Your Finance Team Is the Integration Layer

A founder I worked with last year had a controller who was genuinely excellent. Organized, fast, accurate. She was also spending about 12 hours a week pulling data from NetSuite, reformatting it in Excel, uploading it to a separate AP portal, reconciling against bank exports, and emailing summaries to department heads. None of that work required her judgment. All of it required her presence.

She had become the integration layer between systems that weren't connected.

What this actually looks like

This shows up in every growing company with a heterogeneous finance stack. You have an ERP that doesn't natively connect to your bank, a payroll tool that exports CSVs the GL can't import cleanly, and an expense platform that codes transactions differently than your chart of accounts. Each gap in the data flow creates a manual step. Manual steps need a person. So the controller becomes the API: authenticate, pull the data, reformat it, upload it somewhere else, verify the totals match. Repeat across six systems, daily.

This is why finance headcount at growing companies often scales faster than revenue. It's not complexity. It's friction multiplication.

A $3.6M seed round closed last week for Zalos, a London startup building AI agents to address exactly this problem. Their bet: rather than waiting for mid-market finance tools to build clean APIs (which won't happen), train agents to work through existing systems the way a human would. Screen recordings become repeatable workflows. That's a real problem getting real capital behind it. The underlying diagnosis is worth taking seriously regardless of whether the AI solution matures.

Why buying more tools often makes it worse

Buying more tools is usually the first instinct when finance operations feel slow. Sometimes that's right. Often it adds another integration point without removing the old ones.

Every new tool requires someone to maintain the connection to your existing stack. If the connection is an API, it needs monitoring when vendors change schemas. If it's a CSV export, it needs a human every time the format shifts. When things break, which they do, the controller finds out by running the numbers and getting a variance that doesn't make sense.

Fragmented finance stacks don't fail loudly. They fail through slow accumulation of small errors that nobody catches until the monthly close is already a week late.

The diagnostic question

Before you buy anything, map where a person is the integration. Ask your controller to track their time for two weeks in two buckets: judgment work versus transfer work. Judgment work is decisions, analysis, interpretation. Transfer work is moving data from one place to another.

Most companies find the split is 40/60 in the wrong direction. That's the problem to fix, and the map tells you which integration gaps are actually worth investing in.

Don't aim to eliminate your finance team. Aim to make sure they're doing work that actually requires them.