The Ecomm Analyst

Enough attribution vendors have shipped assistant connectors now that it is worth sorting out which ones are real. A connector lets Claude or ChatGPT query your live data directly instead of you exporting it. The catch is that directory sites list a lot of servers vendors did not build, so I checked each of these against the vendor’s own documentation rather than a listing page.

All pricing below comes from vendor pricing pages. Aggregator sites are wrong on this category often enough that I no longer use them, and the spread I saw while checking this post was wide enough to prove the point.

1. Polar Analytics

The most committed of the group. Polar documents official connectors for both Claude and ChatGPT, and the thing that distinguishes it is what sits underneath. Rather than exposing raw API responses, it queries a semantic layer where metric definitions are pre-set, so blended ROAS means one specific thing regardless of who asks or which assistant they ask from. It pulls from a wide set of commerce and ad sources, including marketplaces beyond Shopify.

That governed-definitions approach is the right architecture for this problem. If your team is going to ask questions in plain language, you want the answer to come from one agreed definition rather than the assistant improvising a calculation.

Pricing is not listed on the website. You have to book a demo, which is worth factoring into your evaluation timeline.

2. Triple Whale

Triple Whale documents a connector for Claude and ChatGPT with a sign-in flow that grants read-only access and can be revoked whenever you want. There is also an API key path for tools that cannot do the sign-in handshake, which the documentation correctly tells you to avoid when the better option exists.

Note the tiering. The connector is not part of the free plan. It starts at Foundation, which the pricing page lists at 219 dollars per month, with Automate at 749 dollars per month and Enterprise on request. Prices scale with your annual GMV on top of the tier you pick, plans are twelve-month subscriptions, and paying annually gets you two months free. GMV-based pricing means your cost rises as you grow even if your usage does not, which is worth modeling before you commit.

3. ThoughtMetric

ThoughtMetric, which sponsors this blog, publishes a connector that exposes its metric and dimension catalog to the assistant, so a question like new-customer ROAS by campaign resolves against defined metrics rather than open table access. The catalog covers the usual ad and order metrics plus separate new and returning customer cuts, and dimensions down to individual ad, SKU, and discount code.

Pricing is 99 dollars per month, billed on pageviews rather than revenue, with every feature available at every tier and a 14-day free trial. Annual billing brings it to roughly 83 dollars per month. For brands in the one to twenty million range the pageview basis is usually the friendlier model, particularly if you have high AOV, since GMV-based pricing penalizes exactly that shape of business.

Two things to know going in. It does not cover Amazon, so if a meaningful share of your revenue runs through marketplaces you will need something else alongside it. And like any attribution platform, switching to it means your historical modeled data does not come with you.

4. Northbeam

Worth including precisely because of what I could not find. Northbeam is a serious measurement platform with multi-touch attribution and media mix modeling, and its documentation is thorough. But I could not locate an official assistant connector documented on Northbeam’s own site.

There are community-built servers in the directories pointing at Northbeam’s API. Those may work. They are also somebody’s side project, unsupported by the vendor, and liable to break quietly when the API shifts. If you are on Northbeam and assistant access matters, ask their team what is actually on the roadmap rather than installing something from a listing page.

How I would choose

Do not pick an attribution platform for its connector. That is the tail wagging the dog. Pick it for whether the underlying measurement fits how you actually run acquisition, then treat assistant access as a tiebreaker between finalists. I laid out the broader selection criteria in this roundup of e-commerce attribution tools.

If you do weigh it, the questions that separate these are whether the connector is read-only, whether it authorizes through a sign-in flow or a pasted key, and whether it queries defined metrics or hands the assistant open access. A bounded query surface with governed definitions is the version of this that stays trustworthy after the novelty wears off.

Pricing models matter more than the monthly figure here too. Pageview-based and GMV-based costs diverge sharply as you scale, and which one favors you depends entirely on your AOV. Run both against your own numbers for the next two years before you sign anything annual.