Straight-through processing (STP)

Straight-through processing (STP) is the ability to take a transaction or document from intake to completion without a human touching it at any point. In document processing, an STP case is one where extraction, validation, and downstream action all happen automatically and the result is trusted enough to post, pay, or approve without review. STP rate, the share of documents fully handled this way, is usually the single most-watched metric in an IDP deployment.

How it works

STP is not a feature you buy; it is an outcome that depends on several things working together. First, the system has to correctly classify the incoming document (see document classification). Second, it extracts the relevant fields, whether that is key-value pairs on an invoice or line items in a table (see key-value extraction and table extraction). Third, it validates those fields against business rules and external data: does the total match the sum of lines, does the vendor exist in the ERP, does the policy number resolve to a real customer.

The step that decides whether a document reaches STP or gets routed to a human is the confidence score. Each extracted field carries a score reflecting how sure the model is. If every field on a document clears the threshold set by the business, and the validation rules pass, the document proceeds automatically. If any field falls short, or a rule fails, the document goes to a review queue instead.

Because thresholds are configurable per field and per document type, STP rate is a tuning exercise, not a fixed number. Set thresholds too loose and error rates climb; set them too tight and humans end up reviewing documents that were actually correct. Good IDP platforms let teams see both effects and adjust deliberately, rather than guessing.

Why it matters for document processing

STP rate is what actually drives return on investment. A system that extracts data accurately but still requires someone to check every document has automated the easy part and left the expensive part, human review, untouched. The gains show up in three places: lower cost per document, faster cycle times (an invoice or claim that would sit in a queue for days gets processed in minutes), and more consistent decisions, since automated rules apply the same way every time.

It also changes how teams should think about accuracy. A model that is 95% accurate but never tells you which 5% is wrong is less useful for STP than one that is 92% accurate and flags its own uncertainty reliably. For processes like e-invoicing or KYC onboarding, where volume is high and errors are costly, STP rate paired with a trustworthy confidence score matters more than raw extraction accuracy alone.

Related terms

• Confidence score
• Data extraction
• Document classification
• Intelligent Document Processing (IDP)
• KYC onboarding

FAQ

What is a good STP rate?

There is no universal number: it depends on document complexity, the cost of an error, and how the process is defined. A simple, structured document type (like a standard invoice format from a repeat vendor) can reasonably target a high STP rate. Complex or highly variable documents, like claims or contracts, will always need more human review, and that is not a failure of the system.

Does higher STP rate always mean better performance?

No. STP rate has to be read alongside accuracy on the documents that do go straight through. A high STP rate built on loose confidence thresholds can quietly push errors downstream instead of catching them. The right way to evaluate a system is to look at STP rate and post-STP error rate together, not STP rate on its own.