Start with what customers keep asking
Every shop has a list of questions customer service answers all day: sizes, materials, care, delivery times, returns. An assistant that answers from your own product sheets and policies, not the open internet, handles most of them and hands the rest to a person. The key is that it says "I don't know" instead of guessing, and that you can see every answer it gave.
Custom orders and quotes
If you sell anything made to order, like furniture, prints, or bulk orders, requests arrive as long emails that someone turns into a quote by hand. A form plus an agent that checks the request against your rules (sizes, materials, lead times) and drafts the quote gets that down to a quick review.
Reports that build themselves
Someone usually spends Monday morning pulling numbers from the webshop, marketplaces, stock system, and ads. A small dashboard that pulls all of it automatically and points out what changed is one of the cheapest and most appreciated things I build.
Where computer vision helps a shop
Customers send photos of damaged deliveries and returns. A model that sorts those photos by product and damage type and pre-fills the claim saves time and gives you data to take to your courier. For physical stores and warehouses, cameras can also help with stock checks and packing verification.
What makes it production-ready
- Answers grounded in your own content, with sources, and tested against real past customer questions.
- Clear handover to a person, with the conversation and order already attached.
- Read-only access to orders and stock at first, write actions only with approval.
- Tracking of what it answered, what it escalated, and how customers reacted.
Where to start
Export a month of customer emails, group them by question, and start with the biggest group. Or paste your shop's website into the scanner on the homepage to get ideas tailored to what you sell.