What breaks when people talk to data — and how to design for it instead
Press enter or click to view image in full size
It’s 2:47 PM on a Tuesday. Your CFO opens Teams/Slack and types: “What’s our burn rate if we maintain current hiring plans through Q2?”
Three years ago, this question would have triggered a chain reaction: Slack message to analytics team → someone opens Looker → builds a custom dashboard → schedules a meeting → presents findings on Friday.
Today? The answer appears in 4.3 seconds, complete with a breakdown by department and a confidence interval.
This isn’t science fiction. This is the new baseline expectation.
And if you’re still designing your data warehouse like it’s 2019 — optimized for dashboards, reports, and SQL-literate analysts — you’re building the wrong thing.
The Core Thesis
The LLM is not your bottleneck. Your data model is.
Everyone’s building conversational analytics by bolting ChatGPT onto their existing warehouse. That’s backwards. Your star schemas, medallion pipelines, and pre-aggregated dashboards were optimized for a world where questions were predictable and analysts understood SQL. Conversational interfaces…