Straight-through path
The item is complete, valid and confident enough to proceed on its own, and does.
- Finishes at machine speed
- No reviewer time per item
- Errors are silent unless audited
- Grows with data quality and evidence
STP is the share of work that completes end to end without a person touching it. The number is easy to raise and easy to fake — the craft is raising it while proving the work is still right.
This guide defines straight-through processing precisely, shows what it means in securities settlement, payments, insurance and document processing, gives the published figures from the US move to T+1, and sets out how to calculate an STP rate that survives an audit — then how to raise it one field at a time.
Updated · 6 sources
Straight-through processing (STP) is the completion of a transaction or work item end to end — from capture to settlement, payment or posting — without manual intervention. An item is “straight-through” only if no person reviews, corrects, approves or rekeys it at any step. The STP rate is the share of items that pass this way; the items that stop for a person are exceptions.
Every STP design splits work into two paths. The value comes from making the first path wide and the second path fast — not from pretending the second does not exist.
The item is complete, valid and confident enough to proceed on its own, and does.
Something failed — a rule, a match, a confidence threshold — and a person resolves it.
Trade capture, allocation, confirmation, affirmation and settlement with no manual step. Now written into US rules for central matching providers.
A payment that clears screening, validation and routing untouched. Swift credits ISO 20022’s structured data with enhancing STP.
A claim paid or a policy issued without an adjuster or underwriter touching it — inside an AI governance programme regulators expect.
An invoice, form or statement read, validated and posted with no field reviewed by a person.
When US securities settlement shortened from two business days to one, there was no longer time for manual post-trade steps. Regulators wrote straight-through processing into the rules, and the industry measured its progress publicly.
An STP rate is a fraction, and both halves are easy to bend. Fix the definitions before the first report, and pair the rate with a measure of the errors that went straight through.
Illustrative month: 10,000 invoices received
Touchless (no person at any step) 7,200
Touched (any review, fix or approval) 2,800
STP rate = 7,200 / 10,000 = 72.0%
Blind audit of touchless items: 360 sampled, 4 wrong
Escape rate = 4 / 360 ≈ 1.1%
Estimated errors that went straight through
≈ 1.1% × 7,200 ≈ 80
Clean STP rate = (7,200 − 80) / 10,000 ≈ 71.2%
4 errors in 360 is a thin sample: a 95% Wilson
interval on the escape rate is 0.4%–2.8%, or
about 30 to 200 errors among the 7,200.
Audit more before you quote the clean rate.Lowering a confidence threshold raises STP and lets more errors through; raising it does the opposite. Plotting both tells you which problem you have.
STP rises fastest when inputs get cleaner and when automation is earned per field, with evidence, rather than switched on per process.
Structured messages, standard settlement instructions and validated forms remove exceptions before they start.
A threshold on calibrated confidence, per field, sends only the uncertain values to people. See confidence calibration.
Keep a blind sample of straight-through items and stop automation when the audited error rate climbs.
On Botminds you choose the fields that may go straight through, then set each threshold — or state the accuracy you need and let the platform derive the threshold from evidence.
Pick the fields that may go straight through; every other field goes to a person.
Give a threshold, or a target such as “at least 99% right”; the platform finds the lowest threshold it can certify at 95% confidence.
A document passes when every chosen field passes; a critical field left empty sends it to review.
A share of passing documents goes to a person, shown without scores.
With an accuracy target set and audits running, a climbing audited error rate trips an alarm that switches straight-through off for the field until a person restores it.
Each value carries a verdict — passed or needs review — and a pass is withdrawn if the value is edited, the threshold is raised or a validation rule breaks. Read human in the loop for the full design.
Straight-through processing means a transaction or document goes from arrival to completion with no person touching it — no review, no correction, no approval, no rekeying. Items that need a person are exceptions.
The share of items completed with zero manual touches, out of all items received in the same period. Report it with an escape rate — errors found in a blind audit of straight-through items — so a higher rate cannot hide more mistakes.
Automating a trade or payment end to end: capture, validation, enrichment, matching, confirmation and settlement without manual steps. US rules for T+1 settlement require central matching providers to facilitate STP and report on it.
Claims settled or policies issued without an adjuster or underwriter touching them — usually simple, low-value or low-risk cases — while the rest go to people. Insurers using AI for these decisions are expected to run them under a written AI governance programme.
It depends on the process and its inputs, so compare like with like. DTCC’s recommendation for trade affirmation was at least 90% by the trade-date cut-off; for documents, the right target is the highest rate at which audited accuracy still meets your requirement.
Automation can cover part of a process with a person still reviewing in the middle. STP means the whole item completes without any person; a process can be heavily automated and still have a low STP rate.
By reading documents into validated data, scoring each value’s confidence, and letting values that clear a calibrated threshold proceed — while sending the uncertain ones, and a blind audit sample, to people.
Bring one process and its documents. We’ll show you the agents, the controls and the record, running on your data.