Product-led growth ends at the first pay run

Product-led growth has become the default ambition of business software. The idea is attractive and often right: let customers find, try and adopt a product without waiting for a salesperson, and let its usefulness do the persuading. Companies built that way can grow with lower acquisition costs and more contented customers. Boards hear the phrase in nearly every software strategy discussion, and at Yarrowby we have adopted parts of it with good results.

I want to argue that it has a limit in payroll, and that the limit can be stated precisely. Self-service can carry a customer up to the moment their first live pay run is about to happen. It should not carry them through it alone. I came to that view after a pilot in 2024 that our product team and I designed together, and whose results neither of us expected.

Where self-service earns its place

Our free rostering tier shows what self-service does well. Since 2022 any workplace has been able to use Yarrowby Rosters free for a single site of up to 30 staff. Managers build shifts, employees swap them on their phones and nobody from Yarrowby needs to be involved. By mid-2025 about 11,600 workplaces had signed up. Roughly 14 per cent moved to a paid rostering plan within a year, and a smaller share later moved their payroll to us.

Rostering suits self-service because mistakes are visible and cheap. A badly built roster is noticed by the people working it and can be fixed before the shift starts. Its value appears within days, and a customer who tries it and leaves has lost an afternoon.

Self-service also works for much of what comes before a payroll decision. Prospective customers can explore our product in a demonstration environment, read our guides to award interpretation and Single Touch Payroll reporting, and estimate their likely bill without speaking to anyone. We removed the forms that once stood in front of most of this material, and our qualified opportunities did not fall.

The pay run nobody should attempt alone

Payroll is different because mistakes land in people's bank accounts. A first live pay run on a new system depends on many things being right at once: employee records migrated with correct year-to-date figures, leave balances carried over, award rates and penalty rules configured, superannuation and tax settings checked, and reporting to the tax authority set up. In Australia that reporting runs through Single Touch Payroll; in the United Kingdom it is real-time reporting to HMRC. A mid-market employer with several awards and a few hundred staff has thousands of settings, and the person configuring them is usually running the old payroll in the same fortnight.

In 2024 our product team proposed a fully self-service route, all the way to the first live pay run, for employers with up to 150 staff. I supported a trial, partly because the economics were attractive: assisted implementation costs us several thousand dollars for each new customer. We agreed to run the two routes side by side for six months and compare them on outcomes customers care about rather than on activation.

The self-service route was faster. Customers reached their first pay run in about half the time. But of the 212 employers who went live without assistance, 31 per cent had at least one pay error in their first three pay runs that needed correcting, against 6 per cent of a matched assisted group. Cancellations within ninety days were 9 per cent against 2 per cent. The errors were rarely dramatic: an allowance missing, a leave balance wrong, an overtime rule applied to the wrong employment type. Each was a conversation between a payroll officer and an employee who had been paid incorrectly, and each was charged to our reputation.

We closed the fully self-service route in early 2025 and redesigned the journey instead. Customers still do most of the setup themselves, guided by the product. They then process a parallel pay run, a real pay period calculated on Yarrowby alongside their old system, and an implementation specialist reviews the differences with them before anyone goes live. The specialist's time per customer has fallen by about 40 per cent, because the customer arrives having done the groundwork. Assistance now costs less, and it falls in the one place where it prevents real harm.

The pilot also changed what marketing promised. For two years our advertising had told employers they could be paying staff through Yarrowby within a week. That was true for a small business with one award and tidy records, and misleading for most of the employers we actually wanted. We withdrew the line and replaced it with a description of the parallel run: what the customer does, what our specialist checks and how long it usually takes for an employer of their size. Enquiries from very small businesses fell, as we expected. Enquiries from employers with more than 200 staff rose, and our implementation team stopped beginning every project by correcting an expectation that marketing had set.

Measuring the handover

The more important change was in what we measure. Product-led companies usually track activation, time to value and conversion from trial to paid. Those measures are useful, but in payroll each of them can be met by a customer who is about to make a mistake. We now track how many new customers complete a verified parallel run, how many of their first three live pay runs need no correction, and how many are still with us after a year.

These measures are slower and less flattering. They also align product, marketing and implementation around one result. Marketing cannot claim success for a sign-up that ends in a corrected payslip, and product cannot claim activation for a customer who leaves at the first renewal.

I do not think the argument belongs only to payroll. Any software that moves money, issues legal documents or controls physical equipment has a moment when an error stops being the customer's inconvenience and becomes somebody else's harm. Product-led growth works well up to that moment. Beyond it, a person who knows what can go wrong is not a cost to be engineered away. That person is part of the product, and should be priced and marketed as such.