Anthropic published a writeup on how they use Claude for self-service data analytics. I'm a fan of their blog and find a lot of it valuable, but this one hit especially close to home as I've been working on a lot of similiar issues recently.
I also found this interesting because it didn't read as a gloating post about Claude being able to one shot everything.
The interesting bit was more around how much has to sit around the model (definitions, curated context, evals, provenance, routing, correction loops).
Self-service still has a service
A lot of people/companies say they want self-service analytics.
Usually this means fewer Slack drive-bys for their data team, fewer one-off dashboard/notebook requests, fewer "can you pull this for me real quick" messages that are never actually quick.
All of that is fair (nobody wants the data team to be a human query endpoint), but self-service does not remove the work.
Instead of asking an analyst to write the query, users depend on the definitions, joins, examples, permissions, and caveats the data team already encoded. If those are thin, stale, or contradictory, the chat interface just makes the confusion faster.
It's on the data team to ensure that doesn't happen, and it's on the data infra/platform team to make that as easy as possible.
The model should not guess the business
The obvious failure mode for AI analytics is wrong SQL. The worse failure mode is right-looking SQL built on wrong business logic.
What does "total spend" mean? Which revenue number counts? Are trials included? Is churn based on invoice status, account state, product usage, or something else? Which timestamp is the real event time? Which table has the current truth and which one is a convenient trap?
These are not syntax questions. They are product/business questions.
If the model has to reverse-engineer the answers from table names, the system is already doing the risky part in the wrong place. A better model might guess more often correctly, but it is still guessing.
The semantic layer becomes the UI
In older BI systems, semantic layers often felt like cleanup work.
People relied on them to make a bunch of disjointed dashboards agree, to define metrics once, and to stop five teams from shipping five versions of the same "pipeline".
Still useful, but ultimately, a little thankless.
With AI analytics, that layer moves closer to the center of the product. It's not just there so dashboards render the same number. It is there so the model can understand what the company means when someone asks a question.
That includes metrics and joins, sure, but it also includes the words people actually use, the questions they tend to ask, the tables everyone should avoid, the caveats that never made it into the dashboard, and the permission rules nobody wants Claude to improvise.
Raw warehouse access is not enough because a warehouse schema is not a business interface. It is an implementation detail with years of decisions fossilized into it.
Analysts become product owners
With all of that in mind, I do not buy the idea that analysts will vanish. If anything, the job gets weirder and probably more important.
Someone still has to decide what counts as the canonical metric. Someone has to notice that sales and product use the same word to mean different things. Someone has to look at a bad answer and say, "the model did what we told it to do, and what we told it to do was wrong."
That smells a lot like product work.
The product here is the organization's understanding of its own data, expressed in a way software can use.
Users experience that model every day. They just experience it through a chat box, dashboard, report, export, or agent.
The unglamorous version
You do not need Anthropic's exact stack to learn from it. Most companies could start with one messy domain and do the annoying work:
- pick one domain
- define the ten metrics people fight about
- document the joins and caveats
- collect real questions from users
- write evals for those questions
- require answers to show where they came from
- review failures weekly
This will not look impressive in a launch video, but it will make the system less embarrassing.
Side note: you'd also be surprised at how quickly expanding coverage can be once you have some of these pieces in place.
Takeaways
A model alone can make analytics feel conversational, but that conversation is only useful if the system knows what the words mean.
So yes, Claude doing analytics is interesting, but the real lesson isn't as flashy: self-service analytics works only when the data is good enough to be treated like the product.