Match every incoming customer payment to its open invoices, post it in the ERP the day it arrives, and code each short-pay for the deductions desk.
J. Harrison, finance lead, opens the cash queue. Overnight Northbay Logistics paid $184,300 against 14 invoices, and the remittance arrived as a PDF in the shared inbox. At 06:30 Juno read it, matched 12 invoices, applied a 2% discount that sits inside terms and posted the receipt in SAP S/4HANA at A4. The accounting document number is on the card.
Two lines are short. The larger one, $6,200, carries the customer’s freight claim number, so on its second pass Juno opened a dispute case with the PDF attached and routed it to the deductions desk. The smaller one, $410, sits inside the write-off tolerance, and the tolerance is hers, so the write-off waits for her as a one-click draft. She approves it.
On Friday she audits a sample, 20 of the week’s 340 postings, each read from its record. This is the A4 scene of the gamut in chapter 3: the agent posts, re-tries and re-routes; the person audits after the fact. Showcase estate; the role is designed, yet to be built.
Cash application is the only order-to-cash step with two dated analyst benchmarks of its own. Hackett benchmarked 15 cash application vendors and found a median auto-match rate of 70%, same-day matching at 61% of companies and up to $15 million of unapplied cash to release at a typical large enterprise (The Hackett Group, 11 August 2025) 2. Companies that redesign order-to-cash around AI apply 89% more payments automatically and run the process at 52% to 59% lower cost (The Hackett Group, 15 July 2026) 1. Hackett’s 2026 review describes receivables platforms moving to workflows executed by agents with human supervisors (The Hackett Group, 30 March 2026) 3.
The ERP vendors have built it in. SAP made AI payment advice processing generally available with a 70% reduction in document processing effort (SAP, 14 April 2026) 10, and lists a cash management agent for bank statements from the first quarter of 2026 (SAP, H2 2025) 13. BlackLine’s remittance agent processes remittances at about 90% straight-through (BlackLine, 14 April 2026) 27 and is trained on more than $1 trillion of receivables transactions 28.
The named results are vendor-published and high. HighRadius names Kimberly-Clark at 97% straight-through cash application and Johnsonville at 95% (HighRadius, 23 September 2026) 23, and DXP at 90% automation with a 20-day reduction in days sales outstanding (HighRadius, 10 September 2026) 22. Esker reports up to 95% less unapplied cash for its customers (Esker, n.d.) 31, and Gartner’s 2024 assessment of invoice-to-cash applications lists nine vendors that all sell matching automation (Gartner, 6 May 2024) 6. A finance lead recognises the unapplied cash and the lockbox key-in fees at once.
Each morning an accounts receivable analyst opens the bank statement and the lockbox file, then hunts for the remittance advice in e-mail, the customer’s payables portal or an EDI 820 message. The analyst identifies the payer from bank details, keys invoice numbers and amounts, applies cash discounts and tolerances, codes short-pays with a reason and posts the receipt in SAP or Oracle.
Payments that fail to match sit as unapplied cash on the customer account and are chased for days, while lockbox banks charge key-in fees for paper remittances. The median company matches 70% automatically and 61% match on the day of receipt; the rest takes one to five working days of analyst time per batch 2.
The cost sits in three places: the analyst hours per 1,000 payments, the bank’s key-in fees, and the cash that stays unapplied while the customer’s account shows an overdue balance it has already paid, which the collections desk then chases.
A cash applier’s morning by hand, and the days after it. Illustrative times; the steps are the ones a receivables desk runs today.
Chapter 6 gave the six stages behind the PO cancellation agent; the same machinery would carry cash application, with the finance objects in place of purchase order lines. Everything in the rows below is a design: the roles sit on the blueprint, and the tense is “would”.
| Row | Kind | What it tests or does | Data it reads | Then |
|---|---|---|---|---|
| 01 | Check | The payer is identified from bank details or the remittance | bank statement, lockbox file, customer master | carry on |
| 02 | Check | Remittance lines match open items by number, amount and date | remittance advice, open items | carry on |
| 03 | Check | The discount taken is inside the payment terms | payment terms, invoice date, payment date | carry on |
| 04 | Action | Post the receipt and clear the matched items | incoming payment, open items | write-back; the document number confirms it |
| 05 | Limit | The residual against the $500 tolerance, and the reason on the remittance | residual, reason code | inside: a write-off draft · a reason given: a dispute case · otherwise: to the analyst |
A plausible policy sentence for the matching roles, and the planning table it would compile to. Designed, yet to be built; the tolerance, the floor and the sweep are illustrative settings.
Execute within rules (A3) is the market’s usual level: vendors post matched payments at 90% to 97% straight-through and route the remainder to analysts 23 27, and Hackett describes receivables platforms moving to workflows executed by agents with human supervisors 2 3. Lima’s blueprint designs cash application and matching, and bank and lockbox file ingestion, at A4, and remittance capture and unapplied cash resolution at A3. The difference is the second pass and the re-route: ‘self-corrects’ means the agent re-tries and re-routes on its own, and the audit stays with a person. A client could start the matching roles at A3 and move them up on the recorded cases, the way the PO cancellation agent moved from A2.
| Element | Setting | Set by | Changes |
|---|---|---|---|
| Objective | Post every incoming customer payment against its invoices the day it arrives, and code each short-pay for the deductions desk | Receivables team lead, in a sentence | New version |
| Data scope | Bank statements and lockbox files, remittance advices, incoming payments, open items and aging, customer bank details, payment terms and reason codes in SAP S/4HANA and Oracle Fusion Receivables | A. Novak, data lead | New version |
| Rules | The matching policy and its planning table, version 1, simulated on last quarter’s payments | Receivables team lead with the Planner Agent | New version, simulated first |
| Trigger | Each bank file and remittance as it arrives · a sweep at 23:00 for anything still unapplied | Receivables team lead | At once |
| Actions permitted | Post a receipt and clear items · apply discount and tolerance · park unapplied cash with a reason · open a dispute case · draft a write-off inside tolerance | Receivables team lead, owner | New version |
| Limits | Residual tolerance $500 · confidence floor 0.90 for a posting · daily posting budget 600 · two-person rule on any change to customer bank details | J. Harrison, finance lead | New version |
| Escalation | Below the floor or a split payment, to the analyst the same day · a write-off above tolerance or a re-assigned payer, to the team lead · a failed write-back, to the owner | Receivables team lead | At once |
| Level | A4 Autonomous for matching and file ingestion; A3 for remittance capture and unapplied cash, with the analyst on exceptions | J. Harrison, finance lead | Up on evidence, down in one step |
| Feedback | Every corrected match, approved write-off and re-assigned payer is recorded and used in the next cycle | Always on | Fixed |
Every row is a setting or a versioned rule; the level and the limits are the ones a manager signs.
Lima’s agent blueprint is a design of 960 agent roles across five end-to-end chains. It is the map of what Juno can be configured to do, and the first agents are built from it one at a time, the way the PO cancellation agent was.
Additional capability in Order-to-Cash lists the rest of the map for this chain.
| Measure |
|---|
| Auto-match rate: baseline median 70%, top third above 80% [2] |
| Same-day application: baseline 61% of companies [2] |
| Unapplied cash: up to $15 million at a typical large enterprise [2] |
| Straight-through rate: 90% to 97% at named adopters [23][27] |
| Days sales outstanding: 20-day reduction at one named adopter [23] |
| Bank key-in fees and analyst hours per 1,000 payments [23] |
Baseline the auto-match rate first. It is the number the two Hackett benchmarks give you to compare against, a 70% median with the top third above 80% (The Hackett Group, 11 August 2025), and the other measures follow it: same-day application rises as matching rises, unapplied cash falls, and the analyst hours per 1,000 payments fall with it. Count it from the ERP’s clearing documents for the last quarter before the first run, so the first weeks compare like with like.
The market’s vendors post matched payments and route the rest. A4 adds the second pass and the re-route, and keeps the audit with a person. A client can start at A3 and move up on recorded cases, the way chapter 5’s agent did.
It is parked as unapplied cash with the reason found and reaches the analyst’s queue the same day, with the candidate invoices the agent considered. The nightly sweep raises it again until it clears.
It reads them to identify the payer. A change to bank details waits for two named people on any agent whose actions can reach those fields, and this agent’s permitted actions leave them untouched.
The payment is posted and the short-pay is a case. Who chases the invoices that were paid late, or stay unpaid, and in what order?
Ranking overdue accounts by cash and likelihood, the right reminder at the right time, and every promise to pay followed to its outcome.
Bring two systems and one decision your planners make every week. In the first week we register the data, confirm the mapping with your team, answer your questions in ChatLima and run the first agent at the level you choose.