Monday brings 1,240 exception messages to one planner. Inside them sit 38 purchase order lines that can still be cancelled this week, worth $412,000, and the window closes on Thursday.
How much of a planner’s week goes to finding the problem before the decision? This story follows one planner’s Monday and one agent, from the question to the confirmed action, at the level you set.
The list is long, the decision is short, and the filtering in between is done by hand.
L. Tan, supply planner, opens the week’s exception list: 1,240 messages from the weekend planning run. A customer pushed a large order out on Friday, so the purchase orders placed for controller board 7741-X at plant 1210 now stand well above the need.
Some of those orders can still be cancelled at full credit, and for several of them the window closes on Thursday. Which lines remain inside their window, what they are worth and which supplier holds them: those answers sit in other systems.
So the week starts with a search. By 11:00 she has three exports open: open orders from SAP, the acquired division’s orders from Oracle, the coverage plan from the planning suite. The afternoon goes to matching lines in a spreadsheet, Tuesday to e-mails with five buyers, D. Okafor among them. By Wednesday afternoon the picture is complete, and the buyers key each agreed cancellation into the ERP by hand.
What she is looking for is small. Of the 1,240 messages, 212 touch an open purchase order. Of those, 38 lines can still be cancelled this week, worth $412,000 together, and 9 of the 38 run above $25,000 and need a buyer’s judgement. The other 29 follow a rule anyone on the team would state the same way.
Every fact exists in some system; joining the facts by hand is what fills the week.
Every fact L. Tan needs exists somewhere. The order quantity and the price sit in SAP, and the acquired division’s orders in Oracle under other field names. The coverage plan sits in the planning suite, and the cancellation window in a custom field.
The work is joining them. Each export arrives with its own keys, its own units and its own idea of a delivery date. The spreadsheet is where the meaning is rebuilt by hand, every week, by the person who can least afford the time.
The waiting costs more than the joining. Five buyers each receive a list on Tuesday and reply when the day allows; the reply is a short decision, cancel or keep, and the day it waits is a day the window is closing.
The research below describes the same desk. Planning runs in spreadsheets and in days; the money for AI has been spent; the missing return is blamed on the join to the systems that hold the facts.
Dated source: publisher and date in each line.
Each of the four is enough on its own to stall a project.
| The cause | What it looks like on the desk | What the market reports |
|---|---|---|
| Data spread across systems | Her facts sit in SAP, Oracle and the planning suite. A tool that reads one of them, or a month-old copy, answers for that system alone; the join stays with her. | 56% of chief supply chain officers call integrating AI with legacy systems a major challenge (Gartner, 29 April 2026). dated source |
| The same word means different things | A delivery date is the requested date in SAP, the promised date in Oracle and the confirmed ship date on the supplier portal. Treat the three as one column and planners stop trusting the answers. | Organisations that prioritise semantics in AI-ready data will lift generative AI accuracy by up to 80% by 2027 (Gartner, 17 June 2025). dated source |
| Unclear limits on what the AI may do | A tool can recommend a cancellation in seconds. While who may act on it, and up to what value, stays open, buyers check every line themselves. | Trust in fully autonomous agents fell from 43% to 27% in a year (Capgemini, 16 July 2025). More than 40% of agentic AI projects will be cancelled by the end of 2027, for rising cost, unclear value or weak risk controls (Gartner, 25 June 2025). dated source |
| People’s corrections are lost | When D. Okafor keeps a line the tool wanted to cancel, his reason goes into an e-mail, and the tool proposes the same line again next Monday. | 95% of pilots show a profit effect too small to measure, and the tools that succeed retain feedback and learn (MIT NANDA, July 2025). dated source |
Read together, the four describe one project shape: a pilot on a copy of the data, meanings kept in the head of the person who built the columns, limits on a slide, and corrections left in a chat window.
Remove three and the fourth still stalls the project. So any answer has to remove all four at once, and the rest of this story is built around that requirement.
The same lines, the same buyer and the same rule; software does the search.
Now suppose software could act. L. Tan asks, in plain language, which open purchase order lines can still be cancelled this week and what they are worth; the answer arrives in 4.2 seconds and names the three systems it read. An agent drafts the 29 cancellations under $25,000, each with its reason. D. Okafor, the buyer who owns these lines, approves the queue in 22 minutes, and the ERP confirms each write-back. The 9 lines above $25,000 reach him with the reason for each.
In the showcase estate that software is Lima Juno, and the lower row is its Monday. Every step in it is an idea still to be explained: a question answered across three systems as one, an agent drafting under a written policy, a named approver, a confirmed write-back. The rest of the story explains them one at a time, on this same Monday, with the same 38 lines and the same two people.
The volumes come from the showcase estate and follow what planners describe: a long exception list, several systems to join and a deadline that arrives before the answer. Your own numbers replace them in the first week that chapter 9 walks day by day.
Ask which of the four it removed. Most remove one at best, and the other three stall the project. The next chapters show what removing all four takes.
What would software have to be able to do to take Monday off L. Tan’s desk, and how much of it should it be allowed to do?
The difference between a tool that answers and a tool that acts, and why that difference needs rules.
Each one builds on the one before it.
Reads your rows in place.
Connect and move ↗One record per real thing.
Master and govern ↗Apps for every decision.
Operate and decide ↗Asks and acts, at your level.
Ask and act ↗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.