Read customer orders from every channel, create them in the ERP with confirmed dates, and clear the credit, delivery and data blocks that hold them back from shipping.
At 07:50 Helvetia Retail AG e-mailed a PDF order for 2,400 units of material 7741-X, worth about $101,000. Juno read it, mapped the customer’s part number to 7741-X, checked price and terms against the contract and created the sales order in SAP S/4HANA at A4; the availability check confirmed 1,800 units for Friday and 600 for the following Tuesday, and the acknowledgement was drafted.
The order landed on the credit-blocked list, because with 2,400 more units exposure stands 6% above the limit. At 09:30 the customer’s Thursday payment run cleared and the hold J. Harrison signed on Tuesday lifted. Juno refreshed its draft release with the evidence on the card: 92% of what is still overdue is Wednesday’s $23,600 price-protection case, and the account has paid on time in 47 of the last 48 months. The order is above the credit analyst’s limit, so the draft waits for her.
She approves the release at 09:47. SAP records it with her name, and the acknowledgement goes out at 09:49 with the confirmed dates. Tuesday’s hold was a credit decision with a signature, the A0 scene of chapter 3; Thursday’s release sits inside the policy she owns, so the agent drafts it and she approves it. Showcase estate; the roles are designed, yet to be built.
Order intake joins the supply chain to finance, and every large ERP now ships an agent for it. Companies that redesign order-to-cash around AI take 134% more orders digitally and automate 138% more credit decisions (The Hackett Group, 15 July 2026) 1. SAP made sales order creation from unstructured documents generally available in its public cloud edition and added guidance across order processing, sourcing, availability, returns and fulfilment (SAP, 14 April 2026) 10.
Oracle ships a purchase-order-to-sales-order converter agent, included in the subscription (Oracle, 10 February 2026) 16, and a command-centre application for sales orders that summarises exceptions, applies and releases holds, cancels or returns orders on a typed instruction and resolves backorders by substitute item or alternate location (Oracle, 9 April 2026; 8 May 2026) 15 17, with an agent crawling orders for new exceptions (Oracle, 10 September 2026) 18. Microsoft’s sales order agent reads customer e-mails, checks availability, computes capable-to-promise dates and drafts quotes and orders, with a designated user reviewing every outgoing message (Microsoft, 23 July 2026) 34.
The named results are fewer than for the three finance steps, which is why the use case ranks fourth. Deloitte reports a Fortune 500 beverage company at 95% order-entry accuracy with 90% less entry time (Deloitte, n.d.) 35. Celonis reports Cosentino processing up to five times more orders a day with its credit-block assistant (Celonis, 23 October 2024) 36, which releases blocks above a confidence floor (Celonis, n.d.) 37. Esker’s customers report order processing three times faster (Esker, 9 September 2026) 30, Blue Yonder’s customer service agent captures and modifies orders and surfaces payment risks (Blue Yonder, 9 January 2026) 43, and Genpact calls order management the next frontier for agents (Genpact, 17 August 2026) 47.
A customer service representative reads each purchase order as it arrives by e-mail, PDF, portal or fax, keys it into the sales order screen, translates customer part numbers, checks price and terms against the contract and requests an availability check. Orders that fail the incompletion check, hit a delivery block or exceed the credit limit stop there.
A credit analyst then opens the blocked-orders list, reviews exposure, overdue items and payment history, calls sales, and releases or holds the order; acknowledgements go out only after that. One vendor reports order entry time falling by 90% when the reading and keying is automated 35, and Celonis reports a customer moving up to five times more orders a day once credit blocks were handled with evidence at hand 36.
Two desks, two waits: the order waits to be keyed, then waits on the block while the analyst gathers what the release needs from three screens and a phone call.
The same order, by hand and with the intake and release roles. Illustrative durations; the steps are a design.
The same six stages as chapter 6, with the order desk’s and the credit analyst’s objects. Everything in the rows below is a design: intake and acknowledgements sit on the blueprint at A4, the entry exceptions and the credit release at A2, the release policy itself at A0, and the tense is “would”.
Draft-and-approve (A2) is the market’s usual level for orders read from e-mail and PDF and for credit releases: the ERP vendors keep a person reviewing every outgoing document and every release recommendation 17 34. Execute within rules (A3) applies to clean EDI orders and to releases above a confidence floor, which one process-mining vendor automates 37. Lima’s blueprint designs intake and acknowledgements at A4, validation and status at A3, entry exceptions and credit block release at A2, and keeps the release policy human-led at A0, the credit hold decision of chapter 3. The badge reads A2 because the decision this page turns on, releasing a blocked order, is one a named person approves; clean intake is designed a level above the market’s usual for PDF orders, and a client could hold it at A2 until the accuracy tests pass.
| Element | Setting | Set by | Changes |
|---|---|---|---|
| Objective | Create every customer order in the ERP with confirmed dates the hour it arrives, and clear the blocks that hold it, inside the credit policy | Order desk lead, in a sentence | New version |
| Data scope | Inbound orders, sales orders, blocks, credit exposure, open items and disputes, availability, the part cross-reference and contract terms in SAP S/4HANA, Oracle Fusion, the EDI gateway, the inbox, the CRM, the portal and the planning suite | A. Novak, data lead | New version |
| Rules | The intake rules, written by the order desk lead, and the credit release policy, written by the credit manager, each with its planning table, version 1 | Credit manager with the Planner Agent | New version, simulated first |
| Trigger | An order in the inbox, the EDI queue, the portal or the CRM · an order failing validation or landing on a blocked list · a change request · an hourly sweep of open blocks | Order desk lead | At once |
| Actions permitted | Create a sales order · change a schedule line after the availability check · remove a delivery block · release a credit block with the approver recorded · send the acknowledgement · add the order note | Order desk lead, owner | New version |
| Limits | Auto-release up to $10,000 over the limit for accounts with 45 of 48 months on time · the credit analyst to $100,000 of order value · above that, the finance lead · confidence floor 0.85 for an extracted order · daily release budget 80 | J. Harrison, finance lead | New version |
| Escalation | An extraction below the floor or a deviation from the contract, to the order desk · a release above the auto-release limit, to the credit analyst, and above $100,000 to the finance lead · a shortfall on a strategic account, to the account manager · competing claims on the same stock, to J. Ridley | Order desk lead | At once |
| Level | A2 Supervised for entry exceptions and credit release; A4 for clean intake and acknowledgements; A3 for validation and status; the release policy stays human-led | J. Harrison, finance lead | Up on evidence, down in one step |
| Feedback | Every corrected extraction, approved, edited or refused release and re-promised date 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 |
|---|
| Digital and touchless order intake share: 134% higher at leaders [1] |
| Order-entry time and accuracy: 90% less time and 95% accuracy at one named adopter [35] |
| Orders processed per day: up to five times more at one named adopter [36] |
| Automated credit decisions: 138% more at leaders [1] |
| Blocked-order dwell time and release service level |
| Order cycle time: three times faster at one vendor's customers [30] |
| On-time delivery against the first confirmed date |
Baseline blocked-order dwell time first, from the block timestamp to the release entry in the ERP, by block reason. It is the wait the customer feels, it is the measure the first release drafts would move within a week, and it separates the credit decision from the keying. The intake share and the entry time come second, once the desk has a month of extracted orders to score for accuracy against the documents.
The hold is the policy: a credit decision with a signature, as chapter 3 showed. A release inside that policy is a routine case with evidence, so the agent drafts it and a named person approves. The policy itself stays with people.
Below the confidence floor, or where price, terms or part mapping deviate from the contract, the order waits for the order desk before it is created. Every correction is recorded and teaches the next read of that customer’s documents.
Yes. The sales order, the availability check and the block are ontology objects, so L. Tan’s shortage list in chapter 3 and J. Harrison’s release queue read the same line.
The order is in and released, and building it needs parts the estate has to buy. Where does the buying side lose its money, and which purchase-to-pay use cases go first?
Invoice matching, sourcing, the purchase order agent that runs in production, and supplier risk: the four on the buying side.
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.