A while back I wrote about the safe way to give an AI access to your store: read-only, scoped, audited, with the model talking to tools you control rather than the database directly. That post was really describing a thing I'd been building. This is that thing. It's called Ask AI, it's on the Shopify App Store, and it lets you ask questions about your store in plain English.
The problem it solves
Your data is real and useful, and it's scattered across a dozen tools that don't talk to each other. Orders in Shopify. Email in Klaviyo. Spend in Meta and Google. Tickets in Gorgias. The numbers that matter live in the gaps between those tools (true ROAS once you net off refunds, which products are about to stock out given current velocity), and getting at them usually means exporting four spreadsheets and praying.
Meanwhile, the AI assistant you already use is brilliant at reasoning and knows nothing about your business.
Ask AI closes that gap. It plugs your store's data into the assistant so the assistant can actually answer.
What it looks like in practice
You ask the kind of question you'd ask an analyst:
- "What were my top 5 products by revenue this month?"
- "Compare email versus paid social performance for the last 30 days."
- "Which products will stock out in 14 days based on current velocity?"
- "What's my true ROAS across Meta and Google this month? Include refunds."
That last one is the tell. It spans Shopify and two ad platforms, and it nets off refunds. That's the sort of cross-source question that's a genuine chore by hand and instant once the data's connected.
How it actually works
The interesting design decision is that Ask AI doesn't try to be a chatbot. It's a bridge. You keep using the assistant you already like, and Ask AI feeds it your data through whatever integration that assistant supports:
- Claude: a Model Context Protocol (MCP) server.
- ChatGPT: a Custom GPT wired to the same backend.
- Gemini: function calling for live lookups.
- Perplexity: an MCP server connection.
Same data, same tools underneath, exposed in the shape each assistant expects. Setup is API keys or OAuth and takes minutes, not an afternoon.
The connectors
There are around twenty data sources today, and the point is breadth, so a single question can reach across your whole stack:
- Commerce: Shopify, WooCommerce
- Email & SMS: Klaviyo
- Analytics & SEO: Google Analytics 4, Google Search Console, Microsoft Clarity
- Advertising: Google Ads, Meta Ads, TikTok Ads
- Attribution: Triple Whale
- Support: Gorgias
- Subscriptions: Recharge
- Reviews & loyalty: Yotpo, Judge.me, Reviews.io, Smile.io
- Finance & ops: Xero, ShipStation
- Other: YouTube, Swish wishlists
Built on the read-only principle
Everything from that earlier post is baked in, because I wasn't going to ship something I'd warned people against:
- Read-only by default. It can read your data; it cannot change orders, edit products, or issue refunds.
- Encrypted. Credentials and data are AES-256-GCM encrypted at rest.
- Scoped, defined tools. The assistant only ever calls the specific functions I've exposed. It can't free-roam your account.
The model gets to answer questions, not take actions. That boundary is the whole point.
It also works while you're not looking
Beyond ad-hoc questions, Ask AI keeps a self-updating executive dashboard and sends a monthly recap email on the 2nd: headline metrics, sparklines, and a written summary of what moved and what to look at. The same engine that answers your questions quietly surfaces the ones you didn't think to ask.
Try it
There's a 7-day free trial, no credit card, and guided onboarding so you're not left wiring up integrations alone. Paid plans start at £19.99/month, scaling by how many data sources and seats you need.
You can start a trial directly or read the overview on this site. And if you've got a store and a "can it answer this?" question, send it to me. Those questions are how the connector list keeps growing.
Written by Charles Marsh, who runs EC Web. Get in touch or book a chat.