2026 Taager
Taager Merchant AI Assistant
Contextual help for sellers while they work inside the platform, from drafting a customer reply to explaining a payout.
Merchants asked support the same questions every day. An assistant that understood the merchant’s own orders could answer most of them in place. The design problem was trust: when should it act, when should it suggest, and how does it show its work?
01
Context and problem
Support handled a high volume of repetitive questions that the platform already had answers to. Merchants wanted answers inside the flow they were in, not a chat window they had to leave their work for.
02
Approach and key decisions
Three rules shaped the assistant.
Suggest, then confirm
The assistant drafts a reply or an action; the merchant always sends it.
Show the evidence
Every answer cites the orders, payouts or policies it used.
Live where the question is
Entry points sit on the order, the payout and the product, not in a global chat.
03
Prototype and test
I prototyped the interaction in TypeScript against real merchant data in a sandbox, then tested with eight merchants. Confidence rose when the sources were visible and fell when the assistant acted on its own.
04
Outcome
Illustrative figures. Replace with measured outcomes before publishing.
of support questions answered in place
faster customer replies
merchant test sessions
It writes the reply, but I am still the one who sends it. That is the right balance.
Test participant · sample
What I’d carry forward
- Visible sources are the whole trust story.
- Entry points matter more than the model.
- Prototype against real data or you learn nothing.
→ Next project
Damenkom
Payment and the experience after the purchase: a clearer checkout, a calmer post-purchase, fewer “where is my order” messages.
