A bilingual sales OS for an insurance domain that doesn't fit a CRM.
A health-insurance brokerage running corporate accounts through WhatsApp needed structure without leaving the channel. We modeled the domain natively — leads, quotes, approvals, family-member registration, age-band pricing, and commissions — and shipped a mobile-first platform that lives where the conversations already happen.
01The Challenge
Health-insurance brokerages serving Saudi SMEs run a sales motion no off-the-shelf CRM understands. A single corporate quote depends on per-employee age bands, family-member registration, multi-step approvals before a policy can issue, and commission tracking that compounds on every renewal. Brokers were holding all of it inside WhatsApp threads, Excel files, and unattended voice notes — because their corporate clients refuse to leave WhatsApp. Every CRM they tried forced them off the channel where the deals actually close.
02Our Approach
- Modeled the insurance domain natively in the schema — Lead → Opportunity → Quote → Approval → Closed Deal → member registration with family links and age-band-aware product types — instead of forcing a generic deal pipeline to fit.
- Built the WhatsApp bridge at the WA-Web protocol layer so brokers send and receive on their personal numbers, with no Meta Cloud API contracts required and no migration cost for the client.
- Layered AI on top of inbound conversations: voice-note transcription, intent classification, draft replies in the broker's voice, and a 0–100 lead-interest score that learns from feedback.
- Shipped a mobile-first stack — NestJS + Prisma API, Next.js 15 dashboard, and a native Android APK with 40+ iterative releases — bilingual at the schema layer, SAR-precise at the money layer, Asia/Riyadh at the time layer.
Built with
- NestJS
- Prisma
- Postgres
- Redis
- Next.js 15
- React 19
- Baileys (WA-Web)
- Android (Compose)
- Meilisearch
- Docker
What they got
- Insurance-domain schema (40+ models)
- WhatsApp bridge on agent's personal number
- AI: voice transcription, intent, draft replies, lead score
- Web dashboard + native Android app (40+ releases)
- Bilingual AR/EN at the schema layer
Timeline
- 01Domain modeling4 weeks
- 02API + WhatsApp bridge8 weeks
- 03Web + mobile app12 weeks
- 04AI layer + iterationongoing
Mobile-first surface for field reps. Customer data redacted out of respect for client confidentiality.
It feels like the system was built around how we already work — not the other way around.
— Sales OS lead — name withheld
03The Outcome
Brokers stay in WhatsApp; the platform turns those conversations into structured leads, opportunities, and approvals automatically. Managers see real pipeline numbers — including projected commissions on renewals — instead of screenshot piles. The WhatsApp bridge is isolated as a single-purpose service, keeping the core API portable to Meta Cloud API later if call-volume justifies it.
