zlacker

> We wanted to build data cleaning software with LLMs, before we realized no one really cared about messy data.

This sounds like a good problem to solve with LLMs, and honestly I'm a bit surprised to hear that nobody cared about data being of poor quality. Care to elaborate a little bit?

>>kjok+(OP)
I should have been a little more careful in my phrasing. Certainly some number of people care about data quality. The motivation for the project came from my own personal experience -- it was a product that I really wanted.

Our positioning (e.g. principal use cases, target verticals, target persona) and timing were totally wrong. We tried to run top-down sales against relatively important data (e.g. sales forecasts). People often expressed an attitudinal interest in the product, but it was very hard to motivate people actually to use the product seriously. Category creation and behavior change are really hard.

If I were trying to sell an LLM data cleaning product now, I'd focus on more technical buyers, someone more like a data engineer than a strong spreadsheet worker. And I'd try to drive bottoms-up, low ACV adoption first.

Doing the post-mortem justice would probably require a long blog post.