How to deploy a WhatsApp AI agent in Morocco in 14 days

Why WhatsApp is Morocco's #1 channel

Morocco has 22M+ active WhatsApp users — by far the dominant messaging channel, ahead of Messenger and Instagram DM. For customer service, real estate, e-commerce or healthcare, it's the natural channel for leads and retention.

A well-designed WhatsApp AI agent absorbs 60–80% of first-line conversations (hours, availability, booking, order tracking, product FAQ) and cleanly hands off to a human when the conversation leaves scope.

Pre-requisites before you start

A verified Meta Business account (Moroccan entity KYC — RC, ICE, IF), a dedicated phone number not used on personal WhatsApp, an official Business Solution Provider (BSP) — 360dialog, Twilio or Gupshup are the most used in Morocco — and a structured knowledge base (FAQ, catalog, internal procedures) that we'll turn into a vector index.

On compliance: CNDP standard declaration, clear in-conversation notice that the user is talking to an automated agent, and log hosting on EU or Moroccan infrastructure.

The 7 deployment steps

1. Step 1

   Days 1–2 — Scoping & CNDP filing

   Scoping workshop: scope of managed conversations, handoff scenarios, default language (written darija/French), brand tone. In parallel, prep the CNDP file (standard declaration, processing register).

2. Step 2

   Days 3–5 — WhatsApp Business API provisioning

   Verify Meta Business account, link the number to the BSP, configure message templates (HSM) approved by Meta for outbound notifications. Sandbox send/receive test.

3. Step 3

   Days 6–8 — Knowledge base indexing (RAG)

   Convert documents (PDF, web pages, Notion, Sheets) into vector chunks stored in pgvector or Qdrant hosted in the EU. Set up the retrieval + generation pipeline with a multilingual model (GPT-4o-mini or Claude Haiku) in no-retention mode.

4. Step 4

   Days 9–10 — CRM/ERP/PMS integrations

   Two-way connection to business tools: lead creation in HubSpot, stock lookup in Sage X3, booking in Cal.com, order verification in WooCommerce/Shopify. Each action exposed as a tool (function calling) to the agent.

5. Step 5

   Days 11–12 — Darija/French testing & guardrails

   Bench of 60–100 test conversations covering written darija, French, code-switching, and adversarial cases (insults, out-of-scope, prompt-injection attempts). Calibrate the confidence threshold that triggers human handoff.

6. Step 6

   Day 13 — Human handoff and agent console

   Set up the console (Chatwoot, Front, or the BSP UI) where reps take over in one click. Real-time notifications, statuses (bot/human/resolved), quality metrics.

7. Step 7

   Day 14 — Production rollout and monitoring

   Progressive cutover: 10% of traffic in the morning, 50% at noon, 100% in the evening if KPIs are green. Supervision dashboard: deflection rate, CSAT, average response time, cost per conversation in MAD.

Questions fréquentes