Avsnitt
-
In this episode of Knowledge Distillation, Katrin Ribant speaks with Kelly Wortham – founder of Forward Digital and the Test & Learn Community, a network of several thousand experimentation practitioners, and curator of the Experimentation Island conference. Kelly came into the industry from academia and social sciences, where running experiments meant clipboards in shopping malls and t-tests on hiring processes. That grounding in the messy real world is exactly what makes her take on agentic commerce so sharp: she sees the current shift not as a technology problem, but as a measurement crisis hiding in plain sight.
The core of the episode is a problem most experimentation teams haven’t fully reckoned with yet. As AI-referred traffic grows, visitors arrive at websites already persuaded – they’ve done their comparison shopping in ChatGPT or Perplexity and are now just confirming what they already know. A/B tests designed to persuade don’t work on people who’ve already been persuaded. And because most companies have no clean way to separate AI-referred from non-AI-referred traffic, results get blended into noise. Kelly’s practical advice: start segmenting AI-referred traffic now, even if the data is messy, because building that muscle early is more valuable than waiting for a clean solution that doesn’t exist yet.
The episode closes on a category Kelly calls brand impact tests – experiments that happen entirely off your website, in the third-party content ecosystem that AI models are trained on: reviews, product descriptions, social mentions. These are the inputs that shape what an AI recommends before a customer ever lands on your page. And on a provocation both find genuinely exciting: after years of optimizing for clicks and dark patterns, agentic commerce may be forcing brands back to clarity and human-first design – because optimizing for machines increasingly means optimizing for humans.
All episodes on our website: www.ask-y.ai/knowledge-distillation-podcast
Learn more about ASK-Y: www.ask-y.ai
-
In this episode of Knowledge Distillation, Katrin Ribant speaks with Gal Rapoport – co-founder and CEO of Kahoona, a company building what he describes as the context and memory layer the web never had. Gal’s background in AI goes back well before the current hype cycle: he helped build what became AWS Inferentia and AWS Trainium at Amazon, then joined the Alexa Shopping team as its first machine learning hire, working on personalization at a time when transformers were still an internal experiment. After leaving Amazon, he pursued a PhD in multimodal AI focused on human-computer interaction and the coming data scarcity problem – and realized that everything he knew about personalization from Amazon simply didn’t exist in the open web. That gap became Kahoona, which has since won the LVMH Innovation Award for best business impact and a similar recognition from the Kering Group.
The central concept of the episode is digital body language: the idea that how a user moves through a website – the speed of their scroll, where they hover, how long they pause – carries as much signal as what they click on. Gal explains how Kahoona captures this through a lightweight script, then trains models on those behavioral signals to infer intent within moments of a user arriving on a page, for the anonymous visitors who make up 97% or more of traffic. The conversation then takes a sharp turn into agentic commerce. Katrin raises the obvious tension: Kahoona was built to read human behavior, but AI shopping agents don’t hover or browse – they execute with surgical precision. Gal’s answer is counterintuitive: agents are actually easier to model than humans. Humans are noisy and shift intent mid-session. Agents have a clear mission and low behavioral variance, which makes their intent more predictable, not less.
The episode closes on what brands should actually do with this today. Most are blocking agents by default, not out of hostility but because they have no policy framework yet. For analysts working with GA4, Gal offers two concrete signals to watch: declared agent traffic segmented by geography, and the correlation between declining time-on-site and rising conversion rates – a pattern suggesting agents are doing the research upstream and sending humans to the site already primed to buy. His bigger prediction: websites will soon detect whether the incoming actor is human or agent and route them to entirely different experiences – one visual and exploratory, one structured and markdown-readable. The brands that build that infrastructure now, before agents become the dominant traffic source, will have the advantage.
All episodes on our website: www.ask-y.ai/knowledge-distillation-podcast
Learn more about ASK-Y: www.ask-y.ai
-
Saknas det avsnitt?
