Straight-through processing targets are the most useful metric in back office operations and the easiest one to game. The definition is simple enough to fit in a sentence — the proportion of transactions that complete from receipt to posting without a human touching them — and that simplicity is exactly why the number gets abused. Once a target is set, the fastest route to hitting it is usually to change what counts as a touch. The metric became standard in back office scorecards during the mid-2010s, borrowed from payments and securities settlement where it had been in use for decades. In banking, straight-through processing had a precise meaning tied to message standards and settlement chains. In accounts payable, order management and employee administration, everyone invented their own definition, which is why cross-company benchmarks in this area are close to worthless.
What the number is actually measuring
The honest version measures how much of your transaction volume requires no human decision, no human correction and no human intervention of any kind. That is a harder standard than most organisations apply. A common definition counts a transaction as straight-through if nobody manually keyed it, while quietly excluding the person who reviewed the extraction output, the approver who clicked through a workflow step, and the analyst who fixed the supplier master record that caused the match to fail. Under that definition, a process where a human touches every item can still report eighty percent. The second common distortion is scope. Excluding categories that are known to be difficult — intercompany transactions, foreign currency invoices, anything from a particular subsidiary — raises the reported number without changing the work anyone does. The exclusions are usually defensible individually and collectively meaningless. The diagnostic question that cuts through all of it: if the straight-through rate is eighty percent, is the team twenty percent of its original size? If headcount has not moved, the number is measuring something other than automation.
Why the last twenty percent is the whole job
The distribution of difficulty in these processes is heavily skewed. Getting from nothing to sixty or seventy percent is mostly a matter of handling the clean, well-formed, high-volume cases — the electronic invoice from a regular supplier against an open purchase order with matching quantities. That work is genuinely straightforward and it is where every programme starts. What remains is everything that is wrong: the invoice with no purchase order, the quantity mismatch, the supplier who changed their bank details, the credit note referencing an invoice from two years ago, the tax treatment that the rules engine has no case for, the scanned document that is a photograph of a screen. Those exceptions consume the majority of the effort and generate the majority of the risk, and they are also where the actual business problems live. An exception queue is a diagnostic report on the rest of the organisation: it shows you which suppliers were onboarded badly, which buyers raise orders after the fact, and which master data nobody maintains. This is why exception analysis matters more than the headline rate. Categorising exceptions by root cause and volume — and then fixing the causes upstream rather than processing them faster downstream — is the difference between a programme that keeps improving and one that plateaus at seventy percent and stays there for five years.
Setting targets that mean something
A target rate with no denominator definition is not a target. The elements that need to be agreed before a number is chosen: The unit being counted, and what falls in and out of scope. Every exclusion documented and justified. What counts as a touch. The defensible standard is any human action — including review, approval and correction — with a separate metric for approvals that are required by policy rather than by system failure, since a segregation-of-duties approval is a control, not an automation gap. The quality constraint. A straight-through rate improved by lowering matching tolerances or removing validation is a fraud risk disguised as an efficiency gain. Pair the rate with error rate, duplicate payment rate and post-payment recovery volume, and review them together. And a realistic ceiling. For most accounts payable operations, a high but sub-total rate is the right ambition; certain processes will never reach it because the inputs are genuinely unstructured or the decisions are genuinely judgemental. Setting a target of ninety-five percent on a process where forty percent of inbound documents arrive as email attachments from small suppliers guarantees either failure or gaming.
| Definition to fix | Companion evidence |
|---|---|
| Transaction unit and scope | Document exclusions and changes in volume mix. |
| Human touch | Distinguish policy approval from review or correction after failure. |
| Quality constraint | Review errors, duplicate payments and recovery findings beside the rate. |
| Exception category | Track root cause and upstream ownership, not just queue speed. |
| Category-specific target | Use the actual input mix and retain required controls. |
Qualitative summary of this article's source text, not a measured outcome or performance estimate.
Practical Guidance for STP Target Setting Workshop
- Define a touch before you define a target. Write down exactly which human actions disqualify a transaction, and get finance, operations and audit to agree on it in the same room.
- Document every scope exclusion and review it annually. Exclusions are how targets get hit without work getting easier. If a category is excluded permanently, it should have its own improvement plan.
- Always pair the rate with a quality metric. Error rate, duplicate payments, recovery audit findings. A rising straight-through rate with rising errors is a control failure in progress.
- Categorise exceptions by root cause and rank them by volume. This single analysis usually reveals that three upstream problems generate most of the manual work.
- Fix causes upstream rather than processing exceptions faster. Supplier onboarding data quality, purchase order discipline and master data ownership deliver more than any downstream tool.
- Separate policy approvals from failure-driven intervention in the reporting. Conflating them either penalises necessary controls or hides genuine automation gaps.
- Set differentiated targets by transaction category. One rate across a heterogeneous population tells you nothing and encourages the team to work on the easiest segment.
- Recalculate the baseline whenever volume mix changes materially. Acquisitions, new markets and new supplier categories shift the denominator, and comparing against a stale baseline produces false progress.
The Regional Dimension
Straight-through processing targets imported unchanged into a Gulf operation tend to look unachievable, and the reasons are structural rather than a reflection of operational maturity. Document format is the first. A meaningful share of supplier invoices across the region still arrives as a PDF attachment, a scan, or a photograph sent over a messaging app, frequently from smaller suppliers with no electronic invoicing capability. Extraction quality on those inputs is materially worse than on structured electronic documents, and any target copied from an operation running mostly structured invoices will be wrong by a wide margin. Bilingual data is the second, and it is underestimated. Invoices carry Arabic and English text, sometimes both on the same document. Supplier names transliterate inconsistently, which creates duplicate master records, which breaks automatic matching, which sends transactions to the exception queue for a reason that has nothing to do with the invoice. Name normalisation and duplicate resolution in the supplier master frequently moves the straight-through rate more than any change to the matching engine. Regulatory change cuts both ways. E-invoicing under the Saudi regime, with its clearance and reporting requirements, is a compliance obligation that has the side effect of standardising document format — which raises the achievable ceiling considerably for Saudi transaction flows. UAE VAT and corporate tax have added validation requirements that increase exceptions in the short term and structure in the longer term. Organisations operating across both markets should expect the achievable rate to differ by country and should set targets accordingly rather than averaging them. Multi-entity structures add a final complication. Intercompany transactions between a mainland company, free-zone entities and a Saudi subsidiary are usually the largest single exception category in regional groups, and they are almost always excluded from the reported rate. That exclusion is worth challenging — intercompany is repetitive, high-volume and rules-based, which makes it one of the better automation candidates once someone owns the reconciliation design. And payments deserve a specific note: supplier bank detail changes are the highest-risk exception in any regional operation, given how common payment redirection fraud has become. This is one queue that should never be optimised for speed.
The objection worth taking seriously
The strongest criticism is that straight-through processing is an input metric masquerading as an outcome, and that optimising it can actively damage the business. Consider what the metric rewards. It rewards paying invoices without anyone looking at them. In a well-controlled environment with reliable three-way matching, that is fine. In an environment where the purchase order data is unreliable, it means paying for goods nobody confirmed receiving, faster and at higher volume. Several organisations that pushed rates aggressively discovered exactly this during a recovery audit. It also creates pressure against judgement. An experienced accounts payable analyst who notices that a supplier's invoice pattern has changed is, in the language of the metric, a touch that lowers the score. Removing that person from the flow improves the number and removes a control that no rules engine replicates. The balanced position is that straight-through processing is a good metric when it is one of three or four, and a dangerous one when it stands alone on a scorecard with a bonus attached. The right framing is not "how much can we process without humans" but "which transactions genuinely require a human decision, and are those the ones getting human attention." An operation where ninety percent runs automatically and the remaining ten percent receives careful scrutiny from skilled people is a better outcome than ninety-five percent with a queue nobody has time to examine.
Common Questions
What is a good straight-through rate for accounts payable?
There is no portable benchmark, because definitions vary so widely that cross-company comparison is unreliable. The useful measure is your own trend on a fixed definition with a fixed scope. If an external benchmark is quoted to you, the first question should be what they count as a touch.
Should approvals count against the rate?
Report them separately. A policy-mandated approval is a control you have chosen to keep; a review triggered by a matching failure is an automation gap. Combining them means you cannot tell whether an improvement came from better automation or from weakened controls.
Where should a first improvement effort focus?
On exception root causes, ranked by volume. In most operations a small number of upstream problems — supplier master data quality, missing purchase orders, inconsistent unit of measure — generate the majority of manual handling, and none of them are fixed by buying a better matching engine.
Does AI-based document processing change the ceiling?
Yes, meaningfully. Models that read unstructured and bilingual documents handle the input variety that defeated template-based extraction, which raises the achievable rate in exactly the segment that used to be excluded. The governance requirement changes with it: a confidence threshold policy, sampling of automated decisions, and a clear record of what the system extracted versus what was on the document. Higher automation with no assurance layer moves risk rather than removing it.
STP Target Setting Workshop — agree what counts as a human touch before agreeing the target, because every straight-through programme that skips that conversation ends up measuring definitions instead of work.