A couple weeks ago I sat down with Claude. I had no agenda, no client deliverable, no specific question to answer, no dashboard brief.
I had a folder of CSV files about the city of Essen, Germany: population counts, birth and death records, physician data, business registrations, pharmacies, registered dogs. I had a lot of curiosity about what would happen if I handed them to an Claude and said, go.
What followed genuinely surprised me. It was pretty eye-opening.
The first thing Claude did was ask me a question. Not a generic one like "what would you like to do?" but a specific one: “Who's the audience, and what decisions do they need to make?” All I said was the Mayor. I was being vague, like many of our stakeholders. It kept probing for more clarity. Once I provided more clarity, Claude said it understood then “Now show me the files before I propose anything.”
That already felt different. It wasn't jumping straight to output. It was behaving the way a thoughtful analyst should: gather requirements and context before touching the data.
Then it looked at the files and came back with actual observations. It noticed that the physician data only ran to 2020, and flagged that upfront rather than burying it in a footnote. It spotted that deaths were outpacing births across multiple districts. It identified that general practitioner counts had fallen 34% since 2004 and said plainly: “This is the most urgent signal in the healthcare data, and I'd want to lead with it.”
Before building a single chart, Claude asked me three more questions.
- How should the dashboard be organised? By theme, by urgency, or by district?
- How much geographic detail matters?
- What visual style fits the audience?
I answered, it proposed a specific layout with specific metrics, and it paused again and asked: does this plan work, or would you like to change anything?
I said yes, with one note: flag the 2020 data cutoff clearly so the Mayor knows it's potentially outdated.
And then it built the dashboard.
A couple minutes later a full interactive HTML dashboard appeared with four themed sections: Population, Healthcare, Economy, Housing and Community. There were red and amber action flags surfaced prominently, a sortable district table showing all 50 districts with natural growth status indicators, and ten charts rendering correctly with proper aspect ratios. The whole process, from first message to finished file, felt like working with a sharp junior analyst who happened to have no ego about asking for direction.
Here's what got me. The output wasn't what surprised me most. It was the process.
I've seen AI generate dashboards before. They were terrible.
What I hadn't fully internalised until this session is how much the quality of the output is a direct function of the quality of the conversation that precedes it. Think about that…isn’t that exactly what the best analysts do?
Claude didn't just execute instructions. It developed the brief alongside me. It pushed back on generic instructions and replaced it with concrete decisions. It noticed things I hadn't asked it to notice. The 34% GP decline was something I saw, but I didn’t flag it for Claude; It found this critical insight because it understood the context well enough to know it mattered.
That's a different kind of tool. That's something closer to a thinking partner. That’s something closer to a great data analyst.
I want to be honest about what this wasn't. It wasn't magic. It wasn’t perfect the first time. There were errors along the way. Working through those required me to understand what had gone wrong and describe the fix clearly.
The AI without the analyst in the loop would have produced something broken and left it there.
But the analyst without the AI in the loop would have spent most of a day cleaning, calculating, and laying out what we built together in under an hour.
That gap, the difference in what you can produce in the same amount of time, is what I keep coming back to. This was true productivity improvement.
And I think it's the most important thing happening in our field right now. Not the question of whether AI replaces analysts. It won't, not the good ones. The question is whether analysts who embrace these tools start pulling so far ahead of analysts who don't that the distance becomes impossible to close.
I think the answer is yes. And I think the window for getting comfortable with this, rather than just curious about it, is shorter than most people realise.
Which is exactly why I want you at the free Next-Level Tableau Live Conference April 28th.
Two sessions in particular are sitting at the center of this conversation.
Tim Ngwena is opening the discussion with AI won't replace analysts — but analysts using AI will, which is precisely what I walked away with from this project.
And Celia Fryar is giving us a first look at Tableau Next and the Semantic Data Model. This is the infrastructure shift that's going to determine how AI and analytics actually connect in the tools most of us use every day.
Those are two sessions I'd attend even if NLT wasn’t running the conference.
If you've been meaning to sign up, I hope this is the nudge you need.
I'll see you there. Register here, completely free.
— Andy