← Retail Media Breakfast Club
Loading summary
Transcript1 lines
- [00:00]
A
Chris Dermott of Spark Foundry got my vote for Line of the Day at X nurta's Signal to Scale Summit this week. She said, we did AI. To what end? That is not a verb, but that is what every marketer and retailer is now hearing from solution providers. We'll just AI it. But as Chris said, AI exposes the cracks in your organization. It does not heal them. And that was really the through line at Exnerta's Signal to Scale Summit on Tuesday. This week, every vendor across the ecosystem is selling agentic workflows of some kind, and brands are honestly very motivated to adopt it. I heard from one attendee that for better or worse, their leadership is asking for every new initiative to include how AI will be used. That kind of mandate means that brands generally want to get smarter on agentic workflows, and this event helped to put some of that theory into practice. I'm going to share some observations from this event with you and make sure you stay tuned to the end to hear about a great initiative that xnurture is opening up to the industry. So number one, some real use cases for agentic media buying. The clearest use cases came from Shah Atakanov, who is the VP Marketing and E Commerce at Voyron, the homeopathy and OTC brand, and he runs retail media with just two FTEs and they in their media campaigns are running roughly 50,000 keywords and he says of his team, they just can't possibly optimize all 50,000 keywords even if they were working 24 7. So AI isn't a productivity game here. It's the only way that that work gets done at all. The second use case from Boiron is creative. They started using an AI tool to score product and lifestyle images for conversion potential on one product that was marketed for babies, a cold symptom product. They kept iterating images, various images against the model score until they hit a great score of 90% and the conversion outcome was that sales increased 29% week over week. So a really great outcome. The funny detail was that the model kept downgrading an image with a mum holding the baby and they didn't understand why that would be, but they did indeed tried that campaign without the mum in the image and the score jumped and sales actually followed. So an interesting use case of AI really having a better insight than a human team might. Shah also gave a really great decision framework for actually selecting AI tools. There's three parts to this. One it has to contribute to growth. Two it has to save time or increase productivity and three reduce cost. Anything that doesn't do at least two of those things you don't buy. And critically, he says that guardrails are part of the spec. With AI you have to have good compliance, good boundaries. You can customize the AI and kind of set your goals and specific objectives because if you don't, the AI could do optimization on its own and it might not be contributing to your profitability. Next up, yi Chiang at BetterBeingCo was the other brand voice that really stood out, particularly on measurement. And he said that ROAS alone is an efficiency metric, not something that is a result that he is trying to achieve. And to that end he talked about share of voice and search query performance as their most underutilized levers, and about simplifying the dashboard noise by sorting metrics into inputs versus outputs. That's not a flashy answer, but it's a real practitioner's answer. And finally, Alex lands at Freebird and is one of the brand side practitioners really using a genticad buying in production today? His mental model is worth stealing for yourself. Think of it as a difference between gears in an automatic car and a stick shift or manual gear transmission. The car is going to drive either way, but if something goes wrong, the person who knows how gears work is the only one who can actually drive. Miracle Ads is the ad tech solution trusted by rakuten and over 50 global enterprise retailers. That's because Miracle Ads was built with both 3Pmarketplace sellers and 1P suppliers in mind. Both advertiser audiences demand a seamless advertising journey from onboarding to reporting. You can offer everything from sponsored products to to video ads all in one solution. Learn more@miracle.com that's M I R A K L.com now there's a little backdrop here from Amazon. None of this is happening in a vacuum. Liz Joyce, who is a technical product marketing lead at Amazon, gave a quick tour of the substrate. Amazon Ads has an MCP server which is live. They they launched Skills in March for those agentic processes, and it gives agencies a way of handling more complex scenarios. The same infrastructure that powers Amazon's own ads agent now lets brands plug in their own. But Liz also had some good advice on a more practical level. She's cautioned that too many signals are not better, and her advice is is to start by setting up AI evals or evaluations. More on that later. Map your workflows, figure out what's high frequency and repeatable versus what always needs a human. And finally Agent evals what is this? This is where the day got really interesting. Xnota launched Insight Agents, which is a redesigned agent that handles open ended queries against Amazon ad accounts, doing things like data pools, audits, root cause diagnosis and making recommendations. But the real banger to me was this eval framework. It's a hundred plus question framework and for those unfamiliar, an eval or evaluation is a structured way to test whether an AI system is actually performing well against the outcomes that you care about, using defined tasks, scoring criteria and benchmarks. So X notre shared how their agent compared against three Frontier AI models which were connected directly to Amazon ads via the MCP server. On data accuracy, The Frontier models scored 50 in the 50% range. Xnurta's purpose built agent scored over 80%. But X notice says it doesn't just want to grade its own homework, it it is open sourcing that eval framework and launching what they're calling the Agentic Retail Media Council, a panel of industry experts and practitioners, and explicitly inviting competitors too who can help to evolve that framework. X Notice CEO Kashif Safar said that this really is a version one and that they built it because someone had to start and he hopes that version two will will be built with the industry. And this really matters because like Liz Joyce from Amazon said, you need to set up your evals. That was the recommendation that she gave to brands and agencies in the room. But there's really a chicken and egg problem with that. Most brands and even many agencies don't have the engineering muscle to build agent evals from scratch. So an open industry built framework is the closest thing to a shared yardstick that I have seen proposed so far. And if the rest of the category shows up to it, this is a real piece of infrastructure. So this is worth watching closely and just a note out to everyone listening if you're interested in joining the AgentIQ Retail Media Council, Xnurda is recruiting members. I'll provide a link in the show notes of this episode. Thank you for listening and I'll catch you next week.