The human API: what swivel-chair work really costs your P&L
Every enterprise has an integration layer made of people. Here is how to put a number on it — and why the cost is mostly not where finance leaders expect.
| Author | Retypd Research |
|---|---|
| Published | |
| Reading | 5 min read |
In every enterprise there is an integration layer that never appears on an architecture diagram. It is made of people: talented, expensive people who read a value on one screen and type it into another, all day, because the two systems were never connected.
We call it the human API. It is one of the largest unmeasured costs in the modern enterprise — and, because it is spread across hundreds of roles and dozens of systems, one of the hardest to see.
Start with the numbers you already have
Most organisations can estimate three figures from data they already hold:
- The share of time spent moving data between systems — from a week of time study on one team, or from task logs.
- The error rate on keyed fields — from rework queues, QA sampling and downstream corrections.
- The fully loaded cost per hour of the people doing the work.
Multiply them out and you get a labour figure. It is usually large. It is also the smallest part of the cost.
The cost is not the typing
Finance leaders tend to model swivel-chair work as hours multiplied by rate. The larger costs sit elsewhere:
Latency. A claim that waits two days in an intake queue isn’t just two days of slower service. It is a later reserve, a longer cycle, a more expensive settlement and a less happy customer.
Error propagation. A single mis-keyed bank detail can cost more than a year of the salary of the person who keyed it. Most organisations add layers of checking to compensate — which adds more human API.
Capacity ceilings. When volume spikes — month-end, open enrolment, catastrophe season — the only lever is overtime or temporary staff. Both are expensive, slow to arrange and error-prone.
Talent drain. The people doing this work are usually overqualified for it. Attrition in high-volume back-office roles is high, and every departure takes process knowledge with it.
A better way to model it
When we build a business case with an operations team, we look beyond hours saved to four outcome measures they already track:
- Cycle time — from arrival to completion
- Cost per case — fully loaded, including rework
- First-time-right rate — cases that never need touching again
- Elastic capacity — how much additional volume can be absorbed without new hires
Hours returned matter, but they are rarely the reason a programme gets funded twice. The second budget comes from the cycle time and quality improvements that show up in customer and financial metrics.
Where the time should go
Automation sold primarily as headcount reduction tends to stall: the people who know the work have every reason not to teach it. The stronger plan is to reinvest reclaimed capacity in backlog reduction, quality assurance, customer-facing work and growth that would otherwise have required hiring.
The human API is expensive. But the people currently running it are not the cost — they are the opportunity.