Why Running a Service Business First Gave Fyxer Instant Product-Market Fit

What Everyone Says

Build fast. Ship fast. Get to market before someone else does. The AI gold rush rewards speed above everything else. Every week a new competitor launches. Every month the models get better. If you're not shipping product, you're falling behind.

VCs want to see traction now. Founders panic about being too slow. The pressure to skip straight to building software is enormous.

Why That's Wrong

Speed without insight is just expensive guessing. Most AI products launch, get initial buzz, then die because they solve problems that don't actually matter to customers. Or they solve them just well enough that nobody pays.

The hidden assumption is that being first matters more than being right. But in AI, the model layer is commoditized. GPT is available to everyone. The real moat is knowing exactly what to build and having the data to build it better than anyone else.

What Richard Did Instead

Richard Hollingsworth and his brother Archie spent six years running what became the UK's largest executive assistant agency before writing a single line of code for Fyxer AI. That wasn't an accident. It was the plan from day one.

On the first day of the agency, they asked every assistant to log and describe every task they performed for clients. Six years of that data told them exactly which workflows people actually use assistants for. Not what customers say they want. What they actually pay for.

They bootstrapped that agency to around $5 million and hired hundreds of executive assistants along the way. Three serious attempts at building technology to automate the service failed. Then GPT-3 arrived.

The first workflow they built was inbox organization. People were paying the agency $60 an hour for this. Fyxer charges $30 a month. They knew the demand was real because they'd been selling it as a service for years.

Result: instant product-market fit. They went from $1M to $18M ARR in nine months. Not because they moved fast. Because they knew exactly where to aim.

The Principle Underneath

Service businesses generate two things that pure software startups don't have: verified demand and training data.

When you run a service, you learn what customers actually pay for (not what they say they'd pay for in a survey). You learn which workflows are repeatable. You learn where humans are slow and expensive. And if you're logging everything, you're building a dataset that no competitor can replicate.

Richard put it simply: "We know more about what customers want from this product than anybody else in the world." Six years of time-tracking data gave them that confidence.

The agency also taught them what doesn't work. Three failed attempts at building tech-enabled services showed them exactly where the technology wasn't ready. When GPT-3 arrived, they recognized immediately that the price of automation had just dropped 99%.

Should You Do This?

Do this if you're entering a space where understanding the workflow matters more than the technology. If the AI layer is commoditized (and it usually is), your advantage comes from knowing exactly what to build and having data to make it more accurate.

Skip this if you're in a market where speed genuinely matters more than accuracy. Some products reward being first. AI products that replace expensive human work reward being right.

One question to ask: Can I describe the exact workflow my AI will automate, step by step, from watching humans do it hundreds of times? If not, you might need more time in the trenches.