Freemium killed a friend’s startup. Eighty thousand free users, a Product Hunt standing ovation, $4,000 in MRR, and then the AWS bill arrived. The model was not wrong; the arithmetic was never done. What follows is the math, honestly stated, including the parts most freemium cheerleaders skip.
Freemium is the most misunderstood model in SaaS: founders copy the surface (a free tier!) without the math that makes it survive. The ending is always the same, 100,000 free users, $0 in expansion, and a burn rate that kills the company. Value capture starts with arithmetic, so here is the freemium equation, honestly stated.
The Core Equation
Freemium works when: (free users × viral coefficient × conversion rate × ARPU) > (free users × cost to serve). Every variable matters. A product with zero virality needs paid-level conversion (3–5%) to survive; a product with strong virality (each user brings 0.3+ users) can survive on 1–2% conversion because acquisition is nearly free. Most failed freemiums have neither: no virality, 0.5% conversion, and full support costs.
Design Rules That Protect the Math
- Gate the expensive axis, free the viral one. Free users should be able to invite, share, and publish, actions that acquire more users, while hitting paywalls on storage, seats, and advanced features. This is feature gating done surgically: the free tier is a distribution engine, not a charity.
- Cap support cost per free user near zero. Self-serve onboarding, community support, no live chat for free tiers. If a free user can consume $40 of support time, your model is broken regardless of conversion. Automate or remove every human touchpoint below the paywall.
- Time-limit the value cliff, not the account. Pure free-forever tiers accumulate dormant users who cost storage and skew metrics. Better: free-forever for the viral core, trials for premium depth, the hybrid that combines endowment psychology with freemium reach.
- Price-discriminate inside the funnel. Students, startups, and nonprofits on free; enterprises fenced into annual contracts. Deliberate price discrimination keeps willingness-to-pay segments from collapsing into the free tier.
The Numbers to Watch Weekly
Free-to-paid conversion by cohort (target 2–4% for SMB, higher for prosumer), cost-to-serve per free user per month (target: cents, not dollars), viral K-factor by channel, and “free forever” cohort size, dormant accounts older than 6 months with zero activity should be archived, not subsidized. Add one business-health metric founders skip: revenue per employee including the infrastructure free users consume. If that ratio decays as free grows, you are scaling a liability.
When Freemium Is the Wrong Model
Skip freemium when marginal cost per user is high (anything with humans or heavy compute in the loop), when the buyer is not the user (enterprise security tools, free users never convert because they never buy), or when usage-based pricing captures the same low-end with less subsidy. The bravest freemium decision is killing the free tier: several public SaaS companies doubled growth the year they replaced free with a 14-day trial.
The bravest freemium decision is killing the free tier.
Frequently Asked Questions
What free-to-paid conversion rate is healthy?
Two to four percent for SMB self-serve, higher for prosumer, lower acceptable only with strong virality subsidizing acquisition. Below one percent with no viral loop means the free tier is charity.
How do you keep free-tier support costs near zero?
Self-serve onboarding, community support, no human touchpoints below the paywall, and automated dormancy archival. If a free user can consume forty dollars of support, the model is broken regardless of conversion.
When should you kill a free tier?
When marginal costs are high, when buyers aren’t users, or when usage-based pricing captures the low end with less subsidy. Several public SaaS companies doubled growth the year free became trial.
Continue Reading: Pricing Power & Value Capture
This article is part of our series on Value Capture: How Much Created Value Can You Keep?. Related reading: