I stopped answering questions.
Every client question came to me, and only to me. I decided to give someone with no SQL background access to all of it — and see what happened.
I was building a B2B SaaS analytics platform. Every time a client asked something new, it came to me.
I'd go back to their engineering team, understand whatever part of the database I hadn't mapped yet, write the query, come back with an answer. It worked. But it meant nothing could move without me.
The product guy understood the business better than I did. He knew what clients were trying to measure, why certain numbers mattered, what questions were likely coming next. But he couldn't write SQL. So his understanding stayed in conversations and never made it into the data.
The problem I had been thinking about was this — analytics knowledge lives in people's heads. When an analyst joins a company, the first thing they do is learn the data model. When they leave, that learning leaves with them. The next person starts from zero. I had built a semantic layer to solve this — a model that encoded the meaning of a company's data in plain terms. Measures, dimensions, relationships. Once it was there, anyone who understood the model could query it without knowing SQL.
I hadn't tested whether this was actually true.
I gave the product guy access to the query layer of the first client's model. Not the modeling work itself. Just the ability to explore what was already there.
He started slowly. But querying the model required understanding it, and understanding it required engaging with how the client's data was structured. He started making connections. He came back with more than what was asked — patterns adjacent to the original question, correlations he had spotted while exploring. The friction of SQL had been sitting between him and the thinking he was capable of.
By the second client he was querying the first client's data on his own. By the third he was handling modeling and querying for both. I jumped in when something was complex. Everything else was his.
This was the first real test of the hypothesis. If the semantic layer worked the way I thought it did, someone without a technical background should be able to pick up a company's data context and run with it. He did. The knowledge stayed in the model. He built on top of it.
I stopped answering questions because he could answer them. And knowing that the approach worked was worth more than any single answer I had given before.