Seven decisions a retail group makes every day, taken properly, every time.
This is a working document rather than a deck. Set the group size in the dark panel and every number on this page, from the arithmetic inside each function to the payback month, recalculates against it. Every figure carries the calculation that produced it, and any small letter i opens a plain English explanation. Every function is delivered on the DeployOne platform, and section 04 sets out the retail products and platform arms that run underneath each one.
How it works, in four steps
The same shape sits behind all seven functions. Data in, one decision layer, an instruction to a named person, and a loop that closes.
Three inputs
Transactions, stock positions and people signals arrive continuously from the tills, the warehouse and the roster. Nothing is keyed in by hand.
One decision layer
Every module reads the same numbers, so a markdown decision and an allocation decision can never disagree about what is in the building.
An instruction, not a report
The output is a named action for a named person: move these units, call this customer, cover this shift. A dashboard nobody acts on is worth nothing.
The loop closes
Whether the instruction was followed, and what happened after, feeds straight back in. The model gets sharper every week without a consulting engagement.
The assumptions, all of them editable
Nothing here is hidden. These are the benchmark ranges for Gulf multi-brand retail; move any of them and the entire document follows.
Benchmark assumptions
Typical ranges for Gulf multi-brand retail. Each slider opens at the market figure, which is also marked on the track, so any deviation is visible. Drag one and the whole document updates.
| Line | Percent | Value |
|---|---|---|
| Discount given away i percent of gross sales | 17.0% market 17.0% · at market | AED 624.3m |
| Store payroll i percent of revenue | 10.0% market 10.0% · at market | AED 367.3m |
| Rent and occupancy i percent of revenue | 13.0% market 13.0% · at market | AED 477.4m |
| Marketing and trade spend i percent of revenue | 3.5% market 3.5% · at market | AED 128.5m |
| Supply chain and logistics i percent of revenue | 3.5% market 3.5% · at market | AED 128.5m |
| Head office and shared services i percent of revenue | 5.5% market 5.5% · at market | AED 202.0m |
| Inventory held i percent of revenue, balance sheet | 18.0% market 18.0% · at market | AED 661.0m |
| Customer care cost i percent of revenue | 0.30% market 0.30% · at market | AED 11.0m |
| Operating profit margin i percent of revenue | 9.0% market 9.0% · at market | AED 330.5m |
Derived shape of the business
| Group revenue i | AED 3.67bn |
| Gross profit i | AED 1.65bn |
| Operating profit i | AED 330.5m |
| Stores i | 120 |
| Shop floor headcount i | 5,000 |
| Hires per year i | 2,000 |
| Loyalty members on file i | 2,000,000 |
| Average transaction value i | AED 349 |
Store count is estimated from revenue per door unless you type the real number into the panel, in which case yours is used everywhere. The first working session replaces all of this with the client's own trial balance.
The seven functions
Pick a function below. Each one states the problem in plain language, shows what breaks today, shows the screen the client would actually use, exposes the arithmetic, and names the agents that run it.
Merchandising, buying and markdown
In plain English: this is the team that decides what to buy, how much of it, and which shops it goes to. They commit the money six to nine months before the product arrives, based mostly on what happened last year. When they get it wrong, the only fix available is to cut the price until the stock moves. That price cutting is called markdown, and across a group this size it currently gives away AED 624.3m a year.
| Line i | How the base is calculated i | Improvement applied i | Result i |
|---|---|---|---|
| Discount currently given away | AED 3.67bn x 17.0% | reduce by 6 to 12 percent | AED 37.5m to AED 74.9m |
| Full price sell-through improvement | AED 3.67bn x 1 to 2% x 45% margin | range applied to the base | AED 16.5m to AED 33.1m |
| Terminal stock written off | AED 661.0m stock x 8% terminal | reduce by 15 to 25 percent | AED 7.93m to AED 13.2m |
| Annual envelope i | sum of the lines above, at steady state | midpoint AED 91.6m | AED 61.9m to AED 121.2m |
The buy is committed six to nine months ahead, so year one only catches one season of buying decisions. The markdown clock lands sooner than the buy does.
| Product | Doors | Weeks cover | Recommendation |
|---|---|---|---|
| Linen shirt, mid blue | 42 | 14.2 | Hold, no discount |
| Cotton chino, stone | 38 | 21.8 | Cut 20 percent in week 3 |
| Occasion dress, black | 26 | 31.4 | Cut 35 percent now |
| Knit polo, navy | 51 | 9.6 | Reorder, selling ahead |
| Lightweight jacket | 33 | 27.1 | Move 340 units to Abu Dhabi |
The product stack that runs them
Ten platform arms underneath, twelve retail products on top. Each one states what it reads, what it writes, how it is modelled and which of the seven functions it powers.
Predict · Demand Forecasting
A hierarchical forecast that reconciles from SKU-store-day up to brand and group, so the number the buyer sees and the number the chief financial officer sees are the same number. New products with no history are forecast by attribute similarity rather than by a planner's guess.
- Gradient boosted trees plus a temporal fusion transformer, ensembled and reconciled with MinT so the hierarchy adds up
- 500+ features: price, promo depth, weather, footfall, school terms, Ramadan and Eid shift, payday, mall events, competitor price moves from Sense
- Cold start on new lines by attribute embedding nearest-neighbour over past seasons
- Backtested rolling-origin, held to weighted MAPE and pinball loss at the 80th and 95th percentile, not to a single point estimate
Predict · Price and Markdown Optimiser
Discounting stops being a July calendar and becomes a per-product clock. Elasticity is estimated per product family and per emirate, then an optimiser picks the discount path that clears the stock by the cut-off date while giving away the least margin.
- Bayesian hierarchical elasticity with partial pooling, so a thin-data line borrows strength from its family instead of producing nonsense
- Mixed integer programme over the markdown path with constraints on ladder steps, brand floor prices, contractual partner rules and terminal date
- Competitor price and range scraped continuously by Sense, matched by product embedding rather than by exact name
- Every recommendation ships with a counterfactual: hold, cut now, cut later, and the modelled margin of each
Custom · Allocation and Size Curve
Initial allocation and in-season replenishment set from each door's own selling history rather than from square footage. Size curves are estimated per catchment, which is why the small sizes stop piling up in the wrong mall.
- Constrained flow optimiser across warehouse and store nodes, minimising expected lost sales plus transfer cost
- Dirichlet size-curve estimation per door cluster, clustered on demographics and past size mix
- Inter-store transfer proposals costed against the markdown they avoid, so a transfer only fires when it beats the discount
- Runs against the same forecast object, so allocation and buying never disagree
Context · Shelf and Store Vision
The cameras already installed become a sensor: on-shelf availability, planogram compliance, queue length, dwell by zone and staff presence on the floor. No customer is identified and no face is stored; the output is counts and events, not people.
- On-edge detection and tracking at 4 to 8 frames per second per camera, only events leave the store
- Planogram compliance by template matching plus a vision language model for the awkward cases
- Queue and dwell analytics feeding the roster optimiser directly, so a long queue at 6pm changes next week's shift
- Privacy by design: anonymised counts, no face templates, regional data residency
Predict + Connect · Workforce and Roster
Footfall is forecast in fifteen minute buckets per door, converted into cover requirement, then solved into shifts inside labour law, visa class, contracted hours and employee preference. The store manager gets a proposed roster to approve, not a puzzle to solve.
- Fifteen-minute footfall forecast per door with event and holiday regressors
- Shift construction as a set-cover integer programme, constrained on UAE and KSA labour rules, breaks, prayer times and maximum consecutive days
- Conversion attribution splits under-performance into traffic, staffing, stock and price before an alert is raised
- Daily voice check-in with every store manager in their own language via Connect, transcribed and scored automatically
Brain · Customer Data and Identity
Most groups have the same shopper five times over, once per brand. Identity resolution stitches them into one profile, then churn, lifetime value and next best action are modelled on the stitched record rather than on fragments.
- Probabilistic identity resolution on phone, email, card token and device, with a deterministic override table
- Survival model for churn and a gradient boosted lifetime value model, retrained weekly
- Next best action ranked by expected incremental margin, not by propensity, so it stops discounting people who would have come anyway
- Consent and preference held per channel, enforced at send time
Connect · Retail Customer Care
Order status, exchanges, refund eligibility, stock checks across doors, appointment booking and win-back calls. The agent calls the order system, writes to the CRM and hands to a person the moment it should.
- 10,000 concurrent calls, sub-400ms end to end, 74 languages including Gulf and Levantine Arabic with natural turn taking
- Real tool use: order lookup, stock check, refund policy engine, CRM write, live warm handoff with full context
- Every human conversation scored by Converse for objection handling and compliance, feeding coaching
- Barge-in, interruption handling and accent robustness tested per market before go-live
Engage · Marketing and Trade Spend
Marketing mix modelling for the long run, geo holdouts for the truth, and an always-on creative pipeline built from what customers actually said on calls and in reviews rather than from a brand workshop.
- Bayesian marketing mix model with adstock and saturation priors, refit monthly
- Geo-based holdout tests for genuine incrementality, run continuously rather than once a year
- Trade income tracked against supplier agreement terms, so the money owed is claimed
- Creative generation for reels, images and copy, with performance agents reallocating budget on a weekly cadence
Custom · Supply Chain Control Tower
Purchase order to shelf, one timeline. Arrival dates are predicted rather than promised, and when a container slips the system already knows which drop, which campaign and which doors are affected.
- ETA prediction per shipment from carrier events, port congestion and historic lane performance
- Exception routing with an owner, a deadline and an escalation path, not an inbox
- Safety stock set per SKU-door from forecast uncertainty rather than one blanket rule
- Landed cost and duty modelled so the margin impact of an air-freight decision is on screen when it is taken
Custom · Back Office and Shared Services
Supplier invoice capture and three-way match, candidate screening by voice at Gulf turnover rates, internal helpdesk deflection and month-end reconciliation drafted for a human to approve.
- Document understanding over invoices and delivery notes with three-way match against PO and receipt
- Voice screening interviews with structured scoring and bias controls, shortlisting to the hiring manager
- Internal helpdesk RAG over policy, HR and IT documents with citation back to the source clause
- Human-in-the-loop approval on anything that moves money
Sense + Predict · Network and Property
Catchment modelling, cannibalisation between your own doors, site scoring for new locations and a renewal model that says what a given store is worth to you at what rent.
- Geospatial catchment model on mobility, population and income layers, with a Huff-style gravity allocation
- Cannibalisation estimated by holdout comparison when a nearby door opened, not by assumption
- Store-level profit and loss projection under rent scenarios, including a walk-away point
- Mall footfall and competitor opening intelligence crawled continuously by Sense
Dash · The Command Centre
Every module above surfaces into a single command centre: the buyer sees their lines, the store manager sees their doors, the chief executive sees the group. Same numbers, same definitions, one place to accept or override.
- Role-scoped views with row-level permissions inherited from Brain governance
- Every figure traceable to the query that produced it, so a disputed number is settled on screen
- Accept and override captured with reason, feeding the retraining set
- Alerting by exception, on the channel the person already uses
The value stack
Envelope at steady state on one side, the number we are prepared to put in a contract on the other.
What it costs and when it pays back
Priced as a share of value claimed rather than a flat licence, so the client pays out of the saving.
Commercial model
12% value shareDelivery is charged once per module and scales with group size. The recurring fee is priced as a share of the value actually claimed in year one, so the client pays out of the saving rather than ahead of it. Drag the share to test the pricing.
| Module | Build status | Delivery, once | Annual fee | Net to client |
|---|---|---|---|---|
| Assortment OS | New build, forecasting spine exists | AED 3.49m | AED 1.32m | AED 9.67m |
| Store OS | Adaptable from OccupancyOS | AED 2.20m | AED 1.34m | AED 9.80m |
| Customer OS | Built and demoable today | AED 918k | AED 1.05m | AED 7.67m |
| Spend OS | New build, Listen layer exists | AED 3.49m | AED 595k | AED 4.36m |
| Inventory OS | New build | AED 3.49m | AED 398k | AED 2.92m |
| Operations OS | Built and demoable today | AED 918k | AED 986k | AED 7.23m |
| Network OS | Adaptable from OccupancyOS | AED 2.20m | AED 144k | AED 1.05m |
| Total, 7 modules in scope | AED 16.7m | AED 5.82m | AED 45.6m |
The first twenty four months
Value is ramped over six months rather than switched on, because no module reaches its run rate in week one. Delivery is taken as a single charge at the start, which is the harshest reading of the case.
The target list
Tap any group to load its approximate revenue into the model and see the case sized against it.
Who this is built for
tap a row to size the model on itEach row loads that group's approximate revenue into the model above, so the value stack rebuilds against a real target rather than a round number.
| Group | Base | Revenue used | Year one value | Why them |
|---|---|---|---|---|
| Alshaya Grouplarge | Kuwait | $6bn to $12bn | AED 51.4m | Very large franchise portfolio, high store count, heavy seasonal discount exposure across fashion and food. |
| Majid Al Futtaim Retaillarge | UAE | $7bn to $9bn | AED 51.4m | Grocery scale plus lifestyle. Already invested in analytics, so the conversation starts further along and the bar is higher. |
| Lulu Grouplarge | UAE | $7bn to $8bn | AED 51.4m | Grocery scale, thin margins, and the largest labour base of any group on this list. |
| Al-Futtaim Retaillarge | UAE | $5bn to $7bn | AED 51.4m | Multi-format across fashion, electronics and furniture. Strong existing data estate to build on. |
| Landmark Grouplarge | UAE | $3bn to $4bn | AED 51.4m | Own brands rather than franchise, so full control of buying and pricing decisions. Large loyalty base. |
| Chalhoub Grouplarge | UAE | $3bn to $3.5bn | AED 51.4m | Luxury, where retention economics are strongest and service quality is a brand issue rather than a cost line. |
| GMGlarge | UAE | $2bn to $3bn | AED 51.4m | Sport, food and health across several markets. Distribution plus retail creates the stock problem in its sharpest form. |
| Al Tayer Groupmid | UAE | $1.5bn to $2.5bn | AED 51.4m | Luxury and department store formats with a mature loyalty programme to work against. |
| Apparel Groupmid | UAE | $1.5bn to $2bn | AED 51.4m | Very high brand count and store count relative to head office size, which is exactly where a decision layer pays. |
| Azadea Groupmid | Lebanon, UAE | $1bn to $2bn | AED 51.4m | Multi-country franchise operator with a complex transfer and allocation problem across borders. |
| Liwa Tradingmid | Abu Dhabi | $300m to $700m | AED 51.4m | Mid-size, faster decision cycle, likely the most realistic first signature on this list. |
| Rivoli Groupmid | UAE | $300m to $500m | AED 51.4m | Watches and luxury accessories. High value, low volume, so retention and clienteling dominate the economics. |
| A single brand inside a groupsingle | Any | typically $60m to $250m | AED 51.4m | One fascia rather than the whole group. Smaller committee, faster decision, and a live reference we can then take upstairs to the group. |