At $50M and above, service cost stops being a line you can read. It is spread across payroll in four offices, absorbed into producer compensation, buried in the overtime that renewal season generates, and sitting in a headcount number that grew for reasons nobody wrote down.
Ask a CFO what service costs and you will get a defensible answer. Ask what a certificate costs, or which office runs renewals most efficiently, or whether last quarter’s two new hires actually absorbed the volume they were hired for, and the answer is usually an estimate.
That gap is the subject of this piece. Not that service is expensive, which every operator already knows, but that at enterprise scale it is expensive in places the P&L does not show you.
The Four Places It Hides
Inside producer compensation. When a senior producer spends part of the week on service work, that cost lands in production expense, not service expense. The P&L shows a well-paid producer. It does not show that a meaningful share of that salary bought certificate processing. This is the largest hidden cost in most large agencies and the hardest to see, because nothing about it looks wrong.
In the offices you do not compare. Four offices doing the same work four different ways produce four different cost profiles, and a consolidated P&L averages them into one number that describes none of them. The efficient office subsidises the struggling one, and neither shows up until someone leaves and the difference becomes visible.
In rework. A task done twice costs twice and appears once. Endorsements that come back, certificates issued against stale data, renewals requoted because the first pass used the wrong information. None of it is recorded as rework. It is recorded as work.
In the data underneath. This one surprises people. Across the books consolidated through COVU Connect, 57% of raw customer records were duplicates or fragments, and roughly 1 in 2 customers existed in more than one system. On the same policy held in two systems, the carrier disagrees 32% of the time. Every one of those discrepancies becomes someone’s afternoon, and that afternoon is charged to service payroll rather than to the data problem that caused it.
| Where it hides | Why the P&L misses it | What makes it visible |
|---|---|---|
| Producer time on service | Booked as production expense | Route service work off producers, then measure what is left |
| Cross-office variation | Consolidation averages it away | Compare offices on the same task definitions |
| Rework | Recorded as work, not as rework | Track a task from open to close, including reopens |
| Data discrepancies | Charged to payroll, caused elsewhere | Reconcile the book so the work stops being generated |
Why Headcount Is the Wrong Lever
The instinct at enterprise scale is to manage service cost through headcount, because headcount is the number the P&L makes easy to see.
It is a blunt instrument. Cutting heads reduces cost and capacity in the same motion, which is why the saving rarely survives the next renewal season. Adding heads absorbs volume without changing the cost per unit of work, so the ratio drifts in the wrong direction while the absolute number looks under control.
Neither move touches the thing that determines the cost: how the work is defined, who it routes to, and how often it has to be done twice.
What Visibility Actually Requires
Cost per task is not a report you can generate from an AMS. The AMS records the policy. It was never designed to record who did what, how long it took, or whether it came back.
Three things have to be true before the number exists.
- The work has to be decomposed. A renewal is not a unit of cost. It is a sequence of tasks with different requirements and different costs. In COVU OS a certificate playbook runs 6 discrete tasks, and a new customer acquisition runs up to 16. You cannot cost what you have not defined.
- Every task needs an owner and a record. Not who owns the account, but who did this piece of this job, when, and whether it was completed or handed on. Every action, human or AI, logged with actor, timestamp, and reason.
- The definitions have to be the same across offices. Otherwise you are comparing four different measurements and calling it a benchmark.
Once those hold, the cost per task falls out of the system rather than being estimated in a spreadsheet. So does the answer to which office is actually more efficient, and why.
What the Production Data Shows
Across the agencies we operate:
- AI triage filters roughly 68% of inbound as noise, and classifies each item in under 10 seconds at a 98% triage success rate. That volume never reaches a person, so it never reaches payroll.
- Escalations, meaning a task where someone is stuck and has to hand it on, are down 20%, from 435 a day to 348, and still declining. Escalations are rework in its most expensive form.
- Median handle time on a task is 3.8 minutes, down 16% from 4.5.
- Roughly 44% of completed service tasks are now executed by AI.
- The platform routes more than 40,000 tasks a week with no manual intervention.
Cost per task is measured, and it comes down. Across the agencies we operate it has been cut roughly in half since launch. We do not publish the figure itself, because our internal cost is not a price, and the number that matters to you is the one from your own book.
Service is your largest cost line. COVU OS is the system that runs it like one, on top of your AMS rather than instead of it, with 8 AMS integrations live.
S&G Mitchell went from 17.9% to 60%+ EBITDA in 12 months. Same agency, same customers, same book of business. What changed was the operating model underneath.
Start by Measuring One Workflow in One Office
The enterprise version of this is not a platform migration. Take one high-volume workflow, certificates or endorsements, in one office. Define the tasks. Route them. Measure what each one costs and how often it comes back.
You will learn two things quickly: what that workflow actually costs you, and how far the other offices are from the same number. Both are more useful than another headcount conversation.
SEE WHAT SERVICE ACTUALLY COSTS
Walk through one workflow in one office and see the cost per task come out of the system rather than a spreadsheet.
Frequently Asked Questions
Why is service cost so hard to see in a large agency P&L?
Because it is not in one place. It sits inside producer compensation when senior people do service work, it is averaged across offices that operate differently, and rework is recorded as work. The P&L shows a total that is accurate and not actionable.
What is cost per task and why does it matter at enterprise scale?
It is what it costs to complete one unit of service work. It matters because it is the only service number that is comparable across offices, across quarters, and against volume. Headcount tells you what you spend. Cost per task tells you whether the operation is getting better.
Can we get cost per task out of our AMS?
Not usually. The AMS is the system of record for policies. It was not designed to record who did which piece of work, how long it took, or whether it was reopened. That measurement lives in the layer above the AMS.
Does reducing service cost mean cutting service staff?
No, and cutting heads usually fails, because it removes cost and capacity together. The durable change is in how work is defined and routed, so the same team absorbs more volume and the most expensive people stop doing the cheapest work.
Where should a multi-office agency start?
One workflow, one office. Define the tasks, route them, measure cost and rework. Then compare that office against the others on the same definitions, which is usually where the real variance shows up.