Tally Turned a 2% Conversion Into $5M ARR

The Case: $5M ARR on a 2% conversion

What she does. Marie co-runs Tally, a Notion-flavored form builder positioned against Typeform and Google Forms.

Backstory. Public in 2020, Product Hunt in 2021, no funding, no growth hacks. Four years of manually answering Reddit threads, seeding help-center content, and getting mentioned in comparison pages. In 2025-26, AI search jumped from mid-tier channel to #1 without them changing the strategy — because LLMs started citing the corpus they had been quietly building since day one.

Product. - Block-based form builder with Notion-style editing - Free tier includes almost every feature, with Tally branding on every submission - Pro tier removes branding, adds logic, integrations, payments - Templates for waitlists, surveys, applications, order forms - Native integrations, webhooks, API

Pricing. Free forever with Tally branding. $29/mo Pro removes it and unlocks the feature stack.

Key numbers. - ARR: $5M (July 2026) - Time to $5M: 5 years, fully bootstrapped - Free-to-Pro conversion: 2% - Implied paying accounts: ~14K on ~700K free base - #1 acquisition channel 2026: AI search (LLM citations) - Prior stacked channels: Reddit answers, brand mentions, help center, comparison pages, listicles - Team: still tiny (undisclosed, "still small" per Marie) - New posture: revenue targets dropped, optimizing for product quality - Tooling for LLM visibility: Promptwatch + Ahrefs

Stack. Reddit social listening, help center CMS, Ahrefs, Promptwatch, Product Hunt community, comparison-page network.

Why this works (and what most readers will miss)

1. The free tier is a media network, not a loss leader. Every free form ships with Tally branding into a stranger's inbox. Each submission is a native-context ad seen by someone who already trusts the sender. Free users aren't a cost — they're the ad inventory that funds the whole machine.

2. AI search didn't create the moat, it revealed it. Four years of Reddit answers and comparison pages became LLM training and retrieval data. When ChatGPT or Perplexity gets asked "best form builder," it says Tally because Tally was named a million times in the corpus. New entrants in 2026 cannot retro-seed that.

3. The 2% conversion is a filter, not a leak. Most operators would panic and A/B test their way to 4%. Tally leaves it alone because the 98% who don't pay are still running the distribution loop. Optimizing conversion would strangle the ad network.

The Distilled Read

Kill the form-builder framing. Tally is not a form builder. Tally is a branded envelope that ships itself to hundreds of eyeballs every time a free user hits publish. The product happens to build forms. The business is what happens after the form is sent: a respondent sees "Powered by Tally," a fraction investigate, a fraction of those sign up, and 2% of those pay. Read the case at the wrong altitude and you'll build a better form builder. Read it right and you'll build a better envelope.

The templatable pattern is branded-freemium, not SaaS. The playbook is: pick a category where the output artifact gets shared with third parties, ship a generous free tier that watermarks every artifact, price the paid tier at watermark removal plus a real feature stack. Works for form builders, resume builders, invoice tools, calendar links, one-pagers, quiz makers, changelogs, waivers. Does not work for internal tools, backend software, or anything the user never shows another human. First filter for any category you're weighing: does the deliverable get sent to a stranger's inbox?

The economics only close because the team stays tiny. $5M ARR at 2% conversion is roughly 14K paying accounts on ~700K free users. Service that with a 40-person team and you'd need enterprise pricing and enterprise motion, and the whole distribution loop collapses under features nobody asked for. Tally holds margin by holding headcount. That's not humility, it's math. Every hire past "the product just works" would eat the surplus that funds the free tier that runs the ad network.

Distribution was seeded manually for four years before it went automatic. Marie's throwaway line — "actively investing in social listening and providing helpful answers since 2020" — is the whole case. That's a founder who personally answered thousands of Reddit threads for four straight years before any of them became an LLM citation. There is no shortcut. You cannot buy your way into the training corpus. You cannot spin up a Reddit farm in 2026 and expect Perplexity to trust it in 2027. The moat is time-shaped, and the window to start pouring is now, for the AI search of 2029.

Here is the observation nobody in the thread caught. Marie thinks the moat is community. It's not. The moat is that every branded form is simultaneously three assets: a customer acquisition unit, a brand mention that will be scraped by the next model, and a training example teaching LLMs the phrase "form builder → Tally." She's running a customer flywheel that doubles as an LLM-retrieval flywheel that doubles as a brand-recall flywheel — and from inside the company it just looks like "the free tier works." Anyone copying this playbook in a new category in 2026 who does not weaponize the branded-artifact-as-LLM-signal on day one will finish 2028 wondering why the models won't say their name.

Steal-the-Playbook

1. Pick one artifact-sharing niche, not the category. Not "productivity" — "invoices freelancers send to clients," "waivers dance studios email to parents," or "onboarding forms bootcamps send to enrollees." One niche where the artifact lands in a stranger's inbox by design.

2. Ship v1 in 72 hours on boring stack: Next.js, Supabase, Stripe. Free tier hard-watermarks every artifact. Paid tier at $19-29/mo removes the mark plus adds one feature the niche actually asked for.

3. Answer 3 Reddit threads a day in that niche for 90 days. Same account, no burner, helpful first and product-linked second. This is your 2028 LLM corpus, not your 2026 traffic play.

4. Publish 8 comparison pages ("Tool X vs Tool Y for [niche]") and get named in 12 listicles by writing them yourself on Medium, Dev.to, and niche blogs, then pitching to niche newsletters. This is your citation graph.

5. At month 6, install Promptwatch or run manual GPT/Perplexity/Claude queries. Track how often each model names you unprompted for niche questions. That number is your real growth metric — MRR is the lagging indicator.

Stack: Next.js, Supabase, Stripe, Postmark, Ahrefs ($108/mo), Promptwatch (free tier). Cost: ~$130/mo. Setup: one weekend for product, 90 days for corpus.