The Real Cost of a 20-Person Document Review Team

If you run a document review operation, you already know roughly what it costs: it’s on the payroll report. But payroll is the smallest of the three numbers that matter, and budgeting off it alone is how review teams quietly become one of the most expensive functions in the company.

Here’s the honest math, using a 20-person team as the working example. The specific figures are illustrative assumptions (swap in your own), but the structure of the calculation is the point.

Bucket one: loaded labor

Take 20 full-time reviewers at a fully loaded cost of roughly $85,000 each: salary plus benefits, payroll taxes, software seats, management overhead, and workspace. That’s about $1.7 million per year before anyone reviews a single document.

Add the costs payroll doesn’t capture. Review roles in judgment-heavy operations turn over, and each replacement carries recruiting cost plus months of below-full productivity while institutional knowledge gets rebuilt one escalation at a time. A team of 20 with typical operations turnover is effectively re-training a meaningful slice of itself every year.

Bucket two: error cost

Manual review of multi-document transactions produces errors at a fairly predictable rate. In operations we’ve measured directly, experienced teams ran around a 7% error rate, not because the people were bad, but because humans doing repetitive verification under queue pressure make mistakes.

Each error costs something: rework hours, a transaction reopened, a customer contacted again, and in underwriting-style decisions, direct financial leakage: a deal approved that shouldn’t have been, or killed when it was fine. If your team touches 50,000 transactions a year and 7% need some form of rework or correction, that’s 3,500 flawed outcomes annually. Even at a conservative $100 of fully-costed impact each, that’s another $350,000 per year, and in high-stakes decisions the per-error cost is usually far higher.

Bucket three: the capacity you never had

This is the bucket most cost analyses skip, and it’s often the largest. A manual review team is a throughput ceiling. When demand exceeds it, you don’t just pay overtime: you delay revenue, lose transactions to slower cycle times, and shelve growth initiatives because the back office can’t absorb them.

We saw this concretely at an automotive marketplace running about 1,000 vehicle evaluations per week through a manual review team: their expansion of mobile purchasing units was constrained not by demand or by field staff, but by how fast documents could clear the queue. The queue was the growth ceiling. What is your review team’s ceiling costing you in transactions you didn’t do?

Adding it up, and two rules for what to do next

For our illustrative 20-person team: roughly $1.7M in loaded labor, on the order of $350K or more in error and rework cost, plus a capacity ceiling whose value depends on your growth situation but is rarely zero. Call it $2M+ per year of economic weight for a team of this size, before counting the opportunity cost.

Two vendor-agnostic rules for deciding what to do with that number.

The $1.2M rule. A process is a serious automation candidate when the total quantifiable opportunity (labor plus error reduction plus incremental capacity) reaches about $1.2 million per year. Below that, the project can still work, but the economics get thin once you account for implementation effort, change management, and the exceptions humans will still handle. A 20-person review team clears this bar easily; a 3-person team usually doesn’t, and that’s fine.

The 20% rule. Whatever you deploy, whether you build, buy, or hire a partner, the all-in cost should not exceed roughly 20% of the value it captures. If the opportunity is $2M a year and a vendor’s proposal works out to $1M a year, the math doesn’t work no matter how good the technology is. This rule also protects you from the opposite failure: a tool cheap enough to buy without scrutiny but too shallow to move any of the three buckets.

What “good” looks like

For calibration: in the automotive operation above, converting the review process to AI operation, where documents come in, data is extracted and verified, rules are applied, decisions go out, and exceptions go to humans, cut per-transaction review time from about 20 minutes to 1–2 minutes for cases humans still see, dropped the error rate from roughly 7% to roughly 1%, and let the company halve the review team while redeploying the rest to growth work. All three buckets moved at once, which is what made the economics decisive rather than marginal.

Run the three-bucket math on your own operation before talking to anyone, us included. If the number clears $1.2M, you have a real decision to make. If you want to see the full before/after from the case above, it’s in our AI Deal Engine case study.