Attribution Caveats: Correlation vs Causation

In this episode, Scot digs into a question he keeps getting from readers and listeners: What’s actually going on inside Amazon’s Rufus, and what should brands do about it?

Ian brings fresh analysis from Sensor Tower’s panel-based methodology (privacy-compliant, double opt-in), walking through Holiday ’25 shopping sessions to show where Rufus shows up in the funnel, how usage spikes during peak moments, and why it appears to correlate with dramatically higher conversion. Along the way, they zoom out to the bigger shift: as more people learn to “talk to AI” (thanks to ChatGPT-style habits), conversational shopping becomes increasingly normal, and increasingly hard to “optimize” using old keyword-era tactics.

Highlights:

• “The consumer is trained.” A year of daily conversational AI use has taught people how to prompt, so when they see Amazon Rufus, they already know how to use it.
• Rufus sessions show a major conversion lift. In Ian’s read of the report: ~3.5× lift vs. non-Rufus sessions, with Rufus-touch sessions far more likely to end in purchase.
• Rufus had outsized influence during Holiday ’25. A large share of purchases included Rufus interaction, even if not every session did.
• Their team mapped ten distinct Rufus-assisted shopping paths, including the standout “Research Conversationalist” profile.
• Correlation vs causation is real. Ian flags measurement caveats: not every “non-Rufus session” is a shopping mission; intent bias exists; so the takeaway is directional, but meaningful.
• Brands should shift from “keyword jail” to “product truth.” Better attributes, clearer specs, stronger review signals, and real storytelling matter more in conversational shopping.
• AEO anxiety + the brand-level rebound. Ian argues the future isn’t just “optimize for prompts”, it may reward brands with stronger reputation and social proof signals (think forums and communities).
• AI can talk consumers out of premium. They’ve seen examples where AI steers price-sensitive users away from expensive brands, an early warning system for brand teams.
• Retail media won’t vanish, but it will mutate. In a lower-click world, retailers will experiment heavily to preserve value, without turning the experience into ad soup.

If you’ve been treating Rufus like a curiosity, this data makes it hard to ignore: conversational commerce isn’t “coming”, it’s already shaping how high-intent shoppers decide.

Timestamps:

00:00 – The Consumer Is Now Trained

02:09 – Why This Episode Focuses on Rufus

05:49 – Ian Simpson’s Background

06:47 – Founding a Bottled Tea Startup

09:14 – Pathmatics → Sensor Tower Acquisition

10:11 – How Retail Media Intelligence Was Born

13:50 – How Sensor Tower Delivers Its Data

15:36 – Why Sensor Tower Started Tracking Agentic Commerce

21:08 – How the Rufus Analysis Was Done

22:48 – The 3.5× Conversion Lift Explained

24:18 – Amazon’s $10B Rufus Claim vs. Independent Data

28:16 – Correlation vs. Causation in Rufus Usage

31:23 – Rufus as a Research Companion

32:20 – The Cup Holder Story (Problem-Based Shopping)

39:55 – Why Rufus Usage Sticks After the Holidays

40:19 – The 10 Rufus Shopping Profiles

41:35 – The “Research Conversationalist” Funnel

45:00 – Why “AEO” Makes Ian Nervous

46:47 – Brand Matters Again in Agentic Commerce

48:47 – When AI Talks Consumers Out of Premium Products

58:21 – Retail Media’s Future in a Low-Click World

01:01:31 – Avoiding the Minority Report Ad Nightmare

01:06:21 – Instacart: The Wild West of Retail Media

01:07:15 – Where to Follow Ian

👉 Connect with Ian: https://www.linkedin.com/in/iansimpson/
👉 Learn more about Sensor Tower: https://sensortower.com

Check out the materials we discuss:
🔗 Sensor Tower Black Friday AI Trend Update 
🔗 How Amazon’s Rufus Shaped Holiday Shopping

🧠 Want to stay ahead in AI commerce? Subscribe and follow along:
📰 Subscribe to the free Substack: retailgentic.ai
📺 Watch episodes on YouTube & Subscribe for updates: https://www.youtube.com/@Retailgentic