This worked case is part of exception handling records for AI agents. That page sets out why exception histories matter. This one follows a single promotion failure from the rule edit to the fix, then shows retail systems consultants how to recognize similar cases at a client.
The weekend two offers stacked (Illustrative)
Invented for this page. A fictional 900-person regional grocery chain with 38 stores runs its promotions from a central engine that feeds the POS. For a four-week campaign it sells a breakfast cereal on a buy-two-get-one-free offer. Separately, its loyalty app issues a coupon for 20% off the same cereal to lapsed members. The campaign plan says the two do not combine.
On Friday afternoon a pricing analyst edits the loyalty coupon to extend its end date. The edit resets the coupon's combination setting, and from Saturday morning the POS applies both discounts. A shopper who buys three boxes pays for two and then gets 20% off what is left. By Sunday, store managers have noticed margins on the shelf report and customers asking why the price is so low.
Nothing is broken in any one system. The engine did what its configuration said, the POS applied what the engine sent and the receipts are accurate. The exception is the gap between the rule that was intended and the rule that was in force.
Who decided what, in order
Each row is one decision, the role that made it and the record it left.
| When | Decision | Decided by | System | Record left |
|---|---|---|---|---|
| Campaign launch | The multibuy and the coupon do not combine | Category manager | Promotion engine | Rule versions with combination settings |
| Friday 3 pm | Extend the coupon's end date | Pricing analyst | Promotion engine | New coupon version; combination setting changed |
| Saturday to Sunday | Honor the price at the till; no refusals | Store managers, following store policy | POS | Receipts with both discount lines |
| Sunday | Correct a few mis-scanned baskets on request | Store managers | POS | Returns and price overrides with manager ID and reason code |
| Sunday evening | Escalate the margin loss to head office | Regional operations manager | Email and back office | Incident note with store list |
| Monday 9 am | Restore the non-combining setting; do not claw back from customers | Commercial director | Promotion engine | Corrected coupon version; approval note |
| Monday onward | Confirm single-discount receipts in every store | Pricing analyst | POS reports | Receipts with one discount line |
Two decisions carry the case. The store managers chose to honor the advertised price rather than argue at the till, a customer decision. The commercial director chose to fix the rule and absorb the weekend's loss, a commercial decision. Both are judgments that a rulebook alone would not predict.
Which record holds each decision
A reviewer needs four kinds of record, and each sits in a different place.
- What was intended, and what was in force: the rule versions. Shopify, for example, gives each discount its own Combinations setting, and applies product discounts before order discounts, a sequence that also covers Shopify POS orders. A stacking error is a configuration fact. It is only reconstructable if the version in force on Saturday was kept, not just today's corrected one.
- What actually happened at the till: the Saturday and Sunday receipts with both discount lines printed against the same box. Receipts that show only a net price would make the weekend look like a pricing typo rather than two offers combining.
- What stores did about it: the corrections. In Shopify POS, a return or refund is started from the original order, so the correction points back to the receipt it fixes. Reason codes and manager IDs show who decided.
- How it was settled: the corrected rule version and the approval note, then the receipts that show the fix held.
If the Saturday rule version is gone, the case collapses into "prices were low one weekend". The system map for promotion records sets out the keys that join these records.
Why a stacking case teaches more than a clean sale
A clean promotion teaches one rule: the engine sets the price. This case shows people handling a rule that failed silently. There is an edit with an unintended side effect, a policy choice at the till, a set of corrections that cite their receipts, a decision not to recover money from customers and a fix that was checked in every store. Each step can be tested against a record made by the person responsible.
That is the kind of history AI developers describe needing when they need examples of real work rather than idealized procedures; why AI agents need work data explains the demand in general terms. It does not mean any buyer wants a given company's records. That is decided only after SourceX reviews them. The finance side of similar cases appears in deductions management, where supplier-funded promotions end up.
Signs a retail client has cases like this
Retail systems consultants hear about these cases during promotion engine or POS projects. Without opening a record, ask the retail operations director:
- Does the promotion engine keep dated versions of each rule, or only the live one?
- Do receipts record which promotion applied on each line, or only the final price?
- Do overrides and returns carry a manager ID, a reason code and the original receipt number?
- When a promotion misfires, where is the decision written down: the engine, email or a meeting?
- Roughly how many promotions a year, and how often do they misfire? A range is enough.
The retail chain assessment guide covers the wider picture. Then ask who can authorize a licensing discussion: the CEO, or the owner-CEO in an owner-run chain; the who-can-license decision tree maps that path. The company needs 50+ full-time employees at peak (contractors excluded), documented operating history, rights to license and an authorized contact. Companies in the U.S., Canada and the EU qualify, as do companies in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia when at least 70% of the data is in English. Loyalty identifiers, supplier-funded promotion terms and franchised stores all need rights review, so what remains licensable is unknown until SourceX reviews it. Partners never ask for or handle raw client records. More sector context is on the retail operations industry page.
Next step
Refer a company. No cap on rewards. Default reward on a company's first qualifying transaction: the greater of $50,000 plus 5% and 15% of Actual Supplier Payout, with no cap; further qualifying transactions first agreed within 12 months of the first Closing pay 15%. Eligibility and payment conditions apply, including the USD 100,000 Minimum Transaction, and a signed Schedule A or addendum can differ. A lead, meeting or signature alone does not trigger payment.
If a retailer you work with can show the rule in force, the receipts and the corrections for cases like this, register as a partner, check it against who qualifies and read the program terms before introducing its decision-maker.