The problem.
For businesses selling through WhatsApp, customer conversations drive the sale. I built Timi to connect that experience with customer management and a clearer view of which ads lead to customers and seller-confirmed sales. AI replies support the workflow, but sellers stay in control.
Key decisions.
Focus on WhatsApp.
There was a choice between expanding across platforms and going deeper on WhatsApp. I chose WhatsApp: understanding one seller workflow well mattered more than collecting integrations. Becoming a Meta Tech Provider and requesting additional permissions required patience, repeated submissions, and careful attention to platform requirements.
Get the seller’s tone right.
Getting the AI to sound natural for Kenyan sellers took iteration. English, Swahili, and Sheng needed to feel like a conversation, not a translated script. Tone was only part of the problem: replies also needed to respect the catalog, know their limits, and hand the conversation back to a person.
Draw a boundary around money.
Collecting buyer payments initially felt like a natural extension. It also introduced a different set of operational challenges. I removed buyer checkout and kept payment collection with the seller. Timi shares payment details or the seller’s website; its own subscription billing remains separate.
Adapt the strategy to platform costs.
Changes to Meta’s pricing prompted another look at the strategy, audience, and business model. The focus became product businesses already running ads to WhatsApp. Attribution distinguishes chat origin, seller-confirmed outcomes, and campaign credit rather than treating them as interchangeable.
How it works.
- Railway app and database hosting, with Cloudflare infrastructure and email integration
- WhatsApp Embedded Signup, coexistence, and encrypted integration tokens
- Live conversations with Socket.IO, reply-scope controls, and human handoff
- Background processing with Redis and BullMQ, plus duplicate-reply safeguards
- Subscription lifecycle, usage metering, grace periods, and grandfathered plans
- Conversation evaluation tools, customer management, and catalog-grounded AI replies
The important work was deciding how much the product should do. I narrowed the scope to give sellers a clearer, more dependable workflow. Search foundations and weekly performance reviews are part of the ongoing work; I’ve also started seeing referrals from ChatGPT, an early discovery signal I’m watching.