6 Ways I Use Claude for Data Analysis

Product analytics is one of Claude’s superpowers. Being able to quickly crunch through product and analytics data means I can make decisions faster, in a more data driven way and with less dependence on the team. Here are 6 ways I am using Claude to help me analyse data.

1. Replacing Dependencies: Claude as a Personal Data Analyst

In my post about Using Cursor and MCP as a Product Manager I wrote about using Cursor to explore analytics in PostHog, document findings in Notion or Google Sheets, and draft tickets for the team in Linear. All without me leaving the Cursor UI. I am now doing all of this in Claude Cowork instead.

Good documentation is one thing that massively improves Claude’s data analysis capabilities. Unlike many humans (myself included), Claude actually reads and sometimes even follows instructions! Our data scientists have written a Claude skill shared organisation wide which captures all the best practice ways of working with our analytics. I grew used to the answer of “have you tried asking Claude” when I asked data scientists for support with analytics, and am now fielding most queries myself via Claude. This frees them up to focus on higher value activities, and helps me get the information I need faster.

2. Validating UI Risks: Analysing Document Title Legibility

Many of the entries in our database contain more than one document. In our passage search UI, I worried that document titles alone wouldn’t be enough to help a user distinguish between text passages from different documents. I even had work planned for the team to fix this.

Instead, using Claude, I analysed all our document titles. I found that most of the variation in document titles was at the beginning of the string, making it easier for users to distinguish documents by title alone. Extra metadata did improve the situation, but not by enough to justify the extra effort and elements on the screen. I was able to resolve this risk myself without asking anyone to build anything.

3. Unlocking Insights Across Multiple Datasets

There is an excellent InBetweeners episode where jay says “BEEPITY BEEP BEEP” every time Simon mentions his new girlfriend Tara. At one stage, I joked about doing the same thing every time our lead data engineer Fred mentioned the data lake. But now that the data lake is in place, it’s clear to me what an amazing tool this is.

Having all of our product databases and product analytics available to query together in one place has unblocked some valuable insights. For example, with a little help from Claude I was able to analyse the relationship between document metadata and the documents that users are most likely to visit and come back to. There is a strong correlation between document summaries existing above a minimum size and engaged usage. This insight has helped shape the direction of my product strategy.

4. Understanding Document Text Quality

HTML is accessible. PDFs very much are not! I wanted to understand whether the text data extracted from PDFs in our database was of good enough quality to recreate the document in a HTML user interface.

I extracted full passage text via Snowflake from a sample of documents and loaded them into Claude. Claude’s analysis of this data helped me spot data issues for the team to prioritise in the next round of text quality improvements. Claude also helped me turn it into a HTML prototype for me to see how it looked for myself. Read more on this in my post on Claude Design.

5. Deciding Defaults with Advanced Query Analysis

Search at Climate Policy Radar enables users to search the full text and detect mentions of 100+ topics in the text of documents via an awesome set of text classifiers. It is a powerful feature, but it is not one I have seen before in many other products. It is not a feature users expect, so there is no normal way of doing it. I had a difficult decision to make about what the default logic should be for when users start combining classifiers (e.g., AND vs OR operators, the level at which co-occurrence should occur, interaction with free text search).

My work buddy Anne (policy analyst, whose great work I am shamelessly mentioning in this post) used Claude to run through the product analytics, identifying how many and which type of users were affected by this decision. Claude also crunched through search logs, categorising different searches by which default would have likely given the user a better set of results. This deep analysis changed my mind about some of the big assumptions I was making, gave me the evidence I needed to make some tough product calls around what not to build, and the data to get the team and stakeholders on board with the decision.

6. Claude as a Thought Partner

Another powerful use of Claude is to help me brainstorm what I should be measuring, and critique my existing ways of measuring product performance. I recently took stock of the end-to-end metrics I was tracking across our full product lifecycle. A deep chat with Claude helped me realise that my core engagement and retention metrics were too tightly coupled with user acquisition metrics. I digested the conversation, consulted with our data scientists, and we are now evolving our top line metrics in response.

Knowing when to ignore Claude is still very much a key skill. If I had followed every recommendation, we would have bloated analytics which tracks many of the wrong things. But I like to think of myself as an expert user who knows my domain, and I frequently find Claude a great thought partner in both strategy and execution.

Parting Thought

Putting analytics and product data in one place. Documenting it so it’s understandable. Chatting with the data via Claude to generate valuable insights. Generating digestible summaries for my team. These approaches have helped me get so much more out of AI than I was this time last year, and is making me a better product manager. Regardless of whether you are using Claude, Codex or Cursor, I’d strongly recommend giving this a try if your work involves analysing data.

More Importantly, some InBetweeners Trivia

The actors who play Simon and Tara in InBetweeners are actually together in real life! They have been for more than 15 years and are engaged. I stumbled on this amazing fact whilst trying to find a good images of “BEEPITY BEEP BEEP” for use in this blog post, which was surprisingly difficult to do.