पाठशाला Pathshala · हिसाब Hisāb, Unit economics · Lesson 20 · Build
The real cost of free users
A free tier is a channel with a monthly bill. Add up what active free users cost in servers, messages and support, divide by the customers they turn into, and decide what free is buying.
Pathshala, The Founder Library · 11 October 2026 · 7 min read
A free tier never appears on the marketing budget, so it rarely gets the scrutiny a paid channel gets. It should. Free users consume servers, messages and support every month, and the only return on that spend is the share of them who eventually pay.
This lesson treats the free tier as what it is, an acquisition channel with a monthly bill. It lists what a free user costs, shows how to measure conversion so it can be compared with cost, works one company’s free tier through to a cost per paying customer, and sets out what free can legitimately buy beyond conversions. Whether to offer a free tier at all is the subject of [freemium, free trial or neither](/library/freemium-free-trial-or-neither); this lesson is about running one honestly once you have it.
What a free user costs
Infrastructure. Compute, storage, bandwidth and, increasingly, AI inference. A free user who uploads files, runs reports or asks an assistant questions costs money every time they do it. Storage accumulates: a free user who left two years ago may still be costing something every month unless the account is archived.
Messages. In India this is often the largest line. Many products remind, confirm and notify over WhatsApp or SMS. Meta’s WhatsApp Business Platform charges per message from 1 July 2025: every marketing template is charged, and utility templates are charged when sent outside an open customer-service window, which is when most reminders go out. A free plan that sends reminders on the company’s account is paying for each one.
Support. Free users raise tickets, join onboarding calls and write to the founder on LinkedIn. Count tickets from free accounts, multiply by minutes a ticket and by the loaded cost of a support minute. It is rarely zero and often surprising.
Abuse. Free tiers attract fake sign-ups, scraping and accounts created to use a free allowance many times. The cost shows up as infrastructure, fraud checks and engineering time spent defending the tier.
Divide each line by the number of active free users, those who used the product this month, and you have the monthly cost of one free user. Count active users rather than registered ones, because dormant accounts cost little beyond storage, and because the active base is where conversions come from.
Free at scale: what Dropbox disclosed
The clearest public account of a free tier at scale is Dropbox’s 2018 IPO prospectus. At the end of 2017 it served over 500 million registered users but only 11 million paying users, at an average revenue of $111.91 per paying user for the year. The filing warned that a majority of registered users may never convert to a paid subscription, and estimated that about 300 million of them had characteristics that made them more likely to pay over time. It also noted that a significant majority of costs are expensed as soon as a user starts using the platform, so an increase in users could mean more cost recognised than revenue in the early part of a subscription.
Three lessons travel from that page to a seed-stage company. Paying users were about 2 per cent of registered users, and the business still worked, because each free user cost very little and each paying one brought in about $112 a year. The company knew which free users were likely to pay and could say so in a public filing. And it said plainly that free users cost money from their first day.
Measure conversion the way cost is incurred
Kyle Poyar and Lenny Rachitsky’s survey of free-to-paid conversion put 3 to 5 per cent as good for a freemium self-serve product and 6 to 8 per cent as great; for free trials, 8 to 12 per cent is good and 15 to 25 per cent great. Those figures describe the share of sign-ups who ever pay. They are the right numbers for comparing products, and the wrong ones for comparing with cost.
Cost is incurred monthly on the active free base. So measure conversion the same way: of the free users active this month, how many started paying this month? The number is much smaller than the cumulative rate, often a fraction of a per cent, and it is the one that sits under the monthly bill. Track both: the cumulative rate by sign-up cohort, to know whether free users eventually pay, and the monthly rate on the active base, to know what each conversion costs.
One free tier, costed
An Indore company sells billing software to kirana and pharmacy owners at ₹900 a month and offers a free plan that sends payment reminders to the shop’s customers on WhatsApp and reads supplier bills with an AI model. It has 40,000 active free users. Messages and AI cost about ₹12 a free user each month and support adds ₹6, so the free tier costs ₹7.2 lakh a month. Each month about 0.2 per cent of active free users start paying: 80 shops. Each conversion has therefore cost the free tier ₹9,000. A paying shop leaves ₹720 a month after its own costs, so the free tier takes twelve and a half months to repay. The company’s paid campaigns on Meta bring paying shops for ₹6,000.
On those numbers the free tier is a more expensive channel than advertising. The figure shows the two ways out. Drag conversion to 0.3 per cent, where the curve meets the dashed line, and free becomes as cheap as ads; that is the conversion the product team has to find. Or cut the cost of a free user: move WhatsApp reminders to the paid plan and halve the AI allowance, and ₹18 becomes ₹9, which halves the cost per paying customer at the same conversion. Most free tiers that pay for themselves do so on cost, not on conversion.
Look at the payback as well as the cost. At ₹9,000 a paying shop, the free tier takes more than a year to earn back what it spent, which means the company is financing its free users out of the bank for a year before each conversion pays off. That is a cash question as much as a margin one, and it belongs in the runway plan next to the payback on paid channels.
A free tier is an acquisition channel with a monthly bill. If you would not pay that price per customer to an ad network, do not pay it to your own servers.
What free can legitimately buy
Conversion is not the only return on free. A free user can bring others: a shared document, an invoice sent to a supplier, a booking link sent to a patient. A free user can create content or data that makes the product better for paying users. A free tier can be the pipeline Dropbox described, a large base with a measurable subset that will pay later. Each of these can justify a free-tier cost per paying customer above paid CAC.
Each must be measured to count. Referrals: track invitations sent by free users and the share that become paying customers within ninety days, and credit that revenue to the free tier. Pipeline: score free users on the behaviours that precede payment and follow each cohort’s cumulative conversion for a year. Brand and search: harder, so be sceptical. A free tier defended only by words like awareness has not been measured.
Cutting the cost of free
Cap the expensive thing. Google’s BigQuery makes the first tebibyte of queries and the first 10 gibibytes of storage free each month; beyond that, the user pays. A cap bounds the cost of every free user, whatever they do. Move costly features behind the paywall. Messages, AI calls and large storage are the obvious ones; keep in free what costs little and shows the product’s value. Archive dormant accounts. After a stated period without use, move data to cold storage and say so in the terms. Make free support self-serve. Help articles and a community forum, with human support for paying customers. Defend against abuse with phone or GST verification at sign-up for business products, which also improves the quality of the free base.
The monthly free-tier ledger
On the same day as the monthly close, put five numbers on one line: active free users, cost a free user (infrastructure, messages, support), free users who started paying this month, free-tier cost per paying customer, and paid-channel CAC for the same month. Add a sixth if you measure it: paying customers referred by free users. Compare the free-tier cost with paid CAC and with the twelve-month line in the [CAC, LTV and payback](/library/cac-ltv-and-payback-the-three-numbers) lesson.
Then decide one thing. If free costs more per paying customer than ads for three months running and referrals do not close the gap, change the tier: a cap, a feature moved, a dormancy rule. If it costs less, consider spending more on the activities that bring free users in. Either way the free tier now has a number, which is more than most of them ever get.
Nothing here is legal, tax or investment advice. The Indore company is illustrative; message and cloud prices cited were checked in October 2026 and change often.
Sources
- Dropbox, Inc., Form S-1 registration statement, February 2018 — Over 500 million registered users and 11 million paying users at the end of 2017; ARPU of $111.91; about 300 million users more likely to pay; costs expensed as users start.
- Kyle Poyar and Lenny Rachitsky, What is a good free-to-paid conversion rate, Lenny’s Newsletter, August 2023 — Freemium self-serve: 3–5 per cent good, 6–8 per cent great; free trials: 8–12 per cent good, 15–25 per cent great.
- Meta for Developers, WhatsApp Business Platform pricing — Per-message pricing from 1 July 2025; marketing templates charged; utility templates charged outside an open customer-service window. Checked October 2026.
- Google Cloud, BigQuery pricing — First 1 TiB of queries and first 10 GiB of storage free each month. Checked October 2026.