Case Study
In-house WhatsApp for 2 lakh customers
Fisdom · Feb 2025 — Aug 2025 · Full Stack Developer
Replaced vendor Kevit with an in-house messaging platform serving 1–2 lakh active customers, plus the support dashboard and gateway work around it.
1–2L
active customers served
200+
conversations routed per day
+40%
support team efficiency
7+
microservices unified behind Kong
Overview
Support conversations ran through Kevit, a third-party WhatsApp provider: recurring cost, UX we could not change, and customer conversations living in someone else's system.
Internship turned full ownership: I built the replacement messaging platform, the chat-to-agent routing, and the support operations dashboard that made the whole org's queue visible in one place.
Problem
Vendor lock-in on the channel where customers actually talk to you. Every limitation in Kevit's UI became a limitation in our support quality, and every month came with an invoice for it.
Constraints
- —Live customers mid-replacement: no conversation could be dropped during cutover
- —WhatsApp Business API rules: template approvals, session windows, rate limits
- —Fair routing: agents could not be overloaded while others sat idle
Approach — what I considered
Manual claim-based assignment
- + Agents pick what they understand
- − Cherry-picking and idle time
- − No throughput guarantee
Cloud-based auto-routing with manual overrideChosen
- + 200+ daily conversations distributed by load and skill
- + Override keeps human judgment available
- − Needs monitoring to stay fair
Standalone ticketing tool
- + Faster to build
- − Agents live in two tabs again
Tickets embedded in the support dashboardChosen
- + One queue: status, resolution times, conversation metrics together
- − Dashboard scope grew
What I Built
- —WhatsApp messaging platform on the Business API with conversation history and templates
- —Chat-to-agent assignment routing with load-aware distribution
- —Support operations dashboard with integrated ticketing and live queue metrics
- —Kong API Gateway integration across 7+ microservices: centralised auth, rate limiting, routing, observability
Results
- —Kevit retired; recurring vendor cost removed with full data ownership
- —Support efficiency up 40% with one consolidated workflow
- —Platform-wide observability improved once traffic flowed through a single gateway
Retrospective
- 1.Replacing a vendor is a product decision, not just an engineering one — ops had to win too
- 2.Centralising the gateway paid off every week after, in debugging time saved
- 3.Cutover monitoring had to exist before cutover, not after the first complaint