Product Led Growth Leaders

There are a handful of genuinely useful ideas—behavior patterns vs. click-level funnels, the 'research shopper' archetype surfaced from looping back to a review page, and the 'AI wash' critique of agents running on bad underlying data—but they're heavily diluted by repetitive pitch language and restatements of the same core argument across the episode.

“the average behavior pattern of a larger e-commerce company is over 50 steps...People are building six and seven-step funnels”

“if you understand that they went all the way to shopping cart and then went back to the review page...We call that a research shopper”

The 'AI wash' framing—agents can't rescue bad underlying data—is a pointed, contrarian take worth hearing, and the argument that funnels and agentic UX are structurally incompatible is moderately fresh; but the foundational critique of funnel analytics over behavioral patterns is a well-worn argument in analytics circles.

“they call it AI wash because all of these sort of modest technology companies...say, oh, agents are the answer”

“Funnels and agents will never cross”

Keith is a genuine enterprise software CEO with 15 years operating at Conviva and prior executive stints at Cisco and Riverbed; the Databricks co-founder connection adds technical credibility. However, the entire conversation is in pitch mode rather than practitioner reflection, so his operational depth is mostly asserted rather than demonstrated.

“My co-founder, Jan Stoika, actually was the founder of Databricks. So that gives you a perspective on Conviva.”

“I've never been a part of a failure. So every company that I was fortunate enough to join had a great ending, great success, great exit.”

A handful of concrete figures appear—8 billion devices, 50+ average steps per e-commerce behavior pattern, 150-step outliers, three minutes of wasted agent re-contextualization—but there are no named customers, no conversion lift percentages, no A/B test results, and no dollar figures tying behavior changes to revenue outcomes.

“we sit today in about 8 billion devices”

“the average behavior pattern of a larger e-commerce company is over 50 steps...We see some consumer behavior patterns in our large e-com companies of 150 steps”

The host occasionally asks a substantive question (on AI accuracy and data integrity) but never follows up with a challenge, never asks for a named customer result, and repeatedly closes with affirmations like 'super, super interesting stuff.' The interview functions mostly as a product demo setup rather than a rigorous dialogue.

“How do you grapple with the problem of correctness, Keith?”

“Wow, super, super interesting stuff and exciting things. And you guys are illuminating the nuances and details of that world.”