What STP is: a complete guide to measuring how much finishes without a human

STP (Straight-Through Processing), also called the no-touch rate, is the share of received cases completed end to end without human intervention, and it is the figure that shows the actual saving from document automation most accurately.

Why to read the no-touch rate rather than accuracy

Accuracy can rise while headcount stays the same.

The figures exchanged in an evaluation meeting are mostly recognition accuracy. Between a value of 98 percent and the hours an operator spends, though, sits a larger gap than expected.

If a case has twenty fields and a person corrects even one, that case did not finish untouched. Field-level accuracy of 95 percent still leaves a substantial share of cases passing through human hands when counted per case. Add system entry after recognition and the human work continues regardless of accuracy. The number that describes the real saving is this one, counted by case.

What STP actually is: how it differs from recognition accuracy

Recognition accuracy and STP across three axes

First, the counting unit differs. Recognition accuracy counts characters or fields. STP counts business cases. Twenty fields in a case with one wrong gives 95 percent field accuracy and a failed straight-through result for that case.

Second, what they reflect differs. Recognition accuracy reflects model performance alone. STP reflects the validation rules, threshold settings, and exception design as well. The same model produces very different STP depending on how the operation was designed.

Third, where they apply differs. Recognition accuracy is for comparing products. STP is for explaining return on investment. The number needed when presenting savings to an executive is the second one.

The two are not alternatives. Low recognition accuracy caps STP. But high recognition accuracy does not carry STP along with it automatically.

Four criteria for measuring STP accurately

Four criteria that hold in measurement

First, define the denominator. Whether it is every received case or only the forms designated for automation changes the value considerably. Excluding out-of-scope documents reports higher than reality.

Second, decide what counts as human intervention. Whether only corrected values count, or whether opening the screen to check also counts, has to be settled. Checking alone still consumes time.

Third, decide how to count reprocessed cases. Whether a case routed to exceptions, corrected, and returned to the flow counts as a success or a failure changes the figure.

Fourth, fix the period. Measure monthly, and where certain documents concentrate in particular seasons, record that effect alongside.

How to raise STP in practice

Group failures by cause

Collecting failed cases by type shows where to intervene. A form problem points at the field definitions, a field problem at the extraction criteria, a rule problem at the threshold.

Adjust validation rules and thresholds together

Tight validation reduces errors but raises exceptions, which lowers STP. Loose validation raises STP but lets wrong values through.

The two have to be adjusted together. Working from review history, relaxing rules that never actually caused a problem and adding rules where errors occurred is the safe approach.

Read it separately from the completion rate

Sometimes the no-touch rate is high and the operator's workload has not fallen. The judgment finished without human hands, but the result never reached the business system.

So read two figures separately. The no-touch rate asks whether a person intervened up to the point of judgment. The completion rate asks whether the case finished through to system registration. A wide gap between them means the integration path is broken, which points at integration rather than recognition.

STP in the Korean environment

In Korean finance and the public sector, internal policy frequently leaves segments where human confirmation is required. Even where automatic processing is technically possible, a case needing regulatory approval did not finish untouched.

When measuring STP, therefore, record confirmation required by policy separately from review driven by quality. The first is a segment technology cannot remove; the second is the target for improvement. Mixing them obscures where the room for improvement actually is.

Frequently asked questions