The free/paid line
Where you draw the boundary between what’s free and what costs money decides who becomes a customer and who becomes a permanent freeloader.
Every product that touches strangers ends up drawing a line it never explicitly debated: some of it is free, and some of it costs money. Get that line right and the free side recruits customers while you are asleep. Get it wrong and it recruits nothing — it just quietly obligates you to answer email from people who were never going to pay.
You already have a number that belongs in this decision. When you measured a project’s maintenance load — the hours per month it takes to hold a product exactly as good as it already is, at zero growth — that number showed up purely as a cost, something to subtract before deciding whether a project earns its keep. It is about to reappear on the other side of the ledger. A free tier is where a slice of that measured time actually goes, and until you can say what it buys in return, you have only priced half the decision.
Free tier as marketing line, not free tier as cost centre
The classic case for giving something away is acquisition: let people try the thing, some fraction get hooked, some fraction of those convert. Getting Real’s “Free Samples” chapter argues exactly this — give a taste away to get noticed and hook people who later upgrade — and stops there. It says nothing about what serving those free users actually costs. That silence is not a flaw in the chapter; it is the shape of almost every pro-free-tier argument you will encounter, and it is exactly the gap this lesson exists to close.
Jason Cohen’s reframe supplies the other half without needing a single benchmark number to do it:
“Retool your expectations of Freemium: It’s a marketing cost.” (Cohen, Reframing “Freemium”)
His concrete proposal: total up the infrastructure and support cost attributable to free users, have the marketing budget actually reimburse the departments that absorbed it, and then judge that spend by the same test as any other acquisition channel — compare its ROI against ads or SEO. A free tier stops being a vague worry the moment it becomes a line item with a number attached, win or lose.
What the free tier has to be for
Under the marketing-cost framing, a free tier earns its keep exactly one of two ways: some of the people using it for free convert to paying, or some of them bring you paying customers who never used the free tier themselves — a referral, a recommendation, credibility that closes a sale somewhere else. If a free tier is doing neither, it is not a funnel with a slow conversion rate. It is an audience you are supporting for free, full stop.
Cohen’s essay adds a scale problem worth carrying into this: free users vastly outnumber paying ones, by his own account “20:1 or even 100:1.” That is offered as observation from running two companies, not as measured data, and it is a ratio of people, not of support tickets — but the implication holds either way. Whatever a free tier generates — feedback, support requests, forum noise — arrives overwhelmingly from people who were never going to pay you anything, which is exactly why it has to be judged by what it buys and not by how much attention it commands.
The solo trap: support with no revenue behind it
The danger this lesson is actually aimed at is specific to running alone. A free tier can quietly accumulate real support hours long before anyone has decided whether that spend is paying off, because there is no team meeting where someone asks. Every hour a free tier eats is an hour that came from the same finite pool you already measured as maintenance load — there is no second bucket of time a solo builder can draw from instead.
That is the connection worth holding onto: the hours-per-month figure you measured earlier was framed as pure cost, unattached to any revenue. A free tier is one of the places that cost concentrates. If none of those hours are buying paying customers — directly or by referral — then the free tier is not marketing. It is unfunded support, and unfunded support is exactly the kind of toil a solo maintenance number was built to surface.
Where to draw the line when the product is small
At solo scale there is no statistical comfort — 400 users behave nothing like 400,000, and a handful of support threads can be the entire free-tier dataset you will ever have. A few things follow from that:
- Draw the free tier narrow enough to demonstrate real value without requiring ongoing person-to-person support — narrow the promise, not just the feature list.
- If a free feature keeps generating the same support question, you have three real options: fix the thing so the question stops, remove the feature, or move it behind the paid line. Leaving it exactly as it is, unpriced and unfixed, is not a fourth option — it is the default everyone drifts into by not choosing.
- Revisit the line on the same cadence as the rest of the ledger, not only when support volume becomes impossible to ignore. The honest quarterly question is whether the free tier bought anything — paying conversions, or paying customers who said someone told them about it — and whether that was cheaper than any other channel available to you. If the answer is no, shrink or gate it.
Why the conversion numbers you will find are not trustworthy
Search for a freemium conversion benchmark and you will find confident numbers everywhere, usually clustered around “2–5%.” Every one traced back far enough resolves to one of: a vendor’s self-interested survey, an SEO agency’s own client roster, or an unattributed sentence borrowed from a paper that measured something else entirely. The agreement between sources is not independent corroboration — it is the same handful of claims being recirculated.
The clearest example of the last failure mode: a peer-reviewed paper on freemium sampling strategies opens its abstract with “most firms suffer from too few premium subscribers (3–5%), which challenges their profitability” — a sentence with no citation of its own. The study behind it measured a 225-subject contest-based online experiment comparing trial-ordering strategies. It did not measure real-world conversion rates at all. Citing that paper for “3–5% is typical” launders a marketing-grade number through an academic venue — exactly the move that gives an unsourced figure false authority.
The vendor numbers fare no better on inspection. One widely quoted set comes from an SEO agency reporting conversion rates across its own client roster — companies that hired and stayed with an SEO firm are not a random sample of software products. Another comes from a venture firm’s benchmark report whose canonical page now returns a 404, while blog posts still cite its numbers as if the source were one click away. None of the datasets in circulation measure a solo or small-indie product at all — every one is built from companies large enough to be surveyed, be a portfolio company, or be an agency client.
Where people get burned
If you want the shape of reality rather than a fake-precise average, the most honest thing available is a spread, not a number: a self-reported January 2026 survey of roughly 200 B2B software products found around a fifth of respondents converting free users to paid below 2.5%, and nearly a quarter converting above 25%. State plainly what that is and is not. It is self-reported — the companies still standing and willing to answer are not a neutral sample — and it measures B2B software companies large enough to take a survey, not a product built alone in the evenings. It is not a forecast for your project. The one real lesson in it is that any single “typical rate” collapses a distribution spanning an order of magnitude, which is a fact about the numbers, not a number you can plan around.
The absence of a trustworthy figure here is itself worth knowing, not a gap to paper over. A reader who goes looking will find precise, confident percentages on nearly every blog post about this topic. Now you know why none of them earn that confidence — and why the marketing-cost test above does not need one to work.
Retrieval check
A free tier has 400 users, 6 of them paying, and eats about four hours a month of your support time. What is that free tier costing you, and what is it buying?
Check your answer
The cost is not mysterious: four hours a month is real time, drawn from the same finite pool as every other maintenance number on the shelf — call it real dollars at whatever rate you charge for your own hours. What it is buying cannot be read off the 6-in-400 ratio alone, because that number says nothing about where the six paying customers actually came from.
If those six converted out of the 400 — or if any of the remaining 394 are the reason a seventh customer showed up elsewhere — the four hours are a marketing cost, and the real question is only whether its ROI beats another channel you could run instead. If the six would have paid regardless, and none of the 394 are bringing anyone in, the four hours a month are not buying anything. That is unfunded support: real cost, on the same ledger as every other hour you have already measured, with no revenue attached to defend it.
Now draw the line
Hands on
Draw the line for one project
Done when: The Free/paid line field in PORTFOLIO.md for one project states, in one sentence, what’s free, what costs money, and what the free side is for.
- Pick one project from the shelf — one that either already has something free about it, or would need to if it shipped a paid version.
- List, plainly, everything currently free (or everything that would be free) and everything that would cost money.
- For each free item, ask one question: does giving this away make someone more likely to become a paying customer, or bring you one? If the honest answer is no, mark that item as a candidate to cut or move behind the paid line.
- Estimate, from your own judgment rather than a benchmark you found online — there is not a trustworthy one to borrow, as this lesson just showed — roughly how many hours a month the free side would cost in support at a realistic scale for this project.
- Add a Free/paid line field to that project’s entry in
PORTFOLIO.md— it does not exist there yet — and write one sentence stating where the line sits and why. - Bring it into the chat. I’ll push on whether the free side is actually funded by anything, or just optimistic.
Check your recall
Answer from memory — no scrolling back.
What this does not cover
Drawing the line settles what’s free and what costs money for a project that already has a price and a stripped-down paid version. It says nothing about how the first people actually cross that line and start paying. The honest answer is that they do not arrive through a funnel — why the first ten customers get sold by hand instead is next.
Read this next — primary source
Reframing “Freemium” by charging the marketing departmentJason Cohen, A Smart Bear (longform.asmartbear.com) — free, about 15 minutes
The page carries a promotion for Cohen’s book at hiddenmultipliers.com and an AI Skills course at skills.asmartbear.com — a normal, disclosable commercial interest, not a disqualifier. Cohen ran two companies (SmartBear, WP Engine) to real scale, so the reframe this lesson borrows is an argued position built on real experience, not a measured study. It is worth reading in full because it is honest about that distinction in a way most freemium advice is not: it names its own sources rather than asserting numbers bare.
Stuck, curious, or think this lesson is wrong? Ask your teaching agent. The lessons are the scaffold; the conversation is where the learning gets unstuck.