Every vendor demo of a commercial real estate document pipeline sells the same number: the straight-through processing rate, the share of documents that go from PDF to structured field with no human touch. It is the wrong thing to optimize first. Straight-through processing is a lagging indicator of a workflow whose real economics live in the other bucket, the exceptions. A pipeline that hits 85% straight-through and handles the remaining 15% badly is slower, riskier, and more expensive than one at 70% that routes exceptions cleanly. Design the workflow around the exception path. The straight-through rate will follow.
The reason is arithmetic. If your straight-through processing rate is 62%, the other 38% falls into exception handling, where a person opens the document, finds the disputed field, and resolves it. According to industry benchmarks reported by KlearStack, most organizations land between roughly 26% straight-through for hard international transaction sets and 67% for best-in-class implementations. The exception bucket is not a rounding error. It is where most of the human hours, and most of the errors that reach the model, accumulate.
Key Takeaways
Straight-through processing rate is the share of documents that go from input to structured output with no human intervention. It is a useful headline but a poor design target.
The exception path, not the straight-through path, holds most of the labor cost and nearly all of the risk in a CRE extraction workflow.
Teams with straight-through rates above 70% process documents at roughly one-third the cost of manual operations, per Ardent Partners, but most mid-market finance teams still sit below 60%.
A high straight-through rate hides a bad exception path. The two numbers must be read together, never one alone.
The right first question is not "how many documents auto-process," it is "when one does not, how fast and how safely does a human resolve it."
What is straight-through processing in CRE document extraction?
Straight-through processing is the percentage of documents that move from raw input to validated structured output with zero human intervention. In commercial real estate, that means a lease, rent roll, or offering memorandum is read, its fields extracted, validated against rules, and written to the system of record without anyone opening it. Everything else is an exception.
The metric came from payments and banking, where it measured trades that settled without manual repair. It transferred cleanly to document automation because the underlying question is identical: what fraction of the work required no person. As IBM notes in its treatment of the term, straight-through processing is fundamentally about removing manual touchpoints from a sequence of steps. For CRE extraction, those steps are ingestion, document extraction, validation, and write-back. A document is straight-through only if it clears all of them untouched.
The trap is treating that single number as the whole scorecard. A pipeline can raise its straight-through rate by lowering its confidence threshold, which pushes weak extractions past review and into the model as silent errors. The rate goes up. The quality goes down. This is why the straight-through rate must always be read alongside the exception rate and the exception resolution time, never in isolation.
Why does the exception path hold the real cost?
The exception path holds the real cost because exceptions consume human time at a rate that dwarfs the marginal cost of an auto-processed document. An auto-processed lease costs pennies of compute. An exception costs an analyst's minutes: opening the file, locating the clause, reconciling the value, and re-keying it. Fifteen percent of documents can absorb most of the payroll.
Consider a portfolio team abstracting 1,000 leases. Suppose the straight-through rate is 80%, so 800 documents clear untouched and 200 route to exception handling. If each exception takes an analyst 12 minutes to resolve, that is 2,400 minutes, or 40 hours, one full analyst-week, spent on one-fifth of the documents. Now improve the exception path so each takes 6 minutes instead of 12. You have cut 20 hours without touching the straight-through rate at all. The same halving applied to the 80% that already auto-processed would save nothing, because that work already costs almost nothing.
This is the compounding gap. Firms that engineer the exception path buy back analyst weeks that scale with volume. Firms that chase only the headline rate optimize the cheap 80% and leave the expensive 20% untouched.
Metric | What it measures | Why it can mislead |
Straight-through rate | Share of documents processed with no human touch | Rises when confidence thresholds are lowered, pushing errors downstream |
Exception rate | Share routed to a human | The inverse of straight-through; says nothing about resolution quality |
Exception resolution time | Minutes to resolve one exception | The true labor driver; often unmeasured |
Silent error rate | Wrong values that passed as straight-through | The hidden cost of over-optimizing the headline rate |
How should an exception get routed and resolved?
An exception should be routed by the reason it failed, not dumped into one undifferentiated queue. A low-confidence field, an unreadable scan, a value that breaks a validation rule, and a document type the model has never seen are four different problems. Routing each to the right handler, with the source clause already surfaced, is what makes resolution fast.
Good exception handling has three properties. First, it is typed: the workflow knows why the document is an exception, whether a confidence score fell below threshold, a data validation rule flagged an out-of-range rent, or named entity recognition failed to resolve a tenant name. Second, it is located: the reviewer sees the exact page and clause in question, not the whole two-hundred-page lease. Third, it is closed-loop: the correction feeds back so the same failure mode does not recur on the next document.
Docsumo, in its work on document automation, frames exception handling as the deliberate design of the human-in-the-loop step rather than an afterthought bolted onto a pipeline. That framing is the point. As one way to put it: a straight-through rate tells you how often you avoided the hard part; the exception path tells you how well you handle it when you cannot. The first number flatters the demo. The second decides whether the workflow survives production.
How high should the straight-through rate be?
The straight-through rate should be as high as it can go without pushing errors past review, and no higher. There is no universal target. According to Ardent Partners, teams above 70% process documents at roughly one-third the cost of manual operations, yet most mid-market finance teams remain below 60%. The right number is the one that holds your silent-error rate near zero.
The reason a blanket target is dangerous is that straight-through rate and error rate trade off directly. Every extraction carries a confidence estimate. Raise the auto-accept threshold and more documents clear untouched, but weaker extractions slip through as silent errors. Lower it and the straight-through rate drops while the exception queue grows. The optimum depends on the cost of a missed error in your context, and in CRE underwriting a wrong base rent or a missed co-tenancy trigger is expensive.
This is why the human-in-the-loop design is not a limitation on straight-through processing. It is the mechanism that lets you push the rate up safely. A workflow that verifies fields at the point of low confidence can accept more automatically, because the ones it accepts are the ones it is sure of. The straight-through rate becomes an output of a well-designed exception path, not a target you chase by loosening thresholds.
Frequently Asked Questions
What is a good straight-through processing rate for CRE documents?
There is no universal number. Ardent Partners reports that teams above 70% process at roughly one-third of manual cost, while most mid-market teams sit below 60%. The right rate is the highest one that keeps your silent-error rate near zero, which depends on the cost of a missed field in your workflow.
Is a higher straight-through rate always better?
No. A straight-through rate can be inflated by lowering confidence thresholds, which pushes weak extractions past review and into the model as silent errors. A rate read without the exception resolution time and silent-error rate alongside it is a vanity metric, not a quality signal.
What is the difference between straight-through processing and exception handling?
Straight-through processing is the share of documents that reach validated structured output with no human touch. Exception handling is everything else: the routing and resolution of documents that failed a confidence or validation check. They are two sides of one workflow and must be measured together.
Why design a workflow around exceptions instead of automation?
Because the exception path holds most of the labor cost and nearly all of the risk. Auto-processed documents cost pennies of compute; each exception costs analyst minutes. Halving exception resolution time buys back more hours than any realistic gain on the already-cheap straight-through path.
Conclusion
Straight-through processing is a real metric, but it is the score, not the strategy. The strategy lives in the exception path, because that is where the human hours accumulate and where a loosened threshold quietly turns automation into a stream of silent errors. A workflow that types its exceptions, locates the failing clause, and closes the loop can raise its straight-through rate safely, because the documents it accepts are the ones it is confident in. A workflow that chases the headline number optimizes the cheap 80% and pays full price for the expensive 20% forever. Read the two numbers together. Design for the exceptions. The straight-through rate is what you get when the exception path is good, not the thing you get by ignoring it.