01Overview
The chain’s stock and sales already lived in its retail system. The analyst we built reads them and suggests what to reorder, move between stores or mark down. Each suggestion comes with its evidence, and a person makes the call.
The evidence layer
Versioned readings of stock and sales, with how fresh and complete each source is.
The recommendation engine
Fixed calculations and a rule book the chain approves, so every figure can be traced.
The working views
A worklist, advice cards and stock, rule and catalogue screens, on desktop and phone.
02The challenge
Repeat orders were decided after the stock ran out.
Repeat orders were decided too late, so best-sellers ran out before new stock arrived. Each decision was put together by hand from the chain’s system, and the knowledge sat with a few experienced people. Stock history was a monthly export, which cannot show how long an item was missing from a shelf.
What the client wanted to achieve
Decide in time
Advice arrives while a repeat order or a move is still possible.
Evidence on every card
Every figure links back to its reading, its rule and its run.
A person always decides
Nothing is ordered, moved or closed by the system itself.
Rules as settings
The team changes any business value on the rule screen, without a developer.
How it worked before
- Repeat orders decided after stock ran out
- Broken size ranges found store by store
- Stock history kept as a monthly export
03How we built it
- 01
Mapping how decisions get made
We rebuilt the buyer’s own rules from his working notes and a discovery of more than 85 questions.
- 02
Every engineering decision first
A decision map settled the hard questions before any code, each one written down.
- 03
Prototyping the risky parts
The transfer method, the order of the evidence and the wording of the AI explanations, each tested on its own.
- 04
The screens, early
The worklist and the working views were built ahead of the engine, so the team could react to them.
- 05
A read-only link to their system
Readers built against a copy of the chain’s system, plus a recorder in the chain’s own cloud that keeps every stock reading.
- 06
The engine at full scale
Repeat orders, transfers, markdowns, carry-overs and findings, tested on a chain of the real one’s size.
- 07
The rules, reworked with the buyer
A working session with the buyer replaced several first-draft rules with his own.
04How it works
Step 01Read
Stock and salesRead-only, from the chain’s own system.Stock historyA recorder keeps every reading.Step 02Check
Fresh and completeStale or partial data stops the advice it affects.Blank is not zeroMissing, invalid and zero are kept apart.Step 03Calculate
Sell-through and sizesPer store and per colour, from fixed calculations.One transfer planNo unit is promised twice.Step 04Decide
Ranked adviceWith its evidence and its limits.RecordedAccept, change or decline, and what was moved.
Reorder in time
Raised as soon, now or urgent, while there is still time to produce.
Move it where it sells
Units from the logistics centre or another store, with quantities per size.
Mark down what is slow
Items that have stopped selling, with the readings that show it.
Learn for the next buy
Excess by size or colour, and lessons for the next purchase.
Every assessment ends in one of five answers
- Advice
- Nothing needed
- No feasible action
- Does not apply
- Could not assess
05The outcome
Each morning the team opens a ranked worklist: what to reorder, what to move and where, what to mark down. Every card carries the readings and the rule behind it, a person accepts, changes or declines it, and what was done is recorded.
| Before | After |
|---|---|
| Conclusions put together by hand | A ranked worklist, each item with its evidence |
| Repeat orders decided after stock ran out | Advice raised while there is still time to produce |
| Broken sizes found store by store | Transfer lines with quantities per size |
| Decisions kept in people’s heads | Every decision, note and rule change recorded |
| Monthly stock snapshots | Stock read twice a day, with gaps in the data shown |
06Key figures
07Technology
What the system runs on, for the technical reader.
- Cloud
- AWS, in the chain’s own accountContainersManaged PostgreSQL
- AI
- Claude, for the written explanations only
- Data
- PostgreSQLPrismaA job queueA read-only link to the chain’s retail system
- How we build
- Next.jsNestJSTypeScriptVitestPlaywrightPython prototypes
08What comes next
Next, the chain’s real stock and sales history is loaded in, and the team reviews real advice cards before the pilot begins.
09About
About the client
A fashion retail chain with its own production, selling through its stores. We name a client only with their permission.
About MYBB
MYBB builds systems around how a business already works. We diagnose it with MYBB ROIQ, build the system, launch it with the team and keep running and improving it.