Why SaaS freemium doesn't work for AI products

I keep watching AI founders open the Dropbox playbook, build a generous free tier, and then panic three months in when COGS catch up to them.

The script is always the same. Big free quota. Paid plan at $19 or $29 a month. A bet that some sliver of free users converts on their own. Then the bills come in. Inference, GPU spend, retrieval, eval. Every one of those “free” users is leaving a real dollar mark on the books.

Classic SaaS freemium worked because the marginal cost of a free user was close to zero. A Dropbox free user took up 2GB of storage. A Slack free user took up a row in a database. Whether they ever paid or not, you weren’t running a real bill on them. Costs were sunk into the build, and adding the millionth free user looked almost identical to adding the first.

AI broke that math.

The variable cost problem

Every interaction with an AI product costs real money. A single GPT or Claude-class generation on a non-trivial prompt is somewhere between fractions of a cent and a few cents. Multiply that by chat-style usage. Multiply again by power users who turn your product into a daily driver. Now the free tier shows up on your P&L as a real number every month, and that number grows with adoption rather than retention.

A recent Lenny’s Newsletter piece walks through the same problem with receipts. Free-to-paid on AI tools is bleeding founders dry because the unit economics don’t match the playbook everyone studied for the last 15 years. Pure freemium assumes the free user costs you nothing. AI assumes the free user costs you something, every time they show up.

This is not a minor wrinkle. It changes what the funnel has to do.

What’s actually working

Three pricing patterns are quietly taking over from pure freemium, and the founders winning right now have stacked all three.

Small free quota, not unlimited free. The free tier is a sample. A handful of generations a day, or a one-time credit balance that runs out. Enough to show value. Not enough to substitute for paying. Anthropic and OpenAI both run this way on their consumer products. Claude.ai and ChatGPT have a free experience that hits a cap. That cap is the thing doing the marketing.

Consumption pricing for power users. The middle tier is metered. You pay for what you use, and the more you use, the more you pay. Cursor learned this the hard way. Their flat $20 plan was getting torched by a few heavy users running it like an unlimited buffet. They’ve rebuilt the pricing twice in the last 18 months to add caps, then surface usage-based add-ons, then explain the math more honestly on the pricing page. The lesson the rest of the market is taking home from watching them: the heaviest users either pay you proportionally, or they sink you.

A paid plan tied to outcomes, not seats. The top tier sells a result. Not access. A managed workflow, a finished deliverable, a guaranteed throughput. AI products at this tier start to look less like SaaS and more like services with a software backbone. Different sales motion, higher price point, real gross margin once the cost of the free tier is contained.

Stack all three and you have a funnel that survives contact with adoption. Strip any one of them out and you’ve got a hole the unit economics fall through.

What this means for founder-marketers

If you’re a founder-marketer at a small B2B AI company, your GTM math is different from what your former CMO friends are still running.

Your acquisition cost can no longer be padded with a generous free tier. Every signup has a cost. Treat free signups the way the old playbook treated trials: short, useful, finite. Your job is to convert intent (signing up) into behavior (burning the credit) into commitment (a paid plan). The window needs to be narrow enough that you haven’t burned more on inference than the user’s eventual ACV is worth.

That changes content too. Generic “try the free version” landing pages leave money on the table. The pages that work harder qualify upfront, surface the consumption math early, and route the user toward the tier that fits their use case. The same logic I wrote about in the rise of the marketing engineer applies here. The person closest to the product, the data, and the pricing engine is the one who can build a funnel that respects the new economics.

The founder-marketers I’ve been watching are also running their own pricing surfaces through AI tooling now. They’re testing tier breakpoints with synthetic personas, running cohort sims, and updating the page weekly instead of quarterly. Most of that work happens in the same setup I described in your terminal is the new marketing department. Pricing is a marketing surface, and AI tools make tuning it faster than the old review cycle ever allowed.

How to think about your own pricing

If you’re an AI founder still running pure freemium, three questions are worth a hard hour with your CFO or your spreadsheet:

1. What does the median free user actually cost you in a month? Real dollars, not “about a dollar.”
2. What is your free-to-paid conversion rate, and how does it compare to the cost of carrying free users for 60 to 90 days?
3. If you capped free usage at the smallest meaningful sample, what changes? Does activation drop? Or does free-to-paid go up because the cap forces a decision?

Most founders haven’t run those numbers because the dashboard wasn’t built for the question. Old SaaS dashboards weren’t tracking inference cost per active user. Yours probably needs to.

If you want to walk through pricing for a specific AI product, book a call. I’ve been spending a lot of time on this with founders shipping in the field, and the conversation tends to be useful both directions.

The freemium playbook had a 15-year run because the math underneath it held. The math has changed. The funnel needs to catch up.

Further reading

Lenny Rachitsky’s piece on why SaaS freemium playbooks don’t work in AI covers the same ground from the founder side and is worth the read.

Jared Castronova is the founder of JAC Growth Marketing, where he builds AI-powered GTM systems for B2B companies.