AI customer assistant
The AI customer assistant is a conversational view over your buyer data. It’s not a chatbot for your customers (that’s Hilal Chatbot, a separate product) — it’s a tool for you to interrogate the buyer dataset without writing SQL.
In this guide:
- Open the AI customer assistant
- Ask a question
- Pin useful answers
- Understand the data scope
Step 1: Open the AI customer assistant
Go to Customer Hub → AI assistant in the sidebar (or open it in a side drawer from anywhere via the ? menu → Ask AI about customers).
The assistant offers example prompts on first open:
- “How many repeat buyers do I have in DE?”
- “What’s my refund rate by SKU last quarter?”
- “Show me buyers who’ve spent more than $500 lifetime.”
- “Which marketplace has my highest review rating?”
Step 2: Ask a question
Type a question. The assistant:
- Translates your question to a structured query (the same pipeline that powers NL2SQL, but scoped to buyer-related tables only).
- Executes the query.
- Renders the answer as a number, table, or chart depending on shape.
- Shows the underlying SQL on demand for transparency.
Step 3: Pin useful answers
If a question is one you’ll ask repeatedly (e.g., “weekly refund rate by marketplace”), click Pin to dashboard on the answer. The pinned card appears on your home dashboard’s deeper-cards row and refreshes on the same date filter as the rest of the dashboard.
Step 4: Data scope
The assistant queries only buyer-related data:
- Orders, order line items.
- Refunds and returns.
- Reviews.
- Buyer messages.
- Marketplace and SKU dimensions.
It doesn’t see ad data, finance data, or vendor data — those belong to other AI tools (AI Chat for general analysis, NL2SQL for arbitrary queries).
Troubleshooting
- “I can’t answer that.” Either the question is outside the buyer-data scope, or the assistant didn’t understand. Try rephrasing more concretely (“how many” / “what percentage” / “show me a list”).
- Answer is wrong. Click Show SQL to see what the assistant ran. If the SQL is wrong, rephrase. If it’s correct but the result is unexpected, the data itself may have a quirk.
- Pinned card doesn’t refresh. Pinned cards refresh on each dashboard load. If a card is stale, hit refresh in the corner of the card.