โ† work
Sertis ยท Client: CP All (7-Eleven Thailand)ยท2019 โ€“ 2021

7-Eleven Conversational AI Chatbot

Built the chat commerce experience inside Thailand's 7-Eleven app, serving 16M+ monthly active users across 15,000+ stores nationwide.

roleproduct manager (first PM role)
teamcross-functional ยท Sertis + 7-Eleven
stackGoogle Dialogflow ยท Python ยท RegEx
industryretail / commerce

16M+

monthly active users

15k+

stores nationwide

3

distribution channels

+15%

customer satisfaction

context

7-Eleven Thailand isn't like 7-Eleven anywhere else. With over 15,000 branches blanketing the entire country, it's less of a convenience store chain and more of a national infrastructure. The mobile app had 16 million+ monthly active users, and the company was looking to unlock new digital sales channels as part of a larger digital transformation initiative.

The goal was to build a conversational layer that could serve across 3 distribution channels: answering customer inquiries, driving product sales through chat, and supporting the app's real-time delivery service. All powered by NLP, before the LLM era.

This was my first PM role. I joined as a junior product manager at Sertis (a data and AI consulting firm) and owned this project end-to-end.

๐Ÿช7-Eleven in Thailand is amazing. We have over 15,000 branches. Seriously, look it up!

what we built

A conversational AI chatbot embedded in the 7-Eleven mobile app, powered by NLP using Google Dialogflow for intent recognition and RegEx-based pattern matching for structured flows. The chatbot served two distinct roles:

customer support agent

Handled general inquiries like store locations, operating hours, promotions, and order status for 7-Delivery (the app's real-time delivery feature).

chat commerce assistant

The more ambitious part. Users could browse products, see real-time stock at nearby stores, get personalized recommendations, and complete purchases, all without leaving the chat.

chat commerce flow

1

Detect user location

Map the user to their closest 7-Eleven stores automatically.

2

Check real-time stock

Query live inventory across 15,000+ stores to show what's actually available.

3

Surface products in chat

Display product carousels with promotions and discounts applied automatically.

4

Personalized recommendations

Suggest products based on purchase history and user profile, making each conversation feel relevant.

5

Complete purchase in chat

The entire shopping flow, from discovery to checkout, happened inside the conversation. No app switching needed.

what i actually did

requirements & PRD

Gathered requirements from the 7-Eleven business team and translated them into detailed product requirements documents, defining user flows, edge cases, fallback behaviors, and success metrics.

conversational design

Designed the conversational flows, mapping out intents, utterances, entities, and response trees. This is where the real product thinking lived: how should the bot respond when it doesn't understand? When should it escalate to a human? How do you make a carousel of snacks feel natural inside a chat thread?

engineering coordination

Ran the engineering team at Sertis while integrating closely with 7-Eleven's internal dev team. Two separate engineering orgs building one product, so alignment was constant work.

testing & iteration

Ran testing cycles and continuously gathered feedback from users to refine conversational flows, improve intent recognition accuracy, and reduce fallback rates.

what made this interesting

scale

16 million monthly active users across a nationwide app. Even small improvements in conversion or deflection rates translated to significant business impact.

chat commerce was still new

This was before the ChatGPT era. Building a commerce experience inside a chat interface required creative conversational design without the benefit of large language models. Every flow was hand-crafted.

real-time inventory integration

The chatbot checked live stock across 15,000+ stores, matched products to the user's nearest location, and applied dynamic promotions. The technical integration was non-trivial.

personalization at scale

Product recommendations were tailored to individual users based on purchase history and profile data, making each conversation feel relevant rather than generic.

why this matters

This was my entry point into AI product management, and it shaped how I think about building AI products to this day. Working with NLP before the LLM era taught me that good conversational design isn't about the model, it's about understanding what users actually need at each moment in a conversation.

The patterns I learned here, intent mapping, fallback design, human escalation logic, conversational UX, carried directly into the RAG Copilot I built years later.

chat commerce before the LLM era.

every flow hand-crafted, every intent mapped, serving 16M+ users across 15,000+ stores in Thailand's largest retail app.