Operating Machine Is Payroll Moat
Jeremy Zhang, CEO of Finch, on building a universal API for employment systems
The real moat in payroll infrastructure is not a prettier API, it is the operating machine behind it. In HR and payroll, every customer sits on a slightly different system, many still move data through CSVs, SFTP, and manual admin screens, and even large buyers like Human Interest historically needed dozens of custom connectors plus large ops teams just to make deductions and employee records line up on each payroll run.
-
Finch is selling relief from connector sprawl. Jeremy Zhang describes roughly 6,000 U.S. HR and payroll systems, with the top 10 covering only about 55% of the market, which means reaching broad coverage takes far more integrations than in banking. That fragmentation is why manual fallback becomes part of the product for many API companies.
-
For a 401(k) provider like Human Interest, this work is operationally critical, not cosmetic. The product has to pull employee census and paystub data, then write contribution deductions and employer matches back into payroll every pay period. Human Interest now markets 500 plus payroll integrations and automated contribution processing because this workflow directly determines plan setup speed, compliance, and audit burden.
-
This is also why universal APIs and embedded payroll are complementary. Finch sits above systems of record and standardizes access across fragmented payroll vendors, while companies like Check and Gusto Embedded create new payroll rails underneath. More embedded payroll products should increase the number of back end systems in market, which makes the normalization layer more valuable, not less.
The market is heading toward more visible automation on the surface and more specialized infrastructure underneath. As retirement, benefits, compliance, and fintech products keep plugging into payroll, winners will be the companies that can hide the messy last mile of payroll setup, data cleanup, and deduction logic well enough that customers experience it as software, even when a large operations engine is still doing part of the work.