Retail and D2C · AI
WhatsApp order tracking and support agent in Gujarati, Hindi and English for a D2C brand
A WhatsApp agent that answers where is my order, handles returns and exchanges, and replies in the language the customer writes in: Gujarati, Hindi or English. It reads live order and courier data and hands the conversation to a person, with full context, when it should.
- Client
- A D2C ethnic wear brand in Surat selling through its own website and marketplaces
- Region
- India
- Sponsor
- Founder and the customer experience lead
- Duration
- 8 weeks to first release, then a monthly tuning plan
Engagement model
O2E: Output Oriented Engagement
Scope was agreed as a set of conversations the agent must handle end to end, tested against real past chats, with acceptance before payment.
How O2E worksThe team
- Delivery lead1
- AI engineer2
- Integration engineer1
- Shared QA1
The challenge
Where things stood
- Most customer messages came on WhatsApp, and the support team answered them by hand from a shared phone.
- The same questions about order status, delivery dates and refunds came in all day, often in Gujarati or Hindi typed in English letters.
- Agents switched between Shopify, the courier portal and Razorpay to answer one question.
- Return and exchange requests were lost in long chat threads, and customers had to repeat themselves.
The approach
How we ran it
Weeks 1 to 2
Read the chats
Went through several months of exported WhatsApp chats with the support team to group the questions, note how customers mix languages, and agree which cases need a person.
Weeks 2 to 4
Connect the data
Connected Shopify orders, courier tracking and Razorpay refunds so the agent can look up a customer by phone number and see the real status of each order.
Weeks 4 to 7
Build and test the agent
Built the agent with approved message templates, language detection and clear handover rules, then replayed past conversations to check every reply before customers saw one.
Week 8
Go live
Launched on the brand's WhatsApp Business number with the support team watching every conversation for the first fortnight, then moved to daily review of handovers only.
What it looks like
Representative screens
Drawn to show the kind of system we delivered. Client screens and data stay private.
WhatsApp, +91 98250 XXXXX
Maro order kyare aavse? Order number 10482
Namaste Hetalben. Tamaro order #10482 (2 kurtis) Delhivery thi nikli gayo che ane kale, 14 October sudhi Vadodara pahonchi jase.
Ek kurti nu size nanu che, exchange thai shake?
Haa, delivery pachhi 7 divas ma exchange thai shake. Kai size joie che? Hu pickup book kari dau.
Exchange request EX-2291 created, pickup pending size confirmation
Support inbox
Agent handling2
- #10482 Hetal Shah, exchange size L
- #10517 Rohit Verma, delivery date
Handed to team2
- #10391 Priya Nair, damaged parcel photo
- #10455 Anil Joshi, COD refund to bank
Waiting on courier1
- #10402 Kavya Rao, RTO marked in error
Resolved today2
- #10377 Sunita Patel, refund sent
- #10468 Imran Shaikh, address updated
What we delivered
Modules and features
- WhatsApp Business agent that replies in Gujarati, Hindi or English, including romanised text
- Order lookup by phone number with live courier status and expected delivery date
- Return and exchange requests with photo upload and pickup booking
- Refund status from Razorpay with the reference number
- Handover to a support agent with the full chat and order history attached
- Support inbox where the team sees open conversations, tags and reasons for handover
- Weekly report of top questions and failed answers for tuning
What changed
Outcomes
- Where is my order questions answered on WhatsApp at any hour, without waiting for the team
- Customers get replies in the language they wrote in
- Support staff spend their time on complaints and exchanges
- Every return request is logged with photos and a pickup date
Solutions
Industry and platforms
Looking back
Customers rarely write in clean Gujarati or Hindi. Most type it in English letters with their own spellings, and some switch language mid sentence, so testing on textbook phrases told us very little. Replaying the brand's real chats was the most useful test we ran. We also set the handover rules too tight at first, which meant the team saw many conversations the agent could have finished; loosening them after the first fortnight, with the team's agreement, was the right call. Next time we would agree the refund wording with the founder before build, as that took several rounds.
Start a conversation
Discuss a similar project
Tell us where your team is today and we will walk you through how a project like this would run for you under O2E. The first conversation is free. Delivery work is paid only after you accept it.
A specialist replies within one working day. The first conversation is free.