Most orders and every invoice get typed in by hand.
A made-up regional distributor of plumbing and HVAC supplies with three branches, an order desk of six and an accounts payable team of four. Customers email their purchase orders as PDFs, and about 140 vendors send invoices to one AP inbox.
The owner brought the order desk to the intro call. Orders were going out right, most days, but the desk kept growing with the business, and the typos surfaced at the dock.
The baseline counts the typing before anyone changes it.
The order desk supervisor and the controller timed the work for two weeks and pulled a month of volumes from the order inbox, the AP inbox and the ERP. Each figure carries how sure they were of it.
| What was counted | How sure | Baseline |
|---|---|---|
| Customer purchase orders, keyed by hand | Measured | 1,900 |
| Lines on an average order | Measured | 14 |
| Minutes to key one order | Measured | 9 |
| Orders with an entry error found later | Estimated | 61 |
| Cost to find and fix one entry error | Estimated | $85 |
| Vendor invoices | Measured | 2,400 |
| Minutes to key and match a routine invoice | Measured | 6 |
| Invoices that fail the match | Measured | 432 (18%) |
| Minutes to resolve one exception | Estimated | 22 |
| Loaded cost of an hour, order desk and AP | Measured | $38 |
| Early-pay discounts missed | Guess | a few thousand |
Measured means read from a system’s own records. Estimated means counted by hand or taken from last year’s books. A guess is written down so nobody forgets it, and it stays out of the math.
Added up, the typing came to 683 hours a month: 285 on orders, 240 on routine invoices and 158 on exceptions. The error count comes from credit memos the team had tagged as entry errors, which is why it is an estimate. Orders that arrive by EDI from the two largest accounts were left out of every count above.
The readout sized the value at stake as a range.
The owner and the controller answered about twenty minutes of questions on a call, then connected the order inbox, the AP inbox and the ERP, read-only, after seeing exactly what would be read. StrikeArc Scout scored the plays in the library for fit against those answers. P06 Cut the manual work from orders and invoices came first, P08 Get quotes and contracts out faster second.
Before the 45-minute readout, one of our people sized the value at stake and signed every figure. The low case assumes 55% of orders and routine invoices go through without a touch. The high case assumes 75%, with exceptions resolved seven minutes faster because each one arrives with its evidence attached.
The missed discounts stayed out: they were a guess. Hours given back are worth their loaded cost only when the time goes to other work, so the owner named that work before deciding: the past-due receivables and a backlog of vendor credits nobody had time to chase.
The readout’s do-not-build list had three lines, each with its reason.
- Releasing payments. The play posts approved bills, and the controller still releases every payment run. Nothing reopens this one. Money leaving the bank stays a person’s decision.
- The EDI orders. They already arrive as structured data, so there is nothing to retype. It reopens if either account moves to emailed orders.
- Quotes for non-stock items, for now. P08 scored second. One play goes in at a time, measured against its baseline, so quotes wait for the second quarter.
The owner decided on the call to start with P06.
The install ran gate by gate.
The play went into the inboxes and the ERP the company already used. Nothing went live until the last gate.
IT gave read access to both inboxes and to the ERP’s customers, items, prices, vendors, open purchase orders and receipts. Write access covered two record types only: sales orders and vendor bills. The order desk supervisor was named the one operator.
The controller set the tolerances for a match: price within 2%, quantities exact, freight under $75. Any bill over $5,000 waits for the controller, and any order line below its price floor waits for the inside sales lead.
A month of past orders and invoices ran through the play side by side with what the team had keyed, and every difference was reviewed. The test can fail, and it did: one vendor billed in cases where the purchase order counted units, so the controller added a unit conversion to the rules.
The controller approved the tolerances and thresholds, and the owner approved the launch with the baseline attached.
Orders went live first. Vendor invoices followed once the supervisor had cleared the first batch of order exceptions.
The supervisor’s time across the install came to about two and a half hours, most of it on the side-by-side test. IT’s part was granting the access and checking its scope.
The first quarter shows up in the ledger.
The play reports the share of documents processed without a touch. The ledger turns that share into hours and errors, then into dollars.
| What the play reports | Baseline | Quarter |
|---|---|---|
| Customer orders processed without a touch | 0% | 56% |
| Vendor invoices processed without a touch | 0% | 60% |
| Minutes to resolve one exception | 22 | 16 |
| Orders with an entry error found later, a month | 61 | 29 |
The cost to run comes from your own invoices and your team’s time records, so the example leaves those two lines blank.
A quarter at this rate runs to about $191,000 a year, inside the readout’s range. The hours went where the owner said they would: past-due receivables and vendor credits.
A person approved every step that mattered.
Every gate is a named human. Agents draft everything; they approve nothing. In this example the names are roles.
- The controller set the tolerances, approves every bill over $5,000 and releases every payment run.
- The inside sales lead approves any order line priced below its floor, with the line’s price history beside it.
- The order desk supervisor clears the exceptions the play cannot match, each with the document and the reason attached.
- The Value & Deployment Specialist and the owner signed each figure in the ledger before it was reported.
Some things did not change, for good reasons.
- The exception rate. 18% of invoices still fail the match. The play sends each one to the right person with the evidence attached, which is why an exception takes 16 minutes where it took 22. The cause sits upstream: one branch records receipts a day or two late, so invoices arrive before the goods are on the books. That is a receiving habit, and the readout named it as the fix.
- Payment runs. The controller releases every one, as before.
- The EDI orders. Untouched, as the readout said.
- The missed discounts. Nobody tracks vendor payment terms, so the guess stays a guess and the ledger counts none of them.
Run the same method on your own numbers.
The baseline above took two weeks of timing and a month of volumes. You can start the same count today.
- The invoice exception log counts the invoices that fail the match and the minutes each one takes.
- The cost per invoice worksheet turns your AP time, errors and late fees into one number per invoice.
- The Workflow Brief writes the workflow down on one page: who owns it, the steps, the systems and today’s numbers, each marked measured, estimated or a guess.
Bring them to an intro call, and the readout starts from your numbers. How the capture, the matching and the approvals work with the ERP you already have is in invoice processing automation.
This example runs on one play.
The play in this example is P06 Cut the manual work from orders and invoices. It reports the share of documents processed without a touch, the first two lines of the ledger above.