AI Customer Operations Platform

01 / Business Challenge

Indonesian small businesses run their sales on WhatsApp, but owners lose hours every day answering the same questions — price, stock, delivery, payment. Slow replies mean lost sales, and most AI chatbot products are too expensive or too generic for a low-margin market.

The business problem was never "build a bot." It was: how does a small business stay responsive and capture every lead without hiring more people?

02 / Solution

A WhatsApp-first customer operations platform that automatically handles repetitive sales conversations, sends catalog and product information, captures leads, and escalates complex chats to a human.

It was positioned around a hard commercial constraint — automation a UMKM can actually afford — rather than around the underlying AI.

03 / Execution

I owned product strategy and build: defined the UMKM sales friction, mapped real Indonesian buyer chat behavior across nine intent categories, and designed a deterministic-first pipeline that uses AI only as fallback.

Key execution decisions:

• No per-message LLM cost on entry tier — deterministic routing keeps pricing at Rp50–59k/month, aligned to willingness to pay.
• Human handoff for sensitive flows — payment, refund, and complaint always escalate, because wrong automation destroys trust.
• Unknown-message inbox as learning loop — unmatched chats feed back into the system instead of being dropped.

04 / Business Impact

The platform made instant, consistent customer response financially realistic for small businesses. It reduced response time, captured leads that would otherwise be lost, and proved a repeatable model for AI-enabled operations in price-sensitive markets.