A lot of product work begins with a solution already hiding inside the problem.
How should we use AI?
What should we personalize?
How do we increase conversion?
What should we automate?
I've spent much of my career backing up from questions like those.
Before deciding what to build, I want to understand the system: what the customer is actually trying to accomplish, what the business is trying to accomplish, what information matters, where the real constraint is, and which decisions actually change the outcome.
Very different
environments
That approach has taken me through very different environments.
Expedia Group
At Expedia Group, it meant moving beyond hundreds of fragmented traveler attributes toward a model of customer context: who the traveler is, what they're trying to accomplish now, and which information is relevant to the next decision. That work evolved across destination intelligence, customer context, generative AI, journey intelligence, and next-best-action decisioning.
Adorama / Printique
At Adorama / Printique, it meant questioning whether asking customers more questions was really the best way to personalize an experience. We instead learned progressively from behavior and the photographs customers uploaded, using Google Vision AI to turn image-level signals into customer-level context. In another case, it meant questioning whether promotional revenue spikes represented actual growth; a longer experiment showed that much of the apparent lift was shifting when customers purchased rather than creating incremental demand.
Big Splash Advertising
More recently, I've been applying the same thinking directly to operating businesses—designing AI and automation around the business system rather than treating automation itself as the objective. At Signature Concrete, for example, the apparent goal of booking more appointments gave way to a different problem: creating customer trust while allocating scarce human selling capacity intelligently.
Different businesses. Different technologies. Different scale.
The pattern is remarkably consistent: understand the system, find the problem behind the apparent problem, assemble the context that matters, and then decide what technology should actually do.
That's the work I'm interested in.