Tally Hits $6M ARR With 10 People as AI Assistants Bring 43% of Its New Users

TL;DR

Tally, the bootstrapped form builder from Ghent, Belgium, published its six-year update on September 22: $501K MRR, which is $6M ARR, 2.5 million users, a team of 10 including both co-founders, and no outside funding. The most useful number for anyone building a SaaS is not the revenue. It is this: 43% of new users now discover Tally through ChatGPT, Claude, Gemini, Perplexity and similar assistants, up from AI search being merely the #1 referral source in mid-2025. The post also discloses 1 million AI tool calls in August, a two-agent code reviewer whose comments get acted on 81% of the time, and a data breach that came through a third-party analytics tool.

The numbers

Tally has published an MRR update at nearly every milestone since launching in 2020, which makes it one of the better-documented small SaaS businesses on the internet. Co-founder Marie Martens wrote this one after a four-week family break, which is itself a data point about how the company runs.

• Revenue: $501K MRR in September 2026, up from $422K in April 2026 (the $5M ARR post) and $338K in October 2025.
• Users: 2.5 million.
• Team: 10 people, co-founders included. The April update said 11.
• Funding: none. Tally describes itself as customer-funded.
• Model: freemium. The free plan has unlimited forms and submissions and carry a "Made with Tally" badge; revenue comes from users who upgrade to Tally Pro.

Divide $6M by 10 and you get roughly $600K of ARR per person, a number most venture-backed SaaS companies with ten times the headcount would like to have.

That flat patch is deliberate. In January Tally said it would optimize for quality, not revenue, and set no revenue goal for 2026. MRR still grew about 19% between the April and September posts.

AI search is now the biggest front door

This is the part to copy into your own planning doc. For five years, Tally grew through a classic product-led loop: someone fills out a Tally form, sees the badge, signs up for free, and a slice eventually pays. The new post puts that upgrade rate at about 1% (earlier updates said 2%).

In the June 2025 update, Tally reported that ChatGPT had become its #1 referral source, with over 2,000 tracked signups a week coming via AI tools, and guessed that GPT-4o's default web browsing caused the spike. Fifteen months later the share has, in Martens' words, "accelerated": 43% of new users now say an assistant sent them.

Two caveats keep this number honest. First, it is self-reported: Tally measures it with an onboarding survey that asks where people found the product, not with referrer headers, which undercount AI traffic anyway. Second, Tally is explicit that it does not control the channel. "We never bought this channel. Well, you can't (yet)," the post says, and "a model update could still reshuffle recommendations overnight."

What Tally actually does for AEO

No secret prompt-injection trick, despite what Martens calls the "wild theories floating around online." The listed work is boring and reproducible:

• Track where LLMs mention and cite Tally, and look for prompt gaps where a competitor gets recommended instead.
• Audit the help center so what models read about Tally is accurate.
• Keep competitor comparison pages current.
• Show up in the public conversations models learn from. The June 2025 post credited years of answering questions on forums and Reddit, noting that user-generated content and forums were "winning in AI search."

Think of an assistant as a very well-read friend who gets asked for a form builder recommendation a million times a day. It recommends whatever it has seen real people praise, in many places, for years. You cannot buy that friend a drink, but you can make sure every page it reads about you is correct.

AI inside the product

Tally is careful to say it is not becoming an "AI form builder." The AI features are opt-in, and the post admits some users pushed back on the company joining "the AI train." The usage still moved fast:

• Tally AI went from 85K calls in April to 767K in August.
• Combined with the Tally MCP server, which lets Claude, ChatGPT and other clients build forms and analyze submissions, August crossed 1 million tool calls.
• 30% of forms created by new accounts are now built with Tally AI.

There is a neat loop here. The same assistants that recommend Tally can now operate it through the Claude connector and ChatGPT integration, so a user can go from "which form builder should I use" to a published form without opening Tally's editor.

AI inside the team

The operational section is the most reusable part for a small team. The engineers write code with a mix of Claude Code, Codex and Cursor. Every pull request gets a first pass from filip[bot], a Claude-powered reviewer:

• It reviews each PR through 33 lenses, such as security, performance and Tally's own conventions.
• Every finding must survive a second agent whose only job is to disprove it by running the actual tests. Only surviving findings get posted.
• About 81% of its comments get acted on before merge, before a human reviewer opens the PR.

That adversarial second pass is the clever bit. Most AI reviewers fail by drowning a PR in plausible-sounding noise; making a second agent try to kill each finding is a cheap filter that keeps the signal rate high enough for engineers to keep reading.

The same setup runs three health monitors, on Sentry errors, Tally AI sessions and the support inbox, that investigate a failure and open a triage-ready Notion ticket. Support tags Claude in Slack to find past cases and help articles, and marketing has skills wired over MCP to tools like Ahrefs. The stated rule: nothing AI-generated is published as-is, and no code reaches production without a human looking at it.

The breach

The update is not all victory laps. In early August, Tally learned that its internal Metabase analytics instance had been compromised through a zero-day, exposing account email addresses and hashed passwords. Tally says no form data was accessed and takes the blame for letting that data sit in an analytics tool at all. It emailed its whole user base, removed Metabase, built internal stats instead, and is pursuing ISO 27001 and SOC 2, expected by the end of 2026. Martens reports thousands of replies, mostly supportive, which may be the only time in history a breach notification email improved a brand.

Why this matters

Most "AI search is eating SEO" commentary runs on vibes and agency decks. Tally gives you a longitudinal series from one real business: #1 referral source in June 2025, 43% of new users in September 2026, same product, same free plan, no paid acquisition. If you sell a horizontal tool that people ask chatbots about, your acquisition mix is probably moving the same way, whether or not your analytics can see it.

The catch is concentration risk. A channel that delivers almost half your new users is controlled by a handful of model providers who change their models every few months. Tally's own hedge is the old playbook: a good free plan, a product people praise in public, and a community. As Martens puts it, AI "raises the bar for craft, brand, and community. Exactly the things that don't scale."

Key Takeaways

• Tally crossed $6M ARR ($501K MRR) with 2.5 million users, 10 people and no outside funding.
• 43% of new users now discover Tally through AI assistants, up from ChatGPT becoming the #1 referral source in mid-2025. The figure comes from an onboarding survey, not referrer tracking.
• Its AEO work is plain: track LLM citations, fix prompt gaps, keep help and comparison pages accurate, and be present in public forums.
• Tally AI calls grew from 85K in April to 767K in August; with MCP, August passed 1 million, and 30% of new accounts' forms use AI.
• A two-agent PR reviewer, where a second agent tries to disprove each finding with real tests, gets 81% of its comments acted on.
• The upside and the risk are the same thing: nearly half of new growth depends on recommendations Tally cannot buy or control.

Sources: Tally: 6 years in, $6 million far, Tally: The road from $4M to $5M ARR, Tally: From 2 to $3M ARR, Tally: In 2026 we're optimizing for quality, not revenue, The Neuron, September 22 digest