PART 01 · PHASE 1 · REVIEWED 31 AUG 2026

Validate the GCC context.

Six research dimensions that may matter across Gulf commerce. None is a universal benchmark. Measure them by store, country, category, channel, customer cohort and season before changing a workflow.

Evidence worksheetFoundation

The earlier regional percentages were retired. Broad averages without a dated primary source, sample, geography, category, denominator and measurement method cannot calibrate your store. This module now tells you what evidence to collect and how to avoid causal overreach.

Start with a data contract

Before comparing markets or channels, define an order, paid order, delivered order, cancellation, return, refund, contribution margin, customer, session and attributed conversion. Record timezone, currency, taxes, shipping, payment fees, discount treatment, lookback window and data freshness. Reconcile daily totals to the commerce and finance systems.

1
Scope — store, market, category, channel, cohort and date range
2
Definitions — event semantics, exclusions, lag and currency treatment
3
Quality — missingness, duplicates, late events, consent and reconciliation
4
Decision — owner, baseline, primary metric, guardrails and stop rule

1. Payment and fulfillment

Do not assume a COD share or cancellation multiplier. Measure payment method selected, authorized, fulfilled, delivered, refused, returned and refunded. Segment by market, carrier, product, first/repeat customer and acquisition source. Use completed delivery—not checkout alone—when the decision depends on realized revenue.

“For orders created in the last complete cohort, what share reached each fulfillment state by payment method, and what is the confidence interval after the normal delivery and return lag?”

2. Order value and contribution economics

A regional AOV comparison is not a budget rule. Calculate median and distribution as well as average; include discounts, tax, shipping, payment, fulfillment, returns, refunds, cost of goods and support. Use contribution margin and cash timing as guardrails. A higher basket can also mean greater inventory, fraud, service and return exposure.

3. Active and converting periods

Build a local-time heatmap from your own sessions, purchases and delivered orders. Separate traffic volume from conversion probability and incrementality. Correct for campaign schedules, budget constraints, day of week, payday, promotion and season. Test time-based allocation with a pre-registered comparison; do not hard-code an evening percentage from this lesson.

4. Arabic, English and code-switching

Language preference varies by market, audience, category and placement. Draft matched variants with equivalent offers and rights-cleared assets. Native speakers review meaning, dialect, cultural fit, claims, readability and RTL behavior. Randomize where practical, monitor quality and contribution economics, and report uncertainty instead of declaring a universal uplift.

Do not infer

An Arabic label alone caused an outcome, one dialect fits every Gulf audience, or automated translation is acceptable production copy.

Evaluate

Comprehension, brand fit, accessibility, policy/claim accuracy, primary outcome and guardrails on a representative audience.

5. WhatsApp as an approved service channel

Measure how many eligible customers choose the channel, the questions they ask, response time, correct-resolution rate, escalation, complaints and opt-outs. Verify current WhatsApp Business rules, approved templates and customer consent. Minimize personal data; never use a model to decide refunds, credit, medical/safety issues or disputed payments.

6. Ramadan and other seasonal periods

Use comparable historical cohorts and operational evidence—not a fixed annual-revenue claim. Document calendar dates, promotions, inventory, delivery capacity, staffing, creative, media and external changes. Plan scenarios with uncertainty, freeze risky changes near peaks, rehearse incident response, and keep a rollback path.

Decision rule A pattern becomes a planning input only after its definition, source, sample, lag, uncertainty and operational explanation survive review. It still remains a hypothesis for the next period.

What comes next

Part 2 should be read as a proposal for separating responsibilities—not a requirement to deploy seven agents. Choose one low-risk, reversible decision with sufficient evidence. Keep the first pilot read-only or proposal-only.

1
Payment — which state represents realized value after normal lag?
2
Economics — what is contribution after all included costs?
3
Time — when does your own audience convert, and why might that be confounded?
4
Language — which reviewed variant helps which audience and task?
5
Channel — who consented, what can be resolved, and when must a human take over?
6
Season — what evidence, capacity, uncertainty and contingency shape the plan?
← Back to hubPart 2 — Proposed service boundaries →