Customer scanning a product with a mobile app inside a UAE mall flagship store

Retail AI · GCC personalisation, forecasting & pricing

Retail AI built for multi-banner, multi-country GCC retail — Arabic-aware, Ramadan-ready, measured through incrementality.

Personalisation that respects Khaleeji search behaviour. Demand forecasting that handles Ramadan, Eid, school terms and payday cycles. Dynamic pricing with merchandiser guardrails. Store-ops CCTV with no face data. Shipped on top of your existing Salesforce / SAP / Oracle / Adobe stack with a measured uplift in 90 days.

Request a 30-day incrementality assessmentWhatsApp our retail lead

Built for Arabic search. Built for Ramadan.

بحث

شو في عبايات جديدة لرمضان

عباية رمضانية — لون رملي

عباية تطريز ذهبي

عباية كاجوال أنيقة

Ramadan kaftan SS line

Sand-tone embroidered kaftan

Gold-thread Ramadan abaya

Casual modest dress

Embedding stack: AraBERT-v2 + mE5 · Khaleeji-dialect head

Ramadan demand · SKU 4-week forecast

Tightening band

MAPE: 6.4 %Lift vs incumbent: +11 pp

11
GCC retail & marketplace clients
+27%
Arabic search NDCG@10 vs Algolia baseline
−11 pp
Ramadan SKU-level forecast MAPE
+3.1%
Gross-margin uplift on 1,200 SKUs (brand house)

The painful problem

The CDP segments 14 M shoppers into ten lookalike tiers that all buy the same Ramadan SKUs.

The recommender shows the same three category banners to every visitor. Demand forecasting overstocks the UAE warehouse and stocks-out KSA every Eid. The merchandising team is rebuilding the buy plan in Excel by hand. The CDO has committed a personalisation revenue number at the investor day.

Failure mode 1

Arabic search

Off-the-shelf engines drop a measurable relevance band on Khaleeji-dialect queries and Arabic-English code-switching. Conversion follows. The shopper switches to Noon or Amazon.sa.

Failure mode 2

Ramadan calendar

Standard forecasting models miss the Hijri-to-Gregorian shift, the payday-pre-Eid lift, the school-term overlay and the tourism flow. A missed Ramadan buy costs 6–9 % of full-price sell-through.

Failure mode 3

Multi-country buy plan

UAE 25th payday and KSA 1st payday are different curves. ADAFSA-monitored items in Abu Dhabi and baby-formula regulation in KSA are different guardrails. The buy plan needs both.

Use-case grid

Nine production plays across discovery, forecasting, pricing and store ops.

Each ships with merchandiser guardrails, an A/B holdout for incrementality, and the integration patterns the marketer and the merchandiser keep using long after handover.

Arabic-aware search & discovery

AraBERT-v2 + mE5 embeddings on the catalogue. UAE retail-group reference: 27 % NDCG@10 uplift vs Algolia, 9.4 % revenue-per-search.

+27 % NDCG@10

Bilingual personalisation

Recommenders on app and web with Arabic and English embeddings. Marketplace reference: incrementality uplift at p<0.05 over 6 weeks.

Ramadan & Eid forecasting

Temporal fusion transformers with a calendar-aware feature store keyed to Hijri-to-Gregorian shift, payday cycles, school terms and tourism.

−11 pp Ramadan MAPE

Dynamic pricing

Bayesian price-elasticity per SKU-store-day. Brand-house reference: 3.1 % gross-margin uplift across a 1,200-SKU range without volume erosion.

+3.1 % margin

RFM & customer intelligence

CDP-agnostic feature store on Snowflake / Databricks / BigQuery with churn-and-CLV models that handle the long retail tail.

Store-ops CCTV

Queue and shelf-availability detection on Jetson edge devices. No face data; privacy-preserving inference pattern.

Catalogue content generation

Arabic and English titles, descriptions and attributes from supplier data at catalogue scale. Style-tone matched to the retailer brand.

Marketplace buy-box intelligence

Listing-quality and stock-availability signals for Noon, Amazon.sa and regional aggregators. Forecasting hierarchies treat marketplace as a first-class channel.

E-commerce fraud detection

Card-not-present fraud scoring at sub-100 ms on the checkout path. Behavioural and graph features tuned to GCC payment patterns.

Commerce-stack integrations

Platform-agnostic. Catalogue-native. Merchandiser-controlled.

We sit on top of the commerce platform you already run. The merchandiser and the marketer keep working in their existing tools; the model layer is portable across platforms.

Commerce platforms

Salesforce Commerce Cloud · Adobe Commerce / Magento · SAP Commerce · Oracle Retail · VTEX · Shopify Plus

Search & CDP

Algolia · Coveo · Bloomreach · Snowflake · Databricks · BigQuery · in-house CDPs

Forecasting features

Hijri-to-Gregorian calendar · Payday cycles (UAE 25th, KSA 1st) · School terms · Tourism (DET / DCT) · Weather · Promo windows

Store-ops edge

NVIDIA Jetson AGX · DeepStream · privacy-preserving inference · no face data

Marketplace channels treated as first-class

Noon · Amazon.sa · Talabat (q-commerce) · regional aggregators. Forecasting hierarchies include marketplace stocking-location as a dimension.

Regulator & standards

Designed for the consumer-protection and data-protection reviewers.

UAE PDPL. Saudi PDPL. Consumer-protection law and regulator-sensitive category overrides for pharma, baby formula and ADAFSA-monitored items. Bias and fairness review on every pricing model.

UAE & KSA PDPL

Consumer data is processed under explicit consent. Cross-border transfer between UAE and KSA respects both regimes. Marketing exclusion lists, opt-out flows and right-to-erasure paths are wired into the personalisation layer.

Consumer protection

Dynamic pricing models enforce MRP / MAP and never breach regulator-sensitive category overrides (pharma, baby formula, ADAFSA-monitored items). The price-elasticity guardrails are part of the model, not a post-hoc filter.

Store-ops privacy

No face data stored. Inference at the edge on Jetson devices. Compliance with UAE PDPL reviewed with the retailer's legal team before any camera is connected.

Fair pricing & explainability

Every pricing model includes a bias-review test pack and explainability via SHAP. The merchandiser sees why a price moved, not just that it moved. ISO 27001, SOC 2 Type II in place.

Reference engagements

Four anonymised retail outcomes.

UAE retail group · Search

+27 % NDCG@10 vs Algolia, +9.4 % revenue-per-search

Arabic-aware embedding stack on the catalogue. Khaleeji-dialect head validated by native speakers. Benchmark methodology published in the lead magnet.

Regional hypermarket · Forecasting

−11 pp Ramadan SKU-level MAPE across fresh & ambient

Hierarchical model with calendar-aware features. The buy plan adjusted automatically; the merchandiser tuned the override layer through a console.

Marketplace · Recommender

Incrementality uplift at p<0.05 over a 6-week holdout

Two-tower retrieval with bilingual embeddings. Revenue-per-session and units-per-basket measured on the treated cohort against a clean holdout.

Brand house · Pricing

+3.1 % gross margin across 1,200 SKUs, no volume erosion

Bayesian elasticity per SKU-store-day. Merchandiser guardrails on MRP / MAP and regulator-sensitive categories.

Differentiation

Brocode vs the four archetypes on your shortlist.

A platform suite locked to one commerce vendor. A search-only specialist. An offshore SI building a custom recommender. A Big-4 customer-analytics programme. Where each fits — and where Brocode is the right shape.

Capability	Salesforce Einstein / Adobe Sensei	Algolia / Coveo / Bloomreach	Offshore SI (custom recommender)	Big-4 customer analytics
Arabic-aware product discovery (AraBERT-v2 + mE5)			Partial
Ramadan / Eid calendar-aware forecasting
Platform-agnostic (Salesforce / Adobe / SAP / Oracle / VTEX)
Search + recommender + forecasting + pricing in one stack
Incrementality A/B holdout (not CTR)	Partial
Store-ops CCTV on Jetson (no face data)				Partial
Merchandiser guardrails on every model		Partial	Partial
Published Arabic search benchmark
UAE-resident engineering team

Free download

Ramadan & Eid Demand Forecasting Playbook for GCC Retail

A 24-page field guide with a reference calendar — covers the 14 calendar effects that break standard forecasting models in the GCC, a worked SKU-level example across fresh and ambient categories, a hierarchical forecasting blueprint, and a buy-plan adjustment template. Headline figure: the average GCC retailer over-forecasts non-fresh Ramadan demand by 18 % and under-forecasts fresh by 22 %.

Ramadan and Eid effects on demand
Multi-event embeddings
School-year, payday, weather and tourism features
Promo-aware uplift modelling
Inventory optimiser sizing
A/B testing in seasonal windows

Retail AI FAQ

The eight questions every CDO asks.

If yours is not here, raise it in the form below. We answer in writing before the first call.

Every personalisation engagement ships with a controlled A/B holdout, sized for statistical power on the retailer's daily traffic. We measure revenue-per-session, units-per-basket, and gross-margin per shopper for the holdout vs the treated group, and publish a p-value on the primary metric. CTR is a diagnostic; it never appears as a headline number on our scorecard. A retail marketplace reference saw incrementality uplift at p<0.05 over a 6-week window.

Start the conversation

Request a 30-day personalisation incrementality assessment on your store.

Tell us the retail format, the countries, the platform, and the priority use case. We come back within one business day with the assessment shape and the benchmark from comparable engagements.

Prefer WhatsApp? Message our retail lead directly.

Continue exploring