Claude Data Is the Reporting Tool I Wanted at Twitter

Claude Data Is the Reporting Tool I Wanted at Twitter

Claude Data offers a new way to build and share dashboards from warehouse data. Dennis compares it with Notion Workers and Dashboards, drawing on revenue reporting at Twitter and his own website metrics dashboard to explore where each approach fits.

Nov 16, 2026
In May I wrote about why Notion Workers and Dashboards had become my default BI stack for teams without a data warehouse. At Twitter, I built dozens of Tableau dashboards for revenue reporting, but no senior execs opened them directly. Ever. Instead, every week, we ended up putting screenshots of the charts into slides. Syncing data into Notion and building your dashboards there gives organisations a way to put reporting alongside the context and other work they look at every day.
Last week, Anthropic made Claude Data generally available on its paid plans. It offers another way to put reporting somewhere people already spend time, with three features that would have saved my team at Twitter a lot of grief.

What revenue reporting looked like at Twitter

At Twitter, building the charts was the easy part. Most of the work came from three other problems.

Getting the data

Revenue reporting meant complex SQL running against our data warehouse, combining Salesforce with billing and ad auction data. Different teams owned different tables and sources, so nobody had a shared picture of the overall data model.

Getting people access

Every viewer needed a Tableau licence, the cost of which adds up fast, and anything that refreshed live also needed permissions on the underlying data. So access was rationed, and the people who most needed a number were often the ones who couldn't open the dashboard.

Agreeing on the numbers

We had to agree how to calculate account churn, including which lookback period to use; how to define product cross-sell; and how to measure ad product performance in the auction. Without shared definitions, different teams could report different answers to the same question, and we'd spend time reconciling the numbers before anyone could use them to make decisions.

Ordinary Folk: the right stack, with the usual overhead

Ordinary Folk is a business that warranted a proper warehouse: several revenue streams, three markets, and data spread across Stripe, ad platforms, marketplaces and internal databases. We set up Fivetran pipelines into BigQuery, built more than 40 intermediate staging tables, and put Tableau Cloud and Pulse on top. It was the right call for that level of complexity, and I'd recommend the same stack to a similar company today.
For a company of Ordinary Folk's size, the cost and maintenance burden are harder to absorb than they are for a large public company. At US$75 per Creator and US$15 per Viewer per month, someone has to decide who gets a Tableau seat. The staging tables need someone who writes SQL to maintain them. And the logic behind each metric lives inside individual workbooks, where only the person who built it can easily check it.

What Claude Data adds

Claude Data turns a question about your company's data into a dashboard, a scrolling data story, a narrated video or a written analysis. Claude writes a SQL query for each chart, and every chart shows its query and when it was last refreshed. Three features make it more than a nicer chart builder.

Direct connections to your warehouse

Notion Workers copy data into Notion databases on a schedule. Claude Data queries the sources directly through warehouse connectors for Redshift, BigQuery, Databricks and Snowflake, plus apps like Salesforce and Notion. Joins and windowed aggregations happen in SQL where they belong.

Viewers can refresh using the editor's access

By default, a shared dashboard loads its data through each viewer's own connections, which is the safe setting. With live sharing turned on, viewers see results from the editor's connections instead, limited to the exact queries the dashboard uses, and those queries can only be changed by the editor. So your sales team can refresh the pipeline dashboard without having to be granted access to the data warehouse itself.

Certified queries

You can certify a query as the approved way to calculate a number. Charts built on it carry a badge, and editing the query removes the certification, so nobody can tweak the logic and keep the badge. Claude also finds and reuses certified queries when it builds new dashboards, instead of writing a different version of the same metric. Teams that already keep metric definitions in dbt, Looker or Cube can bring them in through those tools' connectors.
That doesn't make it a full semantic layer. But it does give people (and Claude) one agreed answer to "how do we count organic sessions?", which is more than most companies have today.

Our website metrics dashboard

To try it out, I built a dashboard for our website. Our Google Analytics and Search Console data already export to BigQuery, so I pointed Claude at that project, described what I wanted to see, and refined it over a few conversations.
It shows headline figures for the last 28 days against the 28 before, weekly site visits by source, Google impressions, clicks and average position by week, how our service pages perform in search, our top search queries, and readers and traffic sources for each blog post. The blog section also pulls each post's tags from our Blog Content database in Notion, so one dashboard reads from both a warehouse and Notion with no sync in between.
Four of those queries are certified: weekly visits by source, Google search by week, the 28-day comparison and blog post tags. The next time I ask Claude, in any conversation, how many visits we had from search last month, it reuses the same query the dashboard does.

A few drawbacks

  • No control over who can certify. Anyone with edit access to a dashboard can certify a query. The badge is supposed to tell people a number follows the approved definition, but without role-based permissions to limit certification to the data team or a metric's owner, it only tells you that someone with edit rights signed off.
  • You still need the data somewhere queryable. Claude Data doesn't move data. If your numbers are spread across a dozen SaaS tools with no warehouse, you still need Fivetran, a native export or a Notion sync worker to bring them together first.
  • Lock-in to Claude. The dashboards live in Claude, so moving to another platform could mean rebuilding them. I'm comfortable with that trade-off when the product and its ecosystem work this well, and all of your certified queries can be copied and moved elsewhere if you need to.
  • Warehouse query costs. Dashboards only re-query when someone opens them and the data is due for a refresh, which keeps repeat queries down. A popular dashboard on large tables can still add to warehouse costs, so it’s worth watching the bill for the first few weeks.
  • The best features are still in beta. Live sharing and certified queries are both labelled beta. They're on by default for Team plans, but on Enterprise plans they stay off until an owner turns them on.

Which one I'd recommend

If your team lives in Notion, pulls its numbers from SaaS APIs, and has no data warehouse, Notion Workers and Dashboards would still be my default recommendation. They cost very little, and the charts live right alongside your context and work. (Workers and Dashboards are a standard part of our Notion consulting work.)
If you already have a data warehouse, Claude Data is a strong fit when multiple team members want to build their own reporting and need to work from shared definitions of complex metrics. And the two aren't exclusive: our own website dashboard reads from Notion and BigQuery side by side.
If you'd like help working out which setup suits your team, book a discovery call and we can talk it through.