Had an interesting experience with Claude this week.

The goal: a weekly report with the most interesting problems and opportunities to investigate through experiments.

This kind of research is crucial for any world-class CRO program, but VERY time-consuming when done the old-school way.

But Opus 5.5 still kinda sucked at separating gold from low-value noise.

It would also very often miss awesome opportunities emerging from changes in customer behavior.

I was getting small improvements by tweaking the workflow and using JEV to help with clustering.

But the output was still much worse than what I’d get from doing the research myself or delegating it to a good researcher.

Then I had this crazy idea:

Ask Claude to "interview" me about how I actually think when looking for opportunities in customer feedback.

That turned into a 2-HOUR conversation.

Claude asked me questions, showed me how it was interpreting sample data, asked for my feedback, and had me analyze real customer comments so it could “watch me working.”

I simply… talked. Literally, in voice mode. A LOT.

The end result was a MUCH BETTER system.

There are still a million things I want to improve.

But the output is already at the level I’d expect from an above-average researcher.

I’m excited about what this will become.

I knew how I did this work.

But apparently, there was a lot I hadn’t explained.

Whenever you feel like AI isn't really getting you when automating a process, try asking it to “interview” you to extract more details on your thinking and intuition.