deliberate shift toward long-term investment over near-term monetization. That is the cycle in this name — sentiment and multiple, not the cash flows.

4. The Metrics That Actually Matter

Forget the income statement for a moment. Five numbers tell you whether this business is winning or breaking.

Why these over the usual superficial ratios: P/E is nearly meaningless here because FY25 net income includes a $257M one-time tax benefit from releasing the US deferred-tax-asset valuation allowance. GAAP margins will appear to drop next year when that tailwind rolls off. EV/Revenue is too blunt because it cannot distinguish "consumer sub with 35% FCF margins" from "enterprise SaaS with 20% churn." The DAU/MAU ratio is where the story lives: it climbed from roughly 28% pre-IPO to 39.6% in four years. That is not a normal trajectory for any consumer app at this scale — it is the specific signature of a habit-forming product.

5. What I'd Tell a Young Analyst

The trap on this name is letting the narrative run ahead of the math in either direction. Bulls frame it as "edtech TAM $123B" — irrelevant, because Duolingo is not competing with a HolonIQ slide. Bears frame it as "AI will commoditize language learning" — possibly true in five years, definitely not true in two, and the filings show AI is more weapon than threat so far (Video Call, Roleplay, content generation at scale). The things to watch are narrow and specific.

Watch these quarterly, in this order:

1. DAU/MAU ratio. If it breaks 40% and keeps climbing, the subscription story has another multi-year leg. If it plateaus near 40% while MAU growth slows, the pipeline starts to clog.
2. Paid-sub net adds versus MAU growth. If subs outgrow MAUs by two-to-one or better, the freemium funnel is tightening — bullish. If MAUs outgrow subs, management is farming awareness at the expense of monetization. Q3 FY25 was the first hint of the latter, which is why the stock fell 25%.
3. S&M as a percent of revenue. Every point above the current 12% means the viral flywheel is losing efficiency and paid acquisition is filling the gap. Cleanest single tell.
4. Stock-based compensation. $137M in FY25 — about 13% of revenue. Reported FCF is ~$360M; SBC-adjusted is closer to $220M. Anyone quoting a 7% FCF yield is ignoring the dilution.
5. DET and adjacency traction. Math, Music, Chess, and the Duolingo English Test are the call option. If any one reaches 5% of revenue, the TAM re-rates. If none does in three years, this is a single-product company at a multiple that assumes otherwise.

What would change the thesis:

• Apple/Google mandated fee reduction (US, EU, or global): instantly adds 5 to 8 points of operating margin with no product change required. Upside-only catalyst.
• A public, peer-reviewed demonstration that a general-purpose chat assistant beats Duolingo on measured learning outcomes — not vibes. This has not happened; filings cite studies pointing the other way. Worth monitoring independent efficacy research, not social media takes.
• Sub-penetration stalling near 9–10% while DAU growth continues — a "running harder to stand still" signal that compresses the multiple fast.

What I'd ignore:

• Analyst downgrades during guide-downs. The business has beaten quarterly consensus in most quarters since the IPO, and the one guide-down in Q3 FY25 was a deliberate investment-cycle reset, not a demand break. Cycle sentiment, not cash flows.
• Breathless coverage of new AI features. The learning flywheel is the moat; AI is the feature set. Don't mistake the two.