An Amazon sales dashboard usually means one of two things. Either the Sales Dashboard inside Seller Central, which shows one account and stops at revenue, or a Looker Studio report that someone built two years ago and nobody dares to touch. Since 10 November 2026 there is a third way: describe the page to Claude and let Claude Data write the queries.
This guide walks through that, step by step, following the flow Anthropic documents for Claude Data (the product that was called Claude Dashboards until this week).
Connect Seller Central, Vendor Central and Amazon Ads to DataDoe. Each account is authorised once, on Amazon's own consent screen. From then on DataDoe fetches daily at 2:00 in the seller's local time, with some tables updated during the day, and keeps the history. Upload your product costs as well, because without them every profit chart later in this guide is a revenue chart with a different title.
There are two routes, and you only need one.
In Claude, open Customize → Connectors → Add and paste the DataDoe server address. This is the usual way to ask Claude questions about an Amazon account, described on our page about Amazon in Claude. Anthropic says Claude Data works with the apps connected to Claude, and once DataDoe is added it is one of yours.
BigQuery is on Anthropic's list of Claude Data sources. In DataDoe, turn on the BigQuery integration and subscribe to the listing in Google's Analytics Hub. You choose Integrated data (cleaned and joined, the right choice for dashboards) or Raw. The result is a linked, read-only dataset in your own Google Cloud project. There is no monthly fee for it on the Base plan; queries use AI Tokens and Google charges for compute. Then connect BigQuery in Claude with a Google account that can see that project. Our post on Seller, Vendor and Ads data in BigQuery explains the dataset.
Both routes are laid out side by side on the Claude Data for Amazon page.
You can start in two places. In any chat, ask for what you need. Or open the Artifacts tab, choose Data and describe the page. Everything you make is saved under Artifacts, so the dashboard is still there tomorrow.
Anthropic's guidance is to describe the question you want answered, not only the chart, and to name the table or app if you know it. For a first sales page, a prompt along these lines works as a brief:
"Build a sales dashboard for my Amazon business from the DataDoe data. I want to know if we are growing and what explains it. Top row: sales, units, ad spend and profit for the last 30 days, each against the 30 days before. Then sales by week for the last 12 months, split by marketplace. Then the 15 parent ASINs with the largest change in sales against the previous period. Use amazon_profit_by_date for account totals and amazon_profit_by_sku_and_date for anything by product. Keep each currency separate."
Three parts of that prompt do the heavy lifting. It states the question. It names the two tables, so Claude does not have to guess between account-level and product-level data. And it tells Claude what not to do with currencies, which is the most common way a multi-country dashboard goes wrong.
Claude writes one SQL query per chart and runs it against your source. Each chart shows its query and the time its data was last refreshed. Open the query behind the headline figures and read what it counts. You do not need to write SQL to do this. You need to check four things:
If something looks off, tell Claude in the chat, or ask it to explain the number. You can also edit titles and captions directly on the page.
Claude Data does not refresh on a clock. It refreshes when someone opens a dashboard and its data is due, and you can set the cadence per dashboard. A Refresh data button asks for new results on the spot. For Amazon sales, a daily cadence fits the data: yesterday is complete after the early-morning fetch, and refreshing every few minutes buys nothing except usage. Claude Data counts toward your Claude plan's usage limits, and on the DataDoe side queries draw AI Tokens.
A new dashboard is private. Open it and click Share to give colleagues access. By default a shared dashboard loads data through each viewer's own connections, so a colleague without the source sees no numbers.
Live sharing, which is in beta, changes that. In the dashboard's Sources menu you choose "Your access" next to the connector, and people in your organisation then see results from your access, limited to the exact queries the dashboard needs. Those queries cannot change without an editor's approval. Anthropic's own warning applies: anyone you share with sees those results even if they have no access to the source, so look at what the page shows first. People with view-only access cannot edit a dashboard or restore an old version.
If you run an agency: sharing Amazon data outside your own organisation is governed by Amazon's data rules and your client agreement, and no buyer personal data belongs in a dashboard.
Claude Data keeps a version history. You can preview an earlier version and restore it, and the restore applies for everyone. If your team settles on one definition of profit or of TACoS, an editor can certify that query, and every chart built on it then carries a certified badge. Editing a certified query removes the badge, which is the point: nobody changes the definition quietly.
"Where does ad spend pay for itself? By campaign and by search term for the last 30 days: spend, ad sales, ACoS, and profit after ads. Flag terms above $50 of spend with no orders. One section per ad type." Background on the ads data is on our Amazon Ads page.
"What do we keep after everything? Monthly waterfall from sales to profit: Amazon fees, ad spend, cost of goods, refunds. Then the 20 SKUs with the lowest margin that still sell more than 50 units a month."
"What will run out, and what is sitting too long? Days of cover per SKU from current FBA stock and four-week sales velocity, inbound units next to it, and stock older than 180 days."
"How healthy is the vendor business? Purchase orders by week, units ordered against units confirmed, and the ASINs with the widest gap. Add shipped revenue by month against last year."
The wider picture of this week's launch is in Claude Data, Sheets and Whiteboards for Amazon sellers, and the general explanation of certified queries and live sharing is in our Claude Data explainer.
No. Claude writes the queries. It helps to be able to read one well enough to see which dates and which table it uses, and Claude will explain any query if you ask.
As fresh as the later of two things: DataDoe's last fetch, which is daily at 2:00 seller time for most tables, and the dashboard's last refresh, which is shown on each chart.
With live sharing turned on, people in your Claude organisation see the results on your access. Without it, they need their own connection to the source.
No. A dashboard only reads. Changes on Amazon go through DataDoe Actions: Actions stay off until you enable them, most AI clients ask before a change, and every change is logged.
Claude Data is part of paid Claude plans and counts toward their usage limits. DataDoe is $97 a month with a 14-day trial.
DataDoe has no affiliation with Anthropic or Amazon and is not endorsed by either. The Claude Data steps follow Anthropic's Help Center as read on 11 November 2026; certified queries and live sharing are beta features and may change.
.svg)
