Open purchase order lines are compared every morning with current demand, supplier confirmations and lead-time fences.
Monday 21 September 2026, 08:50. D. Okafor, buyer, opens his approval queue. The 06:00 run on rule version 3 found 38 open purchase order lines that can still be cancelled this week, worth $412,000, after a customer push-out left them above the need. Twenty-nine sit under the $25,000 threshold and wait as drafts worth $187,300, each with its projected excess, its window and its supplier class.
He reads each row: the action, the three conditions it matched, the value at stake and the confidence. Twenty-eight he approves as drafted; one he edits to a smaller quantity. By 09:12 the queue is done: 22 minutes. The changes go to SAP S/4HANA through its released purchase order interface, and by 09:21 each answer is stored word for word.
The 9 lines at or above the threshold reach D. Okafor as buyer, for his judgement, each with the reason, before the window closes on Thursday 24 September; he is one person in two chairs. L. Tan, who stated the policy on the Friday, finds the 38 lines already found when she opens her 1,240 messages at 07:40. The Monday is the showcase estate; the agent is the one that runs in production.
Managing purchase orders is the transactional work the analysts count. About 28% of procurement staff time goes on it, work Gartner calls suitable for automation (Gartner, in Supply Chain Management Review, 18 February 2026) 4; planners spend over half their time sorting dashboards and filters for supply issues, and agents could cut that effort by up to 80% (Kinaxis, 2 April 2025) 64.
The market’s agents draft, and the buyer applies. Microsoft’s Procurement Agent for Dynamics 365 reads supplier e-mails, identifies changed quantity, price, date and cancellation, assesses the effect on inventory and orders, and drafts chasers for unconfirmed orders; the buyer reviews and applies each change (Microsoft Learn, 10 and 26 August 2026) 34 35. SAP’s Joule mass update of delivery dates by plain-language command is generally available (SAP, 22 October 2025) 16; Oracle ships a Purchase Order Status Advisor (Oracle, 9 April 2026) 22 24.
The named numbers measure the chase. Molex raised its purchase order confirmation rate from 30% to 90% with Celonis process intelligence (IT Brief UK, 16 April 2026) 38; JBT AeroTech reached 86% of purchase orders acknowledged within 72 hours (SourceDay, 8 July 2026) 58. Lima’s PO cancellation agent runs in production: planners author the policy in plain language, the rules engine executes it on federated ERP data, write-back happens under human approval and 100% of actions trace to a rule production · anonymised.
Governance is the gate, and this use case shows it. 6% of supply chain organisations run autonomous AI at scale today and 41% expect to within two years; 52% cite trust in AI decisions as the top barrier (IDC for Kinaxis, 11 August 2026) 63. Said plainly, the surveys speak of managing purchase orders in general; a survey that names cancellation or reschedule agents as deployed has yet to appear, and the named numbers come from process intelligence and a purchase order collaboration platform.
The MRP run posts exception messages overnight: reschedule in, reschedule out, cancel. The buyer filters by MRP controller, opens each purchase order line and checks projected stock and pegged demand in the planning suite, the lead-time fence, the cancellation window, the NCNR flag and the supplier class. That is the 70 minutes of filtering and the four hours of drafting on the showcase Monday.
Then the buyer e-mails the supplier to cancel or reschedule, and the reply takes a day or more. The buyer sets the deletion indicator or changes the delivery date, records the confirmation, tells the planner and logs the line in a spreadsheet: two hours of updating. Unconfirmed orders are chased from a weekly report; confirmations arrive as e-mail or PDF and are keyed by hand.
Four people touch one batch: the buyer, the supply planner, the MRP controller and the supplier. Elapsed in the showcase estate: 2.5 days for one batch of 38 lines, with the cancellation windows closing on the Thursday. Molex started its chase at a 30% confirmation rate (IT Brief UK, 16 April 2026).
The four rows chapter 1 drew, for one batch of 38 lines. Showcase estate.
This is the one place in the use-case family where the six stages are a record and a design at once. Chapter 5 follows the cancellation agent through the Friday, the Monday and the six weeks; the rows below place that agent in its chain and say which of the neighbouring roles, the reschedules, the chasers and the confirmations, are designed and yet to be built.
| Row | Kind | What it tests or does | Data it reads | Then |
|---|---|---|---|---|
| 01 | Check | Projected excess is above 8 weeks of demand edited by L. Tan · 10:10 | stock on hand, stock on order, forecast, sales orders | carry on |
| 02 | Check | The cancellation window is open on the run date | order date, cancellation window in days | carry on |
| 03 | Check | The supplier class is standard | supplier record, supplier class | carry on |
| 04 | Action | Draft the cancellation of the open quantity | open quantity, line value, currency | row 05 decides the route |
| 05 | Limit | Line value against $25,000 | line value | under: the queue · at or above: the buyer |
The policy L. Tan stated on Friday 18 September and the five rows the Planner Agent drafted from it; she adopted four and edited one. Chapter 5 holds the full worked example: the Friday hour by hour, the Monday run, the audit record and the level review. Showcase estate.
The market’s usual level is draft-and-approve: Microsoft’s agent proposes changes that purchasers confirm and apply, with automatic chasers as a setting that stays off until a buyer switches it on (Microsoft Learn, 26 August 2026); SAP’s delivery-date update runs on a purchaser’s command (SAP, 22 October 2025). The blueprint lists PO change and cancellation at A3; the showcase estate started it at A2 and, after 212 approvals with 9 edits, moved lines under $25,000 to A3. Designed at A3, started at A2, raised on evidence.
| Element | Setting | Set by | Changes |
|---|---|---|---|
| Objective | Cancel purchase order lines that stand above the need, inside their cancellation window | L. Tan, planner | New version |
| Data scope | Purchase order lines, schedule lines, demand, stock and supplier class at plants 1210 and 1220 | A. Novak, data lead | New version |
| Rules | The policy sentence and its planning table, version 3 | L. Tan with the Planner Agent | New version, simulated first |
| Trigger | 06:00 each working day, after the planning run | L. Tan, planner | At once |
| Actions permitted | Draft a cancellation · send a change · open a query to the buyer | L. Tan, owner | New version |
| Limits | Value threshold $25,000 · daily action budget 60 · confidence floor 0.80 | L. Tan and J. Harrison, finance | New version |
| Escalation | Above the threshold, to the buyer with the reason · a failed write-back, to the owner | L. Tan, owner | At once |
| Level | A2 Supervised with D. Okafor as the named approver; lines under $25,000 at A3 since the level review | J. Ridley, director | Up on evidence, down in one step |
| Feedback | Every approve, edit and reject is recorded and used in the next cycle | Always on | Fixed |
The nine settings of chapter 4 with the values chapter 5 shows; the owners follow the ruling on who owns what. Showcase estate.
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 Procure-to-Pay lists the rest of the map for this chain.
| Measure |
|---|
| open purchase order value above need ($412,000 on the showcase Monday) |
| cancellations executed inside the window |
| purchase order confirmation rate (30% to 90% at Molex; 86% within 72 hours at JBT) |
| reschedule acceptance |
| expedite and premium freight spend |
| buyer minutes per batch (38 minutes against 2.5 days in the showcase estate) |
| overrides and edits per 100 drafts (9 edits in 212 in the showcase estate) |
Baseline open purchase order value above need first: it is the number the Monday opens on, $412,000 in the showcase estate, and the one the finance lead reads. Buyer minutes per batch and overrides per 100 drafts come next, because they are the evidence a level review needs: 38 minutes against 2.5 days, and 9 edits in 212, are what carried the lines under $25,000 to A3.
The cancellation agent: planners author the policy in plain language, the rules engine executes it on federated ERP data, write-back happens under human approval and 100% of actions trace to a rule, at a high-tech manufacturer. Reschedules, chasers and confirmation tracking are designed roles, yet to be built.
The agent drafted 212 cancellations; D. Okafor approved 203 as written and edited 9. On that record J. Ridley moved lines under $25,000 to A3, where the agent executes inside version 3 and D. Okafor handles exceptions only. Lines at or above the threshold still reach him as drafts.
The rejection returns the line to the buyer with the supplier’s words attached, the line stays open, and the next run raises it again while its window is open. A line is closed only when the ERP acknowledges the change.
The policy checked the supplier class. Who checks the supplier: its finances, the sanctions lists, and the end-of-life notice on the part?
Supplier risk, sanctions and end-of-life monitoring: continuous screening, and the last-time buy drafted before the window closes.
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.