Benchmarking straight-through processing: the KPIs for a cargo capture desk

Straight-through processing is the single number that tells you whether a cargo data-capture desk is working. It is the share of documents — master air waybills, house waybills, manifests — that flow from arrival to export fully automated, with no person touching a screen. Almost everything else you might measure about a keying desk is downstream of it. Yet most forwarders can’t say what their STP rate actually is, which means they can’t say whether an automation project paid off. This is a guide to benchmarking a capture desk the way an operations lead should: pick the few metrics that matter, measure a baseline, change one thing, then re-measure.

Straight-through processing is the headline metric

The STP rate is the percentage of documents that clear the pipeline with no human touch — captured, extracted, validated, and exported automatically. Its mirror image is the exception rate: the percentage routed to a person for review. The two are complements and always sum to 100%. If 82% of your air waybills export automatically, your STP rate is 82% and your exception rate is 18%. Raising one lowers the other, so you can track a single figure and know both. Everything a buyer cares about — throughput, headcount, unit cost — moves with this one number, which is why it belongs at the top of any capture-desk scorecard rather than buried under accuracy stats.

The four numbers to track

A useful benchmark needs more than the headline. These four measure the same desk from different angles — volume automated, human effort, work diverted, and money — and read together they tell you where an automation gain is real versus where it just moved cost around.

Touch time (or handling time) is minutes of human work per AWB. Manual keying of a master air waybill commonly runs on the order of a few minutes each; the goal of automation is to drag the average across all documents toward seconds — not by keying faster, but by only ever touching the exceptions. Cost per AWB is fully-loaded processing cost divided by volume, and it falls as the STP rate rises: fewer human-touched documents means less labor carried per record. First-pass accuracy — the correction rate — is the honesty check on the other three, because a high STP rate built on fields that later need fixing just relocates the cost downstream.

How to measure it: one baseline week

You don’t need a project to get a baseline — you need a representative week. Pick a normal week (not a peak, not a holiday lull) and capture four things. Count the documents that came through: total volume in, split by type if you can. Count the human touches: how many of those documents a person opened, corrected, re-keyed, or chased. That split gives you the STP rate and the exception rate directly. Time a sample — take twenty or thirty documents and clock each from the moment a person opens one to the moment they release it, then average and weight by your exception rate to get touch time per AWB. Total the loaded cost of the desk for the week — salaries, tooling, overhead — and divide by volume for cost per AWB. Four numbers, one week. Then change one thing — automate a document type, raise a threshold — and run the exact same count again. The delta is your ROI, and because you measured a real baseline, it is defensible.

What good looks like

Resist the urge to anchor on a single target percentage; the honest benchmark is your own baseline moving in the right direction. That said, the shape of a healthy desk is consistent: the STP rate climbs and stays climbing as you automate more document types; touch time falls from minutes-per-AWB toward seconds-per-AWB on average, because a person now only sees the exceptions; cost per AWB drops roughly in step with STP; and first-pass accuracy holds steady or improves rather than quietly eroding as volume grows. If STP is rising but corrections are rising with it, the gain is an illusion — you have automated the easy documents and pushed the hard ones downstream. A good desk moves all four numbers together.

It also helps to know where the remaining exceptions cluster, because that is where the next point of STP lives. In most operations they concentrate in a few predictable places: poor-quality scans and photos, a handful of low-volume carriers or agents whose document layouts the pipeline hasn’t seen enough of, and a small set of fields — charges, special-handling codes, unusual routings — that are simply harder to read reliably. Segment your exception rate by those buckets before you chase a higher STP number. Automating a document type that already clears 95% of the time buys you almost nothing; fixing the source of a cluster that fails half the time is where the throughput actually comes from. The benchmark tells you the score; the segmentation tells you the next move.

Where the STP rate actually comes from

In an exceptions-only system, the STP rate is not a report you run after the fact — it is the shape of the pipeline itself. AWBGuru auto-exports a document straight through only when every extracted field clears the confidence threshold the owner set and no field has been flagged by its correction-tracking guard; everything else lands in a review queue. So the straight-through path is the automated path, and the STP rate is literally the proportion of documents that take it. The confidence threshold is the dial an operations lead tunes: raise it and fewer documents auto-clear (lower STP, lower risk of a bad field slipping through); lower it and more clear untouched (higher STP, more trust placed in the extraction). Because the same pipeline extracts the master air waybill data that feeds those thresholds, the four benchmark numbers stop being a manual audit and become a live read on how the desk is set.