Structure predicts the ceiling
PO matching reaches 94% because the comparison is structured and the correct answer is knowable. Multi-line allocation sits at 66% at higher volume, because the context needed is often not on the document.
Free tool
Automation readiness is not a property of your company. It is a property of each workflow, and it depends almost entirely on whether the information an agent would need is present in your systems rather than in somebody's head.
Send several hundred historical transactions with known outcomes. We will score against them and show you the real band.
Tick the prerequisites you genuinely meet. Bars in blue show the published ceiling; grey bars show roughly where a workflow lands when its prerequisite is missing.
How to read it
The ranking matters more than the percentages, because it tells you which workflow to automate first and which needs a data fix before it is worth attempting.
PO matching reaches 94% because the comparison is structured and the correct answer is knowable. Multi-line allocation sits at 66% at higher volume, because the context needed is often not on the document.
Expense coding tops out at 84% and the binding factor is receipt capture quality, not the model. Fixing capture returns more than any amount of tuning.
Two thirds of a large volume is a lot of hours. The point is to build the business case on the real number rather than on an assumption of near-total automation.
Across our customer base, where a workflow lands is predicted mostly by four things, and all four are knowable before you buy anything.
Vendor concentration: how much of your volume comes from suppliers you bill repeatedly. Document quality: whether invoices arrive as structured PDFs or as photographs of paper. Policy clarity: whether coding rules are written down or reconstructed each time. History depth: how many prior examples exist for the pattern to be learned from.
The workflows that plateau lowest are the ones where the correct answer depends on something that was never recorded. A multi-line invoice that should be split across three departments according to a conversation somebody had is not a model problem — the information required to get it right does not exist in any system.
That is why we score ourselves at 66% on allocation and publish it. An agent that guessed confidently at the remaining third would be worse than one that escalates, because the errors would be indistinguishable from the successes until somebody found them.
The highest-return action for most companies is not choosing a better agent. It is deduplicating the vendor master, connecting the bank feeds properly, or putting receipt capture at the point of spend.
Each of those is unglamorous, cheap relative to a software purchase, and moves a ceiling by twenty or thirty points. They also improve your reporting whether or not you ever automate anything, which makes them a safe investment under uncertainty.
Take the highest bar where you already meet the prerequisite, and check that it is also high-volume for you. Automation value is rate times volume, and a 94% ceiling on a workflow you run forty times a month is worth less than 84% on one you run two thousand times.
Ceilings are the median straight-through rate at week 20 from our agent benchmarks, computed across 23 customers running these workflows continuously between January 2025 and June 2026.
Straight-through means the transaction completed with no human interaction and was not reversed within 90 days. Approval clicks are excluded, which is why these figures are ten to fifteen points below what a looser definition would produce.
The grey bars — roughly 62% of the ceiling — are an approximation of where a workflow lands when its prerequisite is absent. That factor is derived from the spread in our sample rather than measured directly per prerequisite, and it is the weakest number on this page.
The bottom quartile of our customers sits materially below every median shown here. If your data resembles that quartile, plan against the lower figure. The only benchmark that predicts your result is one run on your own historical transactions, which we offer before purchase.
Questions
Fix the prerequisite on your highest-volume workflow and the ceiling moves further than any model change would.