BY NUMBER7 AIFormal definition
For a defined AP workflow:
Eligible-population STP = eligible in-scope transactions reaching the terminal state without human intervention ÷ all eligible in-scope transactions × 100
For transparency:
All-received STP = touch-free transactions reaching the terminal state ÷ all genuine in-scope transaction attempts received × 100
The rates answer different questions. Eligible STP measures performance inside a declared operating envelope. All-received STP shows how much of the actual business workload reached the outcome without people.
Define “through” before measuring “straight-through”
AP has many possible terminal states:
- document classified;
- fields extracted;
- invoice validated;
- invoice approved;
- accounting-ready transaction created;
- bill posted to the target ledger;
- payment state recorded;
- payment executed and reconciled.
A system can achieve high extraction STP and low posting STP. Both can be valid metrics if named accurately. The misleading move is to label an upstream stage “end-to-end.”
Number7 recommends a complete metric name:
Eligible invoice-receipt-to-verified-bill-posting STP, September 2026, supported QBO workflow.
The label tells the reader what began, what finished, when it was measured, and where the product boundary sits.
The frozen-denominator protocol
Write and version these rules before the run:
| Rule | Decision to record |
|---|---|
| Transaction unit | Economic payable, invoice, credit note, or another explicit unit |
| Entry event | First controlled receipt, not later system acceptance |
| Terminal state | Observable, verified outcome in the intended target system |
| Follow-up window | How long transactions are allowed to reach the terminal state |
| Supported envelope | Document types, channels, languages, entities, integrations, and size limits |
| Legitimate exclusions | Test files, spam, exact transport duplicates, corrupt files with no recoverable business content |
| Unsupported business inputs | Retained in all-received reporting; optionally excluded from eligible rate if predeclared |
| Human touch | Correction, decision, evidence supply, manual route, manual post, or required verification |
| Retry rule | Automatic idempotent retry may remain touch-free; human-triggered repair does not |
| Reopen rule | A corrected/reopened transaction loses touch-free status for the measured outcome |
The eligibility manifest should be published with a benchmark or available to the buyer. “We excluded unsuitable documents” is not enough.
Worked example: two STP rates, one honest story
An illustrative AP operation receives 1,100 genuine, in-scope invoice transaction attempts. Before the run, it declares 1,000 eligible for the supported workflow and 100 genuine business inputs outside the current product envelope.
Of the 1,000 eligible transactions:
- 760 reach a verified bill-posting state without a person;
- 80 appear automated upstream but require a human correction, verification, or posting repair;
- 160 require other human intervention before the terminal state.
The results are:
| Metric | Calculation | Result |
|---|---|---|
| Eligible-population STP | 760 ÷ 1,000 | 76.0% |
| All-received STP | 760 ÷ 1,100 | 69.1% |
| Eligible transaction-touch rate | 240 ÷ 1,000 | 24.0% |
| Unsupported-input share | 100 ÷ 1,100 | 9.1% |
Reporting only 76% would be incomplete for workforce capacity planning. Reporting only 69.1% would hide how well the supported envelope performs. Publishing both makes the boundary explicit.
This example is illustrative, not a Number7 or market benchmark.
Stage fallout explains why STP is lost
Track the same cohort through stages:
| Stage | Cohort remaining touch-free | Main fallout examples |
|---|---|---|
| Received and classified | 940 | Unsupported type, ambiguous boundary, corrupt source |
| Extracted | 900 | Low-confidence critical field, line structure |
| Identity resolved | 850 | Vendor/entity ambiguity, master-data gap |
| Evidence/control satisfied | 790 | Missing receipt, variance, duplicate candidate, approval |
| Accounting-ready | 775 | Coding/tax/entity decision |
| Posted and verified | 760 | Integration error, timeout, duplicate-safe retry, target rejection |
The final rate is important, but stage fallout tells the team what to improve. It also prevents an extraction vendor and an end-to-end workflow vendor from appearing comparable when they measure different terminal states.
What cannot enter the numerator
- an invoice corrected by an operations team before the buyer sees it;
- a bill manually re-keyed after an integration failure;
- a transaction that reached a target system but was later reopened for an error related to the measured flow;
- an item approved by a person when zero-touch approval is part of the claim;
- a failed or parked transaction merely because the reporting window ended;
- an unsupported business document removed after observing the result.
If a person invisibly rescued the transaction, the system did not process it straight through.
Counterexample: 95% extraction STP, 40% posting STP
A platform reads 95% of invoices without field correction. Half of those still need vendor resolution, evidence, coding, approval, or posting repair. Advertising “95% STP” without the terminal state encourages a buyer to model labor savings that will not occur.
The correct claim is “95% touch-free extraction within the tested field and document scope.” The posting outcome must be measured separately.
How to compare two STP claims
Ask for:
- transaction unit;
- start event and terminal state;
- all-received count;
- eligible count and exclusion manifest;
- touch definition;
- manual rescue and reopen treatment;
- product version and integrations;
- sample composition and measurement period;
- silent-error sampling method;
- stage-level fallout.
If those fields differ, the percentages are not directly comparable.
Limitations
High STP is not automatically safe or valuable. A system could increase STP by weakening controls, auto-approving low-quality proposals, or narrowing eligibility. STP does not measure material error, duplicate exposure, fraud, cycle time, supplier experience, or accounting quality.
Some organizations intentionally require human approval for all items. In that workflow, zero-touch posting may be the wrong goal. Measure the automated path to the control boundary and the efficiency of the required decision instead.
Sources
- IBM: What is straight-through processing? — broad definition of completing transactions without manual intervention.
- Number7 Research protocol: frozen-denominator AP STP, revision 1.0.
How to cite
Number7 Research. “How to Calculate Straight-Through Processing in Accounts Payable.” Number7AI, revision 1.0, 7 September 2026. https://number7ai.com/research/straight-through-processing.
Revision history
| Revision | Date | Change | Reviewer |
|---|---|---|---|
| 1.0 | 7 September 2026 | Initial dual-rate definition, denominator protocol, and stage-fallout model | Number7 AI editorial review |
Next step
Require every internal and vendor STP figure to carry its terminal state and denominator manifest.