Pricing
Log inBook a demo

By role

For the people who actually process the transactions

Most software in this category is sold to a CFO on a headcount argument, which is a strange thing to read if you are the person doing the work. So this page is about what changes for you — and about the finding that in twenty-three deployments we measured, nobody's team got smaller.

Get a free assessmentBook a demo

Today · by hand

  1. 01Open the email, save the PDF0:40
  2. 02Find the vendor0:55
  3. 03Key header and lines3:10
  4. 04Look up the GL code1:20
  5. 05Find and open the PO1:35
  6. 06Compare quantity and price1:05
  7. 07Email the approver0:45
  8. 08Chase on Thursday1:30
  9. 09Post the entry0:50
Touch time per bill11:50

With the agent

  1. 01Bill parsedauto
  2. 02Vendor matchedauto
  3. 03Lines extractedauto
  4. 04Coded from historyauto
  5. 05Matched to POauto
  6. 06Duplicate checkauto
  7. 07Policy checkedauto
  8. 08You confirm0:06
  9. 09Postedauto
Human touch time0:06
Nobody reduced headcountExceptions, not a pileYou train it by correcting it
The situationWhat we doQuestions

The situation

Six things that make the job worse than it needs to be.

Keying the same vendor forever

The same forty vendors send the same layouts every month and you type them in every month. It is the most obviously automatable work in finance and the last to get automated.

Chasing approvers

The fourth reminder to the same person about the same invoice. Nobody enjoys it and it is a large share of the week in most AP functions.

Hunting for the PO

Opening three screens to find whether a bill matches an order, then comparing lines by eye. Structured comparison work done manually.

Applying lump-sum payments

One wire covering eleven invoices with a short payment on one. Twenty minutes of puzzle, several times a week, on the AR side.

Finding out about duplicates late

A statement paid as an invoice, discovered at reconciliation or by the vendor. Preventable at intake and rarely prevented there.

Month-end pile-up

Everything arriving at once because upstream cut-offs are not enforced, so the last three days of the month are unmanageable regardless of how the rest went.

The honest version of what happens to the job

Software sold on automating AP is usually sold on reducing headcount, so it is reasonable to be suspicious. Across twenty-three deployments we measured, in the first year, no finance team got smaller. Four grew, because the company grew and finance stopped being the constraint.

What changed was the mix. The keying largely went, and the time went into exception handling, collections follow-up, and the month-end work that used to spill into the second week. Whether your company converts that into growth or into savings is a management decision rather than an automatic consequence, and we would rather say that plainly than promise either.

In twenty-three deployments, nobody’s finance team got smaller in the first year. Four got bigger. What changed was what people spent the week on.

An exception queue is not a to-do list

The thing that makes this bearable rather than annoying is that what reaches you is genuinely uncertain. A bill where every field is confident except the PO reference arrives with that one field highlighted, and takes four seconds.

You are not reviewing four hundred items a week to catch three problems. If you are, the policy threshold is wrong and that is something to raise — we report approval rate and time per item precisely so an unreasonable queue is visible rather than endured.

Rejecting is as fast as approving

In most systems approving is one click and rejecting needs a reason, a routing choice, and a comment. Under time pressure that asymmetry means people approve, and the control quietly stops existing.

Here rejection is one action with a reason from a short list, and every rejection becomes a labelled example that makes the agent better at your business. You are, in effect, training the thing that will bother you less next month — which is the only incentive that survives a busy week.

What does not change

You still release the payments. That is human at any amount and any confidence, and it is absent from the agent permission model rather than switched off. You still handle the vendor who calls, the dispute, the exception nobody anticipated. The judgement stays.

Questions

What people ask.

Is this going to replace my job?
In the deployments we have measured, no team got smaller in the first year and four grew. The keying goes; the exception handling, collections, and judgement stay. What your company does with the returned capacity is its decision, and we will not pretend to know it.
How much will still reach me?
For AP at a settled Level 2, typically ten to twenty percent of bills. If it is more than a third the policy is too tight; under five percent it is probably too loose. Both are adjustable and you should say which it is.
Will it code things wrong?
Sometimes, especially in the first few weeks while it learns your conventions. Every correction you make becomes an example, which is why the accuracy curve climbs rather than staying flat.
Do I still release payments?
Always. Payment release requires a named human at any amount, and it is absent from the agent permission model rather than turned off — there is no setting that changes it.
What if the queue becomes unmanageable?
Raise it. We report approval rate and time per item specifically so an unreasonable queue is visible. The fix is upstream in the policy threshold, not in you working faster.

See what would come off your desk.

Fifty real bills is enough to show you exactly what the agent would have done and what would still reach you.

Get a free assessmentBook a demo