MarTalks- The #1 Ecommerce and MarTech application podcast

The episode contains a genuinely interesting core thesis — that product intelligence (psychographic taste tags, OCEAN-derived scores) is the missing substrate that explains why personalization has failed — and explores it from several angles including agentic commerce. However, the same thesis is restated repeatedly without meaningful deepening, the opening AI discussion is generic, and multiple digressions (granola bar, travel destinations, Facebook news feed) add length without adding signal.

“everything we've called personalization today is optimization”

“we bypass needing to understand who the consumer is and can personalize right away for guest users through what they're interacting with”

Applying the OCEAN/Big Five personality framework to products rather than just consumers, and building a decisioning layer of 60+ psychographic taste tags as agent-interpretable embeddings, is a genuinely novel framing not commonly surfaced in e-commerce discourse. The distinction between latent representation and surface-attribute optimization is articulated with some precision. However, broader takes on Amazon-as-convenience and emotional shopping are well-worn.

“we are now supplying brands with...60 plus taste tags on top of that which are agent interpretable versions of our embeddings. So like a score of low for status seeking, a score for a higher score for romanticism, playfulness, femininity”

“the entities that win are the ones that capture the best latent representations of people and products. And latent is like the things that you can't see”

Annabelle Maldonado is a genuine practitioner: a founder who built her thesis from fashion copywriting into a deployed product with named clients (Kima Zabète, Toys R Us, Anne Klein) and a ShopTalk competition win. She demonstrates real command of consumer psychology and the e-commerce stack. The limitation is that Psyche is a startup still demonstrating scale, and much of the conversation is inherently self-promotional with limited external validation of her claims.

“we thought going into some of these pilots, we're probably going to do a floor of 2 to 3% and ended up being 22%”

“we're adding like 60 plus taste tags on top of that which are agent interpretable versions of our embeddings”

The episode contains a solid number of concrete anchors: third-party statistics from Deloitte, McKinsey, and Bain; named client deployments; specific figures like 700 dimensions, 60 taste tags, 22% lift, 1.5% conversion rate, and a $16M revenue unlock claim. Weaker points include the $16M figure lacking retailer or timeframe context, and several outcome claims from pilots without naming the brands or methodology.

“McKinsey found 71% of merchants say AI merchandising tools have had no effect on their business”

“In the package we sold recently at least 60 taste tags that we're adding to, I don't know, like two”

The host shows real preparation — referencing specific Deloitte, McKinsey, and Bain data points and using them as question prompts — and lands a few sharp conceptual questions ('are we turning commerce into procurement?', 'what's one data point no agent currently collects?'). However, he talks far too much, frequently answers his own questions, allows unchallenged self-promotion, and the opening framing ('bringing sexy back to online shopping') sets a soft-ball tone that persists throughout.

“McKinsey found 71% of merchants say AI merchandising tools have had no effect on their business. So far, seven in 10 merchants spent money on AI that didn't move the needle. Now in your conversations...what was that the root cause of that? Was it the bad. Was it bad tech or was it bad questions?”

“if an AI agent becomes the front door to that discovery, is loyalty going to just disappear entirely?”