There’s no single “right” cost per transaction in used-car operations. A high-volume vehicle-purchasing platform running roughly 1,000 evaluations a week has a cost structure built for standardized, repeatable decisions. A franchise dealer processing similar volume, but running title work, F&I, and compliance on every deal, does not. Blend the two into one industry average and you get a number nobody can actually act on.
That’s the trap. Operators pull a “cost per unit” or “cost per transaction” figure from a vendor deck, a trade publication, or an informal peer conversation, hold it up against their own P&L, and conclude they’re bloated or lean without ever checking whether the comparison is apples to apples. It usually isn’t.
Why does a single benchmark break down?
Cost per transaction is a ratio: total operating cost for a process, divided by transaction volume. The math is simple. What varies enormously is what sits in the numerator, and that’s a direct function of the operating model.
A purchasing platform’s “transaction” is an instant-offer evaluation: a vehicle comes in, a limited set of documents gets checked (title, lien status, ownership, condition), and a decision comes out. The process is narrow by design, built to run at volume with a small, specialized team.
A franchise dealer’s “transaction” is a full retail deal: the deal jacket, the title jacket, F&I products, compliance disclosures, odometer statements, sometimes a trade-in with its own lien payoff. The document count per transaction is higher, the compliance surface is wider, and more departments touch the file before it’s done.
A wholesale or remarketing desk’s “transaction” is different again: arbitration exposure, condition-report accuracy, and a much lighter documentation load than a retail deal, but a different kind of judgment call on every unit.
Same word, “transaction.” Three different cost structures. A benchmark that averages across them tells you about the average, not about your operation.
What actually drives the difference
Three variables explain most of the spread between operating models, more than volume does:
Documents and compliance touchpoints per transaction. A purchasing platform evaluation might involve about five documents; a retail deal with F&I and compliance review involves substantially more, each one with its own verification step, its own error mode, and its own person or system checking it.
Where judgment lives. In a standardized evaluation, most decisions are rule-based: does the title match, is the lien clear, does the mileage reconcile. In a full retail deal, more decisions require judgment calls that don’t reduce cleanly to a checklist, which pushes cost toward experienced staff rather than junior reviewers.
Capital and carrying cost baked into the transaction. A dealer’s cost per unit often carries floorplan interest and curtailment exposure that a purchasing platform’s evaluation cost doesn’t touch in the same way. That’s a real cost of doing business, but it doesn’t belong in the same bucket as review labor when you’re trying to isolate an operations problem.
None of this means one operating model is more “efficient” than another. It means the two numbers aren’t measuring the same thing, and putting them on the same chart invites the wrong conclusion.
A worked example, illustrative only
Say a purchasing platform runs 1,000 evaluations a week with a review team of six to twelve people, each evaluation touching about five documents. Say a franchise dealer group processes a comparable weekly volume across its rooftops, but each deal moves through F&I, compliance, and title before it closes.
Even if the two operations spent an identical total dollar amount on back-office labor in a given month, the purchasing platform’s cost per transaction would land lower, because its transaction is narrower by design. A COO who benchmarks against that number without adjusting for scope will chase a cost target that was never achievable given what the dealer’s transaction actually requires.
Key insight
That's not the purchasing platform being better run. It's the purchasing platform doing less per unit of work.
This is the mistake we see most often: someone reads a purchasing-platform cost-per-transaction figure in a trade article, applies it to a full-service dealer operation, and concludes the review team is overstaffed. The team isn’t overstaffed. The comparison was wrong from the start.
A more useful way to benchmark
The fix isn’t to abandon benchmarking. It’s to narrow the comparison set to operations with a genuinely similar model, not just similar total volume. That’s the actual value of a NIADA 20 Group: 20 non-competing dealers of comparable size and structure sharing composite financials and operational KPIs monthly. The comparison holds because the operating model is held roughly constant across the cohort. Our 20 Group composite benchmarking guide covers which operations metrics actually get compared this way and which ones get distorted by mixing dealer types in one composite.
For a purchasing platform or a centralized buying center without an equivalent peer cohort, the same principle still applies informally: find operators running the same transaction shape (same document count, same judgment profile, same compliance load) and compare against them, even if it’s a smaller, less formal reference group than a 20 Group.
The broader economic backdrop matters too. Our Dealer Group Operations Economics piece lays out how 2026’s margin pressure, much of it driven by PE ownership pushing headcount reductions across portfolio companies, is forcing more operators to defend cost-per-transaction numbers up the chain, which raises the stakes on getting the comparison right in the first place. If your group is under that specific kind of scrutiny, PE-Owned Dealer Groups and the 2026 Headcount Pressure walks through how that pressure typically shows up in operations first.
It’s also worth separating cost per transaction from gross profit per unit, a related but distinct metric that gets conflated in the same conversations. Gross Profit Per Unit Normalization explains why that number has shifted in 2026 and why it shouldn’t be read as an operations-efficiency signal on its own.
Where AI changes the comparison, and where it doesn’t
AI-assisted document review changes the labor component of cost per transaction, but it doesn’t change which operating model you’re comparing against. In one automotive marketplace’s transaction-processing operation, review time per transaction dropped from about 20 minutes to 1 to 2 minutes, and the error rate fell from roughly 7% to about 1%, after moving document verification and rule-checking into an AI-assisted workflow. The review team went from twelve people to six, with the rest redeployed into growth work rather than laid off. That’s a real shift in the labor line of the numerator.
But that shift happened inside one operating model, a purchasing-style evaluation process. It doesn’t make that platform’s post-automation cost per transaction a fair comparison point for a franchise dealer’s F&I-heavy retail deal, or for a wholesale desk’s arbitration-exposed unit. The rule of thumb we use when evaluating whether a process is worth automating: labor, capacity, and error costs need to add up to roughly $1.2 million a year or more, and the automation should cost no more than about 20% of the value it captures. That threshold is about your own numbers, not about matching someone else’s benchmark.
FAQ
Why is a single cost-per-transaction benchmark misleading? Because the underlying operating model, purchasing platform versus retail dealer versus wholesale desk, drives fundamentally different cost structures even at comparable transaction volume. A “transaction” means a different bundle of documents, compliance steps, and judgment calls in each model, so a blended average measures the mix, not any one operation.
What’s a more useful way to benchmark operations cost? Compare cost per transaction against peers with a genuinely similar operating model, ideally through a 20 Group or comparable peer cohort, rather than against a generic industry-wide average. If no formal peer group exists for your model, an informal reference group of operators running the same transaction shape still beats a headline number pulled from a trade deck.
What this means for your numbers
Before you act on a cost-per-transaction figure from anywhere outside your own walls, ask what’s actually in the numerator: how many documents per transaction, how much of the decision is rule-based versus judgment-based, and whether capital costs like floorplan interest are baked in. If those don’t match your operation, the number isn’t a benchmark. It’s a distraction.
If your document review process is the line item driving your cost per transaction up, and you want a clearer read on where the labor is actually going before you compare yourself to anyone else, Deskflow maps that process end to end so the numbers you’re benchmarking are ones you can actually defend.