How Fyxer used AI-powered experimentation to grow from $1M to $35M ARR

Executive summary

Fyxer did not grow from $1M to $35M ARR in one year by betting on one big idea. They built an AI-powered experimentation engine.

With GrowthBook as its experimentation backbone, Fyxer ran 541 A/B tests across pricing, onboarding, product-led growth loops, usage limits, messaging, and core AI product experiences. AI helped the team compress the work around each experiment, from research and development to QA, analysis, and internal documentation.

GrowthBook’s MCP server helped automate everything from Cursor to GrowthBook to publication on Slack. That combination helped a four-person growth team move at unusual speed without turning product decisions into guesswork.

The results included a lift in free-to-paid conversion from 5% to 35%, a 50% reduction in annual-plan churn, a 65% increase in trial starts for personal email users, and a referral growth loop tied to 33% of accepted invites.

For Fyxer, AI did not replace experimentation discipline. It made that discipline easier to scale.

The challenge: Hypergrowth makes guesswork expensive

Fast-growing AI companies face a hard problem: the market moves quickly, but not every idea deserves to reach every user.

Fyxer was not just testing low-risk UI changes. The team was testing pricing models, usage limits, signup flows, product-led growth loops, and AI product behavior. Those are business-critical decisions, and shipping the wrong version broadly could create real downside. GrowthBook made that risk visible.

The team could see that only about 25% of experiments were winners. That meant a large share of ideas, even reasonable ones, should not be shipped to 100% of users without evidence.

Experimentation became a way to de-risk growth. Instead of debating which idea sounded right, Fyxer could test what happened when real users experienced it.

Why Fyxer chose GrowthBook

Fyxer needed an experimentation platform that could support a technical team moving at startup speed.

GrowthBook gave the team a flexible way to launch controlled experiments, measure impact, and keep a clear record of what worked, what failed, and what needed another pass. It fit into the company’s engineering-led workflow without forcing a heavyweight process.

That mattered because Fyxer was also building an AI product. Offline evals and model tuning could help improve technical performance, but they could not fully answer whether a product change made users more likely to activate, convert, retain, or expand.

GrowthBook helped Fyxer validate AI product changes with real users and real business metrics.

The solution: AI-assisted experimentation

Fyxer combined GrowthBook with an AI and data stack that included Claude, Cursor, OpenAI, Google BigQuery, dbt, Dot, Tembo, Slack, Linear, Apollo, and Metabase. Using AI tools with GrowthBook's MCP server reduced friction across the experimentation loop.

Claude helped with research, strategy, copy, analysis, and internal workflows. Cursor helped the team build and review experiments faster, including visual experiments where AI-assisted workflows could reduce manual QA. Dot helped answer questions about segments, cohorts, and experiment results using documented warehouse schemas and Slack-based workflows.

GrowthBook tied that work back to measured product outcomes. The loop was simple:

Hypothesis → AI-assisted research and build → GrowthBook experiment → Result analysis → Shared learning → Next test

The impact was not just speed. It was repeatability. Fyxer could run more tests without losing the discipline that made the results useful.

Experimentation became part of how Fyxer operated

Fyxer also made experimentation visible across the company. Experiments, releases, wins, losses, and learnings were pushed into Slack so the broader team could see what was changing and why. GrowthBook became part of that operating rhythm, helping the team create a shared view of product progress instead of leaving results buried in dashboards.

That visibility changed how the company treated losing tests. A failed experiment was not wasted work. It was information the team could use to avoid bad rollouts, sharpen its understanding of users, and decide whether to move on, retry later, or approach the problem differently.

In a high-growth environment, that mattered. Fyxer did not need every idea to work. They needed a fast, reliable way to find the ones that did.

The results

Fyxer’s 541 experimentation volume was impressive, but the real story was compounding learning that drove business growth.

One experiment increased free-to-paid conversion from 5% to 35% by adding a credit card gate before the free trial. That kind of change can be easy to debate and risky to ship blindly. GrowthBook helped the team test it with real users.

Another experiment reduced churn by 50% for annual plans by showing a targeted coupon offer when customers were about to cancel.

A third experiment increased trial starts for personal email users by 65% by segmenting trial lengths based on signup type.

The team also used GrowthBook to test product-led growth loops. Instead of launching one broad PLG motion and hoping it worked, Fyxer tested the steps that moved users forward: the email that brought people back, the signup page they landed on, the message they saw, and the path to add seats. 

Each test created the next learning. The AI-assisted workflows to turn experimentation into a core growth system that helped grow the company from $1M to $35M ARR.