Your Cost per Transaction Looks Precise. Is the Allocation Behind It Credible?

Two products share a platform. One looks efficient and the other looks expensive. Finance changes how the shared bill is divided, and their ranking reverses—even though neither workload changed and the company didn't remove a single resource.

That is possible because allocation is a rule for assigning cost, not a direct observation of every product's consumption. A precise cost-per-transaction figure can contain both measured amounts and a choice about how to share the remainder. Before using the ranking to judge performance, the reader needs to know which is which. [1][2]

Watch the ranking change without changing the bill

Here is a hypothetical example. Product A and Product B each have $1,000 of direct cost and complete 10,000 comparable transactions during the period. They also share a platform bill of $6,000. The company's total cost is therefore $8,000.

Suppose A represents 75 percent of the users but only 25 percent of the measured computing use. B represents the reverse. Those are teaching assumptions, not observations about a real service.

The team considers two allocation methods: distribute the shared bill by user count, or distribute it by measured computing use.

Dividing each total by its 10,000 transactions produces the reversal:

Under the first method, B appears cheaper. Under the second, A does. The total remains $8,000 under both methods. The allocation changed, not the underlying bill.

This doesn't demonstrate that either method is right. A platform whose costs mainly track computing use may justify one driver; a service providing substantial per-user capability may justify another or a mixed arrangement. The method needs evidence about the economic question, not a preference for which product should win.

The rule influences what people try to improve

If a team is judged by allocated cost per transaction, it has a reason to improve that number. Under a user-based method, reducing assigned users could improve its reported position without reducing the resources the platform actually needs. Under another method, different behavior receives the incentive.

That doesn't make allocation useless. It can establish accountability for shared services that the provider doesn't bill separately by product. It does mean an allocation movement shouldn't be presented automatically as an engineering saving. [2]

“This product consumes more” and “our method assigns more cost to this product” are different statements. One may support the other when the driver reflects relevant consumption, but the relationship needs to be explained.

Direct, shared, and unresolved costs should remain distinguishable for the same reason. An unassigned remainder doesn't disappear economically because the report omits it. Distributing it arbitrarily makes the view look complete while hiding how uncertain the comparison remains.

Test the assumption that could change the decision

For the two products, the ranking is sensitive to the allocation rule. That is useful information, particularly if someone proposes a redesign based on which product looks inefficient.

I would compare a small number of credible methods and ask whether the preferred action changes. If a proposed improvement remains worthwhile across reasonable allocations, the decision may not need perfect attribution. If the ranking reverses as it does here, a confident performance judgment needs more investigation.

This is a sensitivity test, not an invitation to generate endless scenarios. Its purpose is to identify which assumption matters enough to change the choice. The unit, period, direct-cost coverage, and meaning of a transaction must also stay comparable. [1]

The next step might be to examine workload behavior or improve attribution before committing to a large redesign. It should not be to select whichever rule produces the most convenient winner.

When the method changes, the explanation needs to travel with the numbers. Otherwise a revised cost model can become a story about a team's performance that nobody's work actually supports. An honest range or conditional conclusion is more useful than an exact-looking ranking that reverses when someone chooses a different way to divide the same bill.