Writing

Fivetran is one of the best ETL products on the market. Its pricing on CRM and email connectors is the part I've never been able to justify, and now I save it for sources with less changing dimensions.

The components of a governed LLM workflow, drawn from systems we build and run and written to apply whatever tools you use.

Content libraries are data products. I built the taxonomy, measurement, and pruning infrastructure for one spanning 80+ domains after three acquisitions.

Field notes from building a Salesforce opportunity creation agent with a colleague. The model wasn't the hard part.

Field notes on designing, hiring, onboarding, and building momentum with a data engineering team at a paid performance marketing agency.

A consumer goods brand built their marketing around one customer profile. The data showed a different one had been buying the whole time.

A client project reminded me that the hardest part of shipping AI isn't the model. It's the foundational data work that has to happen first.

AI can cut days off workflows that used to take weeks. The outcomes depend on honest answers to a few questions about the data underneath.

Field notes on the real tradeoffs after three years of fractional data work.

The difference between reporting that gets adopted and reporting that gets ignored is usually upstream of the dashboard itself.

How a book written for corporate CDOs helped me think about consulting projects, team dynamics, and the transition from building to guiding.