Toucan Talks: Winning in the AI Commerce Era

Amazon AI algorithms: exploring COSMO

Sam opened the day by showing how far AI has already reached into Amazon, on both sides of the shelf: AI listing generation for sellers, an AI seller assistant, review summaries, and Rufus, used by more than 300 million customers over 2025, with shoppers who engage with it completing purchases at a rate more than 60 percent higher. The core of the talk was COSMO, Amazon's common sense knowledge system, introduced quietly in 2024. It does not replace the A9 algorithm, which still ranks on keyword matches and performance signals, but adds a layer of intent on top, mapping products to the needs, uses and situations a shopper describes. Sam took the room through COSMO's 15 relation types, grouped into five families: what a product does, who uses it, where and when it is used, what it is, and what it connects to. The practical close was a four-step routine that optimises for COSMO, Rufus and A9 at once: audit the listing, read the reviews against those relations, do the keyword research, then combine all three.

Winning with AMC: unlocking paid features for growth

Mahlah's talk moved past standard Amazon Marketing Cloud analysis to its paid features, which bring organic, non-ad-attributed shopping signals into the clean room. Flexible Shopping Insights set ad-exposed shoppers against those who engaged with a brand without seeing an ad; for one personal care brand, ad-exposed audiences accounted for 82 percent of purchases over 13 months and converted at 41 percent against 1 percent for everyone else. Audience Segment Insights showed that two of the top six audiences driving that brand's purchases were not being targeted at all. Amazon Retail Purchases stretches the look-back from 13 months to five years and needs no advertising data, which opened up five-year lifetime value analysis for a supplements brand and new-to-brand versus repeat trends by ASIN for a beauty brand. Brand Store Insights showed which ad paths into a store actually convert, and where off-Amazon traffic was arriving from. The talk closed on Amazon's Ads Agent, which writes AMC SQL from plain English, and on cost: the paid features are free until 31 December 2026.

Alexa+: agentic commerce at scale

Duncan picked up where Sam and Mahlah left off, with what Amazon is doing with AI on the shopper side and in the ads console. Rufus and Alexa+ are being combined into Alexa for Shopping, pairing Rufus's product knowledge with Alexa's memory and context across every Amazon surface and device, and conversational shopping already converts at around three and a half times the rate of keyword search. The ads have to work inside that conversation rather than interrupt it: Sponsored Prompts already run off existing Sponsored Products and Sponsored Brands campaigns, interactive audio ads let a listener add to cart mid-ad, and branded conversations, due around the first quarter of next year, will let shoppers question a Sponsored Brands collection without rival brands appearing, which is why Duncan urged anyone running Sponsored Products alone to build a Sponsored Brands strategy now. Because Alexa reads a brand's own site and the open web as well as its listings, content needs regular audits and one consistent story everywhere. He closed on the DSP: Performance+ for remarketing that bids on a shopper's propensity to buy rather than just recency, and DVA+, which combines Performance+ and Brand+ next year across display, video and audio on the open internet, with advanced settings keeping room for AMC custom audiences.

Beyond the prompt: the human edge in AI design

Chloe started with a test: which of these designs were made by AI? The point was less the answer than the fact that AI imagery now has a recognisable visual language of its own, excessive symmetry, staged lighting, awkward spacing, too much information on one tile, and shoppers are learning to read it. On Amazon that matters more than anywhere, because there is no smell, no salesperson and no second chance, and the design has to build trust, desire and value on its own. Studios' answer is to use AI as an intelligence tool rather than an image generator: mining reviews for pain points and unmet needs, reading category conventions and competitors, then letting designers decide what the shopper needs answered next. Three case studies showed it working, including a dog food listing set reordered around the nutrition and digestion concerns that AI review analysis surfaced, and a toy brand's Brand Store home page where sales from the page rose 50 percent and dwell time 25 percent in the month after Prime Day. The takeaway: AI plus human, not AI versus human.

Shop, optimise, measure

Kate, from the day's sponsor Xnurta, framed AI in Amazon advertising as three shifts. Shop: two thirds of UK consumers used AI to research or plan a purchase in the first quarter of 2026, so listings need structured attributes and content that answers what a shopper, or their assistant, is actually asking. Optimise: advertising is moving from hand-set rules to AI-assisted recommendations to agents acting inside guardrails the team sets, and Kate laid out four levels of hand-over, observe, recommend, execute on approval and execute, with anything new starting at the bottom. A US drinks brand that took this approach grew its new-to-brand rate by 21.5 percent and ROAS by 10.4 percent while cutting cost per acquisition by 16.7 percent. Measure: when AI runs every stage of the funnel, the funnel has to be measured as one, across Prime Video, Twitch, Fire TV and streaming as well as amazon.com. Her one thing to take away was that AI gives a team more control, not less.

AI vs authenticity: the future of TikTok Shop content

Dania's talk was the counterweight to the AI-heavy sessions before it: shopping is still an experience shaped by human connection, from the first impression to the decision to come back, and TikTok Shop is the one platform that recreates every one of those touchpoints. LIVEs bring real-time answers, demonstrations and limited-time offers, and turn viewers into participants whose questions help other shoppers decide; one beauty brand's 14-hour Christmas LIVE took around 2.7 million. Content is a brand's representation on the platform, built on strong creative briefs, engaging creators, consistent posting and content made for discovery, and backed by a brand that replies to comments and reposts its affiliates' videos. AI creators, she argued, cost a brand its authenticity outright. Where AI earns its place is the affiliate messaging sequence, from sample invitation to approval, content reminder and thank you, so a brand can reach tens of thousands of creators. With affiliates driving at least 80 percent of a brand's revenue on TikTok Shop, the relationship itself, built through WhatsApp groups, creator competitions and affiliate webinars, is still human work, and it is what lasts.

How TikTok Shop is changing the way people shop

TikTok Shop closed the day, with Amal giving the platform's view of where it sits in a brand's channel mix. TikTok has more than 44 million monthly users in the UK, who spend around a quarter more time on it each day than on YouTube; TikTok Shop was the fastest-growing UK online sales platform in 2025, and Black Friday sales were up 50 percent on the year before. The central argument was that TikTok Shop reaches people who were not searching: traditional ecommerce runs need, search, compare, buy, while social commerce runs scroll, discover, buy, creating demand rather than cannibalising other channels. That demand does not stay on TikTok, and in one case study 62 percent of TikTok Shop's incremental impact landed on Amazon, so pricing, stock and messaging have to be planned across channels and the halo measured rather than judged on last-click sales. The talk closed on five shifts to plan for, social becoming search, creators becoming the sales force, LIVE going mainstream with a fixed schedule of around three a week, cross-border selling and AI-shaped buying, and on trust as the thing that decides who wins: choose creators for audience fit over follower count, and deliver on what they promise.

In the room