How to get your first 100 users

A hundred users is not a marketing target, it is an arithmetic result, and the arithmetic is the part that almost nobody ever shows you. Most posts on this answer with a list of channels. This one answers with denominators: how large the top of your funnel has to be before a hundred people at the bottom of it is even possible, at the rates somebody actually measured.

Where I stand, before the numbers start: I have not got a hundred paying users. I run a small SaaS that has made one sale, I know to the cent what that sale cost, and I have measured the funnel around it in far more detail than a funnel that size deserves. So this is not a dispatch from the far side of a hundred. It is the arithmetic, plus what my own logs say about the distance. A smaller promise than the title, and the only version of it I can write without inventing something.

The rate everyone quotes is not a rate anyone measured

Look for a free-to-paid conversion rate for a product shaped like mine and the figure that keeps coming back is a median of around eight percent. So I went and looked at what that median is a median of. Roughly 57% of the sample were free trials — hand over a card up front, get billed unless you cancel — and those convert in the region of 25 to 35%. Mixing trials into a freemium median drags it up by roughly a factor of two. It is a real number about a mixed population, quoted as though it described the population you are in.

Here is the band I could assemble for tools in my own category. Veed converts about 0.25% of monthly actives. OpusClip about 0.9% of registered users. Creatify about 1.0%. HeyGen about 1.9% cumulative. invideo says 2.0% of registered users, which matters more than the rest because it came from the CEO rather than from me doing division. Kapwing lands around 3.3% of monthly users. Canva reports 11.7%, and you should not benchmark against it — different product, different denominator, different decade of compounding.

Every figure there except invideo's is derived by me from public numbers, not published as a conversion rate by the company. So treat the set as a band rather than as points: somewhere between a quarter of a percent and a bit over three percent. Two datasets that do publish freemium-specific figures land in the same place — ChartMogul's analysis across 200 products calls 3 to 5% good, and First Page Sage, working from 80-plus customers' own data, reports averages of 3.0 to 3.7%.

Then the part that decides everything and gets written down nowhere: the denominator. The same product produces four different conversion rates depending on whether you divide by downloads, registrations, monthly actives or people who completed the core action. An app-store download costs the developer nothing. My free user has burned about 26 cents of my money, uploaded a screenshot and watched a rendered video to the end. Those two are not the same human being and their percentages are not comparable.

Pick one denominator, write it on the wall, never move it. Mine is the share of accounts that generated a free video who pay within seven days. It is a harsh denominator, and that is the point of it.

So here is what a hundred paying users costs, in people

At 2% — the one first-hand figure in that band — a hundred paying users needs 5,000 people through the free tier. At 0.25%, it needs 40,000. At an optimistic 3.3%, about 3,030. And that is the last stage only. Above it sit visitors who never sign up, signups who never start, and starters who never finish, each with its own leak.

My measured top of funnel is about 26 real human sessions a day. That comes from our own visits table, and it is far below what the server log suggests, because most of what hits the server is bots probing for /backup and PHP back doors, and bots do not run JavaScript so they never reach the table that counts people. In a separate 48-hour window there were 16 signups.

Those two numbers are deliberately not divided here — different tables, different days, and dividing them would manufacture exactly the sort of rate this post is complaining about. What they support is a ceiling. If every human session became a free-tier user, which it does not, 5,000 of them at 26 a day is 192 days. A ceiling, not a forecast; the real figure is some multiple of it.

Which is the whole reframe. A hundred users is not a thing that happens in a launch week. At the rates measured on products like mine, it is a top of funnel one to two orders of magnitude larger than the goal, sustained for months. If your plan contains no number in the thousands, it does not contain a hundred users.

The free user is not free, and the arithmetic has to survive that

My first real sale was a $5 one, in late August, for the small tool that animates a screenshot.

I know to the cent what it cost to serve, because cost is recorded per project: $0.2565 of language model and $0.0083 of cloud rendering, $0.2648 in total. Ninety-seven percent of that is the model. The rendering everybody assumes is the expensive part is nearly free.

Then the payment processor, which takes 4% plus 40 cents on my account, so a $5 sale nets $4.40. The flat 40 cents alone is 8% of the price — small prices get eaten by fixed fees in a way percentage thinking hides.

Now run the hundred. A hundred paying users at $5 is $440 net. Feeding that at a 2% conversion means 5,000 free runs at 26 cents, which is $1,324 of cost. The hundred users lose me about $880.

Break-even at $5 is a conversion of roughly 5.9%, above every rate measured anywhere in my category. At $9.99 break-even falls to 2.83%, still above the only first-hand figure available. That is not a pricing tip. It is the arithmetic pointing out that "get a hundred users" and "have a business" are two different projects, and that the shape of your free tier quietly decides which one you are running.

One more thing about that sale, because it is the part that is hardest to write. It came from nowhere I built: direct traffic, no account, straight onto the tool's own page — a URL nobody guesses, so it arrived by bookmark or through a link I will never see. With no account there is no email address and no first-touch channel for the only customer I have. That is not luck, it is missing instrumentation, and it is worth fixing before you go hunting for ninety-nine more.

What I found when I read my own checkout log

Two confident beliefs went into that afternoon: the price was too high, and the free tier was too generous. Both died in it.

There were 44 checkout sessions. Stripping out my own test accounts left nine real people who reached the payment page. None of them paid.

The detail that changed my mind is a label. The processor distinguishes a declined card, which it marks failed, from an abandoned session, which it marks expired. There were thirty-four expired and zero failed. Not one card was declined, because not one card was entered. Whatever went wrong went wrong upstream of payment entirely.

So I looked at who the nine were. All nine had accounts. Two had a finished video of any kind. Exactly one had ever made the free video the product exists to make. Two had never opened a project at all.

Eight of the nine clicked buy having never seen what the thing produces.

That kills both beliefs at once. Nobody rejected the price — they never got far enough to reject it. And a free tier cannot be too generous to people who did not use it. The defect was a buy button that sends a stranger to a payment page before the product has shown them anything, and no amount of channel work at the top would have fixed it. It is the same discipline as reading the primary sources behind my own industry's conversion statistics: read the log, not the story about the log.

Nine people is a laughable sample. Reading it still changed what I did next, which is more than a hundred borrowed benchmarks managed.

The top of the funnel has to exist before any of this means anything

All of the above assumes people arrive. Mine mostly did not, and the reason was not the one I expected.

Search Console, read page by page: 46 pages, 11 indexed, 35 not. Of the 35, thirty-one sat under "Discovered — currently not indexed" with the last crawled column empty. The count under "Crawled — currently not indexed" was zero. Google knew those addresses existed and had never fetched a single one. There was no ranking problem, because there had never been any ranking.

The sitemap explained half of it. It reported status Success, which is why I had never thought to check it — but last read was fifteen days earlier, and discovered pages read 37 at a point when the sitemap listed 53. Google's copy was two weeks stale and sixteen URLs short. Hence individual pages returning "No referring sitemaps detected" while sitting in the sitemap.

The line that explained the other half stung more: "Referring page: None detected." My blog posts did not link to each other. A page with no internal links and no external ones gives a crawler no reason to spend budget on it, and crawl demand on a young domain is mostly a function of links from places that already have authority.

Which quietly collapses two channels into one. SEO is not a separate channel from getting mentioned by other people; it is downstream of it, and that is the structural reason the list of channels actually open to you is shorter than the internet suggests.

There is a cheaper lever in the same finding. When the free tools went into the site footer, those pages were indexed within days — the same lever, applied to internal linking rather than to outreach. Those tools have since been run 139 times, two of the seven accounting for 78% of it. Small, but measured, and it is the sort of number that should be choosing your next page for you.

Talking to people one at a time has a measured ceiling too

The standard answer to this whole question is: do things that do not scale, go and talk to a hundred people. The instruction is right. It needs a ceiling on it, because "talk to people" gets treated as though it were unbounded and free, and it is neither.

I costed the cold email version before doing any of it. Seven transactional and marketing email providers, all seven prohibiting cold outreach in their acceptable use policies — not discouraging it, prohibiting it as grounds for termination. That matters more than it sounds, because the account you would be risking is usually the same one that sends your password resets. The cheapest legitimate setup I could find was around eight dollars a month on a separate sending domain.

Then supply, because you cannot email an address you cannot find. Measured across 31 company websites: homepage and footer alone yielded one 13% of the time, and the /contact page alone yielded 6% — the exact opposite of everyone's intuition, mine included. Privacy, terms and legal pages yielded 39%, because a privacy policy is legally obliged to name someone. Twelve standard paths together reached 58%, and scanning JSON-LD and JavaScript bundles pushed the ceiling to 84%. Five of the 31 sites had no address anywhere.

The pool feeding that: the two launch directories I was watching list about 49 products a day between them, roughly 40 after removing duplicate domains. So the honest ceiling on the unscalable version is forty companies a day, 84% reachable, and a reply rate you would be fortunate to see in double figures. Do it — but with the arithmetic in front of you, rather than as a substitute for arithmetic.

What I would count instead of a hundred

A hundred is a poor target because it is a number about other people. Four numbers that are about you, in this order.

One denominator, chosen and frozen. Not signups, if signups are cheap and meaningless in your product. Pick the action that means someone has genuinely seen what you do, and divide by that forever.

The count of people who have seen your output. Not visitors, not signups. Mine was one out of nine buyers, and that single figure reorganised my roadmap more than any traffic number ever has.

Cost per free user, measured rather than estimated. Mine is 26 cents and 97% of it is a single line item. Yours might be nothing, in which case your free tier can be enormous — and that is an advantage you should know you have.

Break-even conversion at your current price. Compute it, then hold it against the measured band of 0.25 to 3.3%. If break-even sits above the band, the plan is arithmetically wrong before it starts, and no amount of posting will rescue it.

Launch day, where most people expect this question to be answered, is worth its own treatment. The point that belongs here is narrow: a launch moves the numerator for a day and does nothing whatever to the rates. If the rates cannot get you to a hundred, a spike cannot either.

The thing I would fix first, which is also the thing I sell

Every counting step above assumes a stranger can tell what your product does. Mine could not, and eight out of nine buyers proved it in the coldest way available.

Static screenshots are bad at that job, and the reason most small products never fix it is that fixing it used to mean either weeks in an animation tool or several thousand dollars with a studio. That is the gap PromoHyper exists to close, and it is mine, so weigh it accordingly. You give it your product URL. It reads your site, writes a script you can edit in plain English, and renders an animated promo — voiceover, music, motion, your own product screens and brand colours. The first video is free, watermarked, no card and no subscription. Plans start at $39 after that.

Make the free one, put it in front of the people who would otherwise reach your buy button having seen nothing, and then go back to counting. The arithmetic will not improve because you wish it to. It improves when a stranger finally sees the product before they see the price.