20 years in, I’m still asking: what if we tested this?

AI made production cheaper.

It made the fundamentals more valuable.

AI made it cheaper to produce more marketing. It did not make it easier to decide what deserves to exist. When a team can generate more variants than it can responsibly test, production stops being the main constraint. The harder part is judgment: which customer behavior matters, which idea is worth testing, and what result would change the decision.

That puts the value back on the unglamorous work. Read the behavior underneath the dashboard. Design a test that can prove the assumption wrong. Compare the result with a meaningful baseline or holdout. Trust what customers do over what they say they might do. Write something someone is glad to receive, not one more message to ignore.

I learned that early in my career. The model was never the whole answer. The harder part was turning its output into a signal marketers could understand, trust, and use when making decisions. That lesson has stayed with me through every role since.

Across ecommerce, SaaS, global industrial tools, and big-box retail, I’ve seen technology create an advantage when it shortens the distance between a customer signal and a better decision. It creates noise when it increases output without improving judgment or measurement.

The tools and playbooks here are built around that distinction. Run them against your own data and constraints. Keep what changes the result. Revise or discard what does not.