Sell-through by style, size-curve health and weeks-of-cover — every door, one board, every day the season is still yours to change.
Not at the end of the season. Not in next week's report. At 09:14 this morning, in six of twenty-two doors — while there is still something to do about it.
Six doors below reorder point on two sizes — detected by rule, not by a buyer noticing.
Allocation, not demand. The four fastest doors took the first drop and were never topped up.
84 units from the three slowest doors — holds full price eleven more days.
4.2 points of margin, given away in a markdown that did not need to happen.
| Rule | Threshold | Now | State |
|---|---|---|---|
| Gross margin · season-to-date | ≥ 48% | 52.4% | clear |
| Repeat purchase rate | ≥ 35% | 38% | clear |
| Full-price sell-through | ≥ 67% | 71.4% | clear |
| Demand forecast accuracy | ≥ 89% | 92.8% | clear |
| Model | Predicts | Accuracy |
|---|---|---|
| Size-curve drift | break risk by size | 93.4% |
| Sell-through | 8-week curve by style | 91.8% |
| Markdown timing | first-cut date | 90.2% |
One board, read three ways. The rule states the fact, the model states the risk, and Iris states the move — with the trade-off priced. Nobody has to assemble the three themselves.
Sell-through by style, size-curve health and weeks-of-cover — every door on one board.
The till a sales associate runs all day — and the twelve core capabilities every Zentallio retail sector ships with, identically.
Size and style live at the till. An associate can see cover across every door, so a lost sale becomes a transfer rather than a shrug.
Returns and exchanges are one flow. Apparel returns at volume, and an exchange that fails to find the right size is a customer lost twice.
Iris upsells against the size curve. The recommendation favours what is actually in stock in that customer's size, not what the planogram wishes were.
Financial, customer, operational and AI-learning on one live board — with Iris naming the style that just broke and the move that holds its price.
The fourth lens is the platform itself. Forecast accuracy and models live are reported alongside margin, because a board that cannot audit its own AI is asking for trust it has not earned.
Every door, one definition. 180 stores report the same measures, so a comparison is real rather than approximate.
Iris names the style. Not that sell-through dipped — which style, which sizes, which doors, and what the move is worth.
Trial balance, P&L, balance sheet and cash flow on a ledger that ties out to the penny — with season inventory valued as it actually sits, not as it was bought.
Inventory valued by age and cover. Season-old stock at four weeks of cover is not worth what it cost, and the ledger says so before the auditor does.
Markdown provision is computed, not guessed. The markdown-timing model feeds the provision, so the number has a basis.
Ask Iris for any number. Any door, any style, any period — answered in the conversation rather than a week after close.
Forecast to buy to allocation to the door — with the size curve carried through the whole chain rather than collapsed into a single unit count.
The buy is made in sizes. A style bought on the wrong curve cannot be rescued by allocation, so the curve is decided at the buy.
Lead time is part of the decision. Whether a break can be reordered or only reallocated changes the recommendation entirely.
Feeds allocation directly. Receipts post into the allocation engine, so the nightly run works from what actually landed.
Store standards, visual merchandising compliance, assets and energy across 180 doors — completion tracked live, exceptions escalated.
Visual merchandising, evidenced. A window set is completed with a photo against the guideline, not assumed from a distance.
Transfers are an operational routine. A nightly allocation run only works if the sending door actually picks and ships it.
Exceptions escalate themselves. The routine that didn't complete surfaces to a human; the rest stays quiet.
Rostering, attendance, payroll and working-time compliance measured against forecast footfall — because in apparel, conversion is a staffing decision.
Staffed to conversion, not to footfall. A door converting at 24.6% is limited by fitting-room service far more often than by traffic.
Peak weekends planned from the forecast. Launch weekends and markdown events are demand events, and the roster treats them as such.
Flagged before the shift. An understaffed Saturday is caught while the roster can still change.
Grouped by what they act on. Pick one and it runs.
Thresholds that fire on the shift the breach happens (the “Zen Rules” layer) — no model lag, no waiting for a report. Two solutions run on this layer, and both of them move stock.
M and L below reorder point in 6 of 22 doors. Detected at the threshold, not at the weekly review — which is the difference between a transfer and a markdown.
| From | To | Units | Reason | State |
|---|---|---|---|---|
| Lille Grand Place | Paris Rue Cler | 38 | conv 2.1× | queued |
| Nantes Atlantis | Lyon Part-Dieu | 26 | conv 1.8× | queued |
| Rennes Colombia | Bordeaux Centre | 20 | breaks S | review |
| Toulouse Capitole | Lille Grand Place | 31 | conv 1.6× | queued |
| Other 138 moves | — | 1,765 | — | queued |
The run is weighted by conversion, not depth — Lille holds the units but Paris sells M and L twice as fast. One move is held back because it would open an S gap at Rennes to close an M gap at Bordeaux; that trade is not worth making.
Two businesses share one shop floor. Core replenishes forever on a predictable curve; fashion has one life and no second chance. The rules that govern them cannot be the same rules.
Every style is classified core or fashion, and the classification decides which replenishment logic it obeys.
Core lines hold a continuous cover target; fashion lines run to a sell-through curve with a planned end.
Floor-space and allocation share between the two is governed as a rule, so a strong fashion drop cannot quietly starve the core basics that carry the margin.
Drift in the mix is surfaced per door, because a store that has become 70% fashion is a different business from the one that was planned.
Markdown exposure is computed on the fashion portion only, which is the number that actually threatens the season.
What is modelled before it lands (the “Zen Models” layer) — modelled early enough to act on, with the cause named and the move costed. Four solutions run on this layer.
| Door | M | L | Cover | Status |
|---|---|---|---|---|
| Paris Rue Cler | 0 | 2 | 1d | gone |
| Lyon Part-Dieu | 1 | 3 | 2d | critical |
| Bordeaux Centre | 2 | 4 | 3d | thin |
| Lille Grand Place | 3 | 5 | 4d | thin |
| Other 18 doors | 14 | 19 | 11d | healthy |
The break is not spread — it concentrates in the four doors that took the first allocation, which sold through fastest and were never topped up. That pattern says allocation, not demand. Moving 84 units from the three slowest doors holds full price 11 more days; markdown instead costs 4.2 points.
An eight-week curve per style, projected from its first days on the floor — so a style is known to be ahead or behind its curve while both are still recoverable positions.
The curve is projected per style and per door from early sell-through, comparable styles and the season's shape.
A style running ahead of curve is a reorder or a transfer decision; one running behind is a markdown clock starting.
The model trace is inspectable — which comparable styles, which signals, how much each contributed — because a buyer will not act on a number they cannot interrogate.
Forecast accuracy is reported on the board itself, currently 91.8%, so confidence in the number is calibrated rather than assumed.
The curve feeds allocation, markdown timing and the inventory provision, so all three work from one projection.
The first cut is the expensive one. Predicting when to take it — and how deep — is worth more than any subsequent decision in the season, because everything after it is recovery.
A first-cut date is predicted per style, currently at 90.2% accuracy, rather than fixed to a calendar week the whole estate shares.
Depth is modelled against remaining cover and residual demand, so the cut is deep enough to clear and no deeper.
Transfer is always evaluated first — on AW-4412, moving 84 units holds full price 11 more days against 4.2 points of margin lost to a cut.
Markdown is planned per door, since a style dead in Paris may still be full-price in Toulouse.
The engine feeds the accounting provision directly, so the markdown number in the ledger is the same one the buyer is working to.
The fitting room is where apparel is actually sold or lost. A garment tried and not bought is the strongest signal in the store — and in most estates, the least recorded.
Try-on to purchase is measured per style and per size, which separates a fit problem from a demand problem.
A style that is tried often and bought rarely is a garment issue, not an allocation one — and no amount of transferring fixes it.
Requests for a size that was not on the floor are captured, which is the demand a stock report structurally cannot see.
Conversion by door is read against staffing, since fitting-room service is a labour decision more than a merchandising one.
The signal feeds back into the buy, so a fit failure is corrected in the next order rather than repeated at scale.
Not a screenshot — the actual agent, reading the same board the rules and models write to.
The sector solutions sit on top of these. They are not an upsell and they are not configured per customer — every Zentallio retail deployment ships with all twelve.
You configure a sector playbook, not a custom project. Iris applies it herself — agentically, from day one.
Onboarding is agentic. Iris connects the till, the product hierarchy and the size curve directly — no manual data mapping.
Rules go live before the models do. L1 thresholds work from day one; the L2 models earn their accuracy over the first season and report it on the board.
Live in weeks, learning from day one. Go-live is a configuration, not a project plan.
The first line of support is agentic — Iris resolves most of it herself. Our engineers pick up from there.
Layer 1 — Iris, 24/7. Configuration questions, anomalies and routine issues resolved directly, instantly.
Layer 2 — our engineers. Anything Iris can't close escalates automatically to a Zentallio engineer.
No blank tickets. Every escalation arrives with Iris's own diagnosis — engineers start from an answer.
The fastest doors take the first drop, sell through, and are never topped up — so the best sellers break first.
A break read as demand rather than allocation, answered with a markdown that was never necessary.
A first cut taken on a calendar week the whole estate shares, regardless of what each style is doing.
Thresholds fire on the shift the breach happens, and a move that would break a size at the sender is held for review.
Size-curve drift, sell-through and first-cut date are modelled at 90–93% accuracy, reported on the board.
“Why, and what do I do?” — answered with the cause, the move and what the alternative costs.
Apparel & Ready-to-Wear is the first of nine Fashion Retail sectors. The layers, the six products and Iris herself are the same ones running across ten Food & Beverage sectors.