When the trial balance freezes, Juno would compare every account and cost centre with the prior period, the prior year and the budget, drill the material movements to the transactions and the supply chain events behind them, draft the explanation in the company’s words and route each draft to the owner who confirms it before the pack goes to the executive committee.
It is workday 3, 14:20, an hour after the trial balance froze. J. Harrison, finance lead, sees 27 variances above the threshold, each with a drafted explanation and its transactions attached. Cost of sales at plant 1210 is up 8.1% on August, and the draft reads: scrap on 7741-X rose to 4.6% after the tool change on the 12th ($212,000); purchase price variance on Brightline Metals SA followed the copper index ($148,000); the idle line in week 3 added an absorption charge of $96,000.
She asks ChatLima which of the three repeat from last quarter, and the answer comes back in 4.2 seconds with the query on show: the scrap is new, the copper index has moved cost of sales for three months, and the absorption charge appeared in June. The draft goes to the plant controller with a clock of one working day.
The plant controller confirms two sentences and rewrites the one on the scrap, because the tool change was planned and the yield loss came from the first lot after it. J. Harrison approves the commentary at 16:05; the pack is assembled with every explanation linked to its owner and its transactions, and the questions she asked are saved as a walkthrough for October.
The one large named result in the chain is here: HPE’s finance team cut its financial reporting cycle time by 40% with agents built with Deloitte, replacing days of analysis and about a hundred pages of slides, and 54% of the finance chiefs surveyed plan to integrate AI agents in 2026 (CFO Dive, 12 February 2026) 47. The initial target was 50% (Deloitte, 18 March 2025) 48.
Reporting shows AI adoption comparable to planning and forecasting, where 19% of finance organisations are scaling and 22% piloting (The Hackett Group, 19 March 2026) 6, and 53% of accounting teams name financial reporting as an automation opportunity (FloQast, 11 August 2026) 9. Deloitte lists drafted variance explanations among the controllership tasks generative AI now performs with a person validating each (Deloitte, 20 April 2025) 44, and KPMG’s agentic month-end runs flux analysis exception-first (KPMG, 19 March 2026) 46.
The vendor field is wide: Oracle’s Ledger Agent alerts accountants to variances defined in plain language and names variance reasoning as its next direction (Oracle, 1 December 2025; 20 September 2026) 18 19. Microsoft explains the drivers in natural language (Microsoft, 20 October 2025) 23, FloQast traces material variances to source transactions (FloQast, 17 September 2025) 36, BlackLine’s Verity Flux and Narrate draft the explanations and the narrative 29, Trintech prepares the first variance report (Trintech, 9 April 2026) 34, and Workiva’s Tie-Out Agent surfaces explanations (Workiva, 29 July 2026) 52.
The pain is described more than measured across the market, which keeps the use case fourth. For a manufacturer the explanation lives in the supply chain: scrap and yield on the production order, purchase price variance by material and supplier, absorption from an idle line, inventory reserves. Those objects are the ontology’s own, and the drill from the ledger line to the production order is one query across the ERP, the planning suite and the shop-floor system.
After the trial balance freezes on workday 3 or 4, analysts in financial planning and the plant controllers export balances by account, cost centre and product line, build the comparison with the prior period and the budget in a spreadsheet, apply a threshold such as 5% and $50,000, and start asking.
Each material movement means a drill into the ledger, then an e-mail to the cost-centre owner, the plant controller or the buyer, then a wait. Explanations come back in different words and at different depths, get rewritten into commentary in a document or slides, are consolidated into the management pack and reviewed with the finance lead and the CFO the day before the executive meeting.
Two to four days pass from trial balance to pack; at HPE it was days of work and about a hundred pages of slides (CFO Dive, 12 February 2026). The supply chain questions are the hard ones, because scrap, yield, purchase price variance and absorption sit in controlling, production and purchasing, and the person who can explain them is on the plant floor.
Workday 3 to workday 6 as it runs by hand: the questions travel by e-mail and the answers are rewritten twice. Illustrative; the mechanics are the note’s.
The six stages carry this use case’s objects, questions, terms, trigger and measure. Everything below is a design; the supply chain objects behind the explanations (production order, material, plant, supplier, purchase order line) are the ontology’s own, and ledger account, cost centre, budget version and variance would be added as a finance extension.
Three movements explain $456,000 of the $471,000 rise. The rest is inside the threshold.
| Driver | Amount | Behind it |
|---|---|---|
| Scrap on 7741-X | $212,000 | yield 95.4% after the tool change on the 12th; first lot after it |
| Purchase price variance · Brightline Metals SA | $148,000 | copper index up 6%; 112 lines received on the 27th |
| Absorption charge · line 3 | $96,000 | idle in week 3; 41 hours below plan |
The copper index has moved cost of sales for three months running; the absorption charge appeared in June; the scrap is new. The drafted explanation now says so, and the plant controller has it with a clock of one working day.
The two questions of the scene, as ChatLima would answer them: the drill to the production order and the supplier, the systems read, the time, and the follow-up from the saved list. Designed, yet to be built; showcase estate.
The market runs this at inform and draft-and-approve: Oracle alerts and summarises 18, Microsoft explains drivers 23, FloQast, BlackLine and Trintech draft explanations for review 36 29 34, and Deloitte recommends a person validating every draft 44. The blueprint agrees: variance commentary drafting is designed at A2, because every sentence that reaches the executive pack carries an owner’s name, while the analytics roles behind it are designed at A3 and A4. It is the one use case here whose lead role stays at A2 by design.
| Element | Setting | Set by | Changes |
|---|---|---|---|
| Objective | Explain every material movement from its transactions before the pack goes to the executive committee, each sentence confirmed by its owner | J. Harrison, finance lead | New version |
| Data scope | Ledger and controlling balances, budget and forecast versions, production orders, standard costs, purchase price variances and inventory values in SAP S/4HANA, the Oracle Fusion division and the planning suite | A. Novak, data lead | New version |
| Rules | The flux policy and its planning table: 5% and $50,000 against prior month and budget; draft from the transactions; confirm within one working day | J. Harrison with the Planner Agent | New version, simulated first |
| Trigger | Trial balance freeze on workday 3 · flash run on workday 1 · budget or forecast load · on demand | J. Harrison, finance lead | At once |
| Actions permitted | Assign a variance to its owner with a clock · write confirmed commentary into the pack · draft an adjusting entry for the reconciliation queue | J. Harrison, owner | New version |
| Limits | Threshold 5% and $50,000 · clock one working day · a sentence reaches the pack with an owner’s confirmation alone · two named people on a restatement | J. Harrison with the CFO | New version |
| Escalation | A clock that runs out, to the plant controller’s manager and then J. Harrison · a variance that is an error, to the reconciliation queue | J. Harrison, owner | At once |
| Level | A2 Supervised with the plant controller or cost-centre owner as the named approver of each draft, and J. Harrison on the pack; designed to stay at A2 | The CFO | Up on evidence, down in one step |
| Feedback | Every edit to a draft is recorded and trains the next; confirmed explanations become templates | 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 Record-to-Report lists the rest of the map for this chain.
| Measure |
|---|
| reporting cycle time from trial balance to pack (HPE: 40% shorter) |
| share of material variances with a confirmed explanation before the review meeting |
| edit rate on drafted explanations |
| time per explanation (adopters: hours to seconds for the draft) |
| adjusting entries found through the flux review |
| questions answered on demand in ChatLima against questions e-mailed |
Baseline the reporting cycle time first: the working days from the trial balance freeze to the pack in the CFO’s hands, for the last six closes, and beside it how many of the material variances had a confirmed explanation before the review meeting. Both come from the calendar and the pack, so they cost an afternoon to collect, and they are the two numbers HPE’s 40% is made of.
The rules engine finds the movements and drills the transactions; the small language model inside your estate words the draft from those facts. The owner confirms or rewrites, and the confirmed sentence is what the pack carries.
Through the ontology: the cost of sales line links to the production order, its material and its scrap, and to the purchase order line and its supplier, read in place across the ERP, the planning suite and the shop-floor system.
The owner replaces the sentence, and the replacement is what goes to the pack under the owner’s name. The edit is recorded, and the next draft for that area starts from it.
The commentary explains what moved this month: scrap after a tool change, a copper index, an idle line. Each began as a product or process decision months earlier. Where does that chain start, and what would Juno do there?
From the roadmap decision to production release: regression triage, engineering change impact, end-of-life parts and launch readiness.
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.