The Ecomm Analyst

Connecting ThoughtMetric to Claude takes about five minutes and the work is all on the Claude side. You add ThoughtMetric as a connector, authorize it against your account, and the assistant can then query your attribution data directly. ThoughtMetric sponsors this blog. What follows is the setup I run, plus the parts that trip people up.

Before you start

You need an active ThoughtMetric account with data flowing. That sounds obvious, but it matters more than usual here, because a connector pointed at an account that has been collecting for four days will answer your questions with four days of data and will not necessarily make a fuss about it. Give it a few weeks of history before you start asking comparative questions.

You also need a paid Claude plan, since connectors are not available on the free tier. On Team or Enterprise plans an owner may need to approve the connector before individual users can enable it, which is worth knowing before you promise a demo.

The setup

Connectors live in Claude’s settings under Connectors, and can also be reached from the tools menu inside a conversation. You add ThoughtMetric there and authorize the connection against your ThoughtMetric login. The authorization is standard OAuth, so you are approving access rather than pasting an API key anywhere, and you can revoke it later from the same settings page.

ThoughtMetric’s help center has the current version of the exact steps if the interface has moved since. I would rather point you there than describe screens that change every few months.

Once connected, enable it in whichever conversation you want to use it in. This is the step people miss. The connector being added to your account does not automatically mean it is active in the chat you happen to have open, and a fresh conversation is usually the fastest way to confirm it is working.

Confirming it actually works

Do not open with an interesting question. Open with a boring one you already know the answer to.

Ask which projects the connection can see. That confirms authentication and, if you run multiple stores, tells you what the assistant is going to default to. Then ask for total orders and total sales for a specific closed month, and check both against your commerce platform. If those match, the pipe is clean. If they do not, you have a data question rather than a connector question, and you want to find that out now rather than three weeks into using it for decisions.

My third check is asking what metrics and dimensions are available. It is a useful inventory to have seen once, and it saves you from later asking for something that does not exist and puzzling over the answer.

Things that go wrong

Multiple stores is the big one. If your login has access to several projects and you ask about ROAS without saying which brand, you will get an answer about one of them, and it may not be the one you meant. Name the store in the question. If you work across clients, name it every single time.

Vague date ranges are the second. Last month, this quarter, and recently all resolve to something, but not always what you had in mind, and month boundaries in particular are a common source of quiet disagreement with a dashboard. Give explicit dates when the number matters.

The third is asking for combinations the data model does not support. Not every metric can be grouped by every dimension. Ad spend cannot be broken out by discount code, for instance, because spend has no discount code attached to it. The connector checks compatibility before querying, so you generally get told rather than getting a wrong answer, but if a question comes back reworded into something narrower than you asked, that is usually why.

What to do with it once it works

The honest answer is quick lookups. Checking a number from your phone. Answering a question in a meeting without opening a laptop. Slicing something an odd way that would not justify building a saved report.

Because it is read-only, the risk surface is small. Nothing you ask changes a budget or writes back to an ad platform. The realistic failure mode is not damage, it is acting on a number you misread, which is why the reconciliation check above is worth doing properly.

One habit I would push. When a query produces a number that surprises you, verify it in the dashboard before you repeat it to anyone. Not because the connector is unreliable in my experience, but because natural language makes it easy to ask a slightly different question than the one you think you asked, and a confident sentence reads more authoritatively than a chart does.

For reference, the connector is included with any ThoughtMetric subscription rather than sold separately. Pricing runs on monthly pageviews from $99 per month for 50,000 pageviews with every feature at every tier, and there is a two-week trial with no card required, so you can test the whole thing before committing.

If you are setting up tracking from scratch rather than adding to an existing stack, the order I do things in is in the analytics setup I run on every new Shopify store.