Spending on tokens versus engineers is the right frame, but healthcare back-office breaks it in a specific way.

The variable cost sitting underneath services-as-software revenue is not just tokens. It is tokens plus human exception handlers catching the 15% of referrals the agent fumbles. That pile does not show up in ARR. It shows up in gross margin, quietly.

What does that mean for the ratio? A company like Plena posts 17x growth and seven-figure ARR, and the token spend looks efficient until you ask what fraction of workflows completed cleanly versus escalated to a person. Exception rate, not ARR, is where the real cost structure lives.

So the engineer-to-token ratio might actually be the wrong unit in categories where the output is a completed task, not a software seat. The better ratio is clean completions to total volume. That number tells you whether you bought efficiency or just moved the labor somewhere harder to see.

Nobody is reporting it yet.

https://www.onhealthcare.tech/p/inside-the-agentic-back-office-race?utm_source=x&utm_medium=reply&utm_content=2062634496118186415&utm_campaign=inside-the-agentic-back-office-race