Latest / The CMO Podcast with Fexingo: Marketing Leadership, Budgets, and Executive Strategy / How CMOs Use AI Agents for Personalization at Scale
Transcript
- Lucas: You know that feeling when you open a retail app and it shows you something you actually wanted, but you never searched for it? That moment of 'how did they know' is becoming the baseline expectation, not a nice surprise. Luna: Right, and the gap between 'how did they know' and 'why are they showing me this random thing' is basically the difference between a good customer experience and a lost sale. Lucas: Exactly. And increasingly, that gap is being closed by AI agents — not just recommendation algorithms, but autonomous pieces of software that learn, adapt, and act on behalf of the brand in real time. I want to look at how CMOs are deploying these agents for personalization at scale. Luna: Let's get specific. Stitch Fix is the canonical case here — they've been doing this longer than almost anyone. Their recommendation engine isn't a simple 'people who bought this also bought' model. Lucas: No, it's a reinforcement learning system. Each time a client receives a Fix — that's their curated box of clothing — they keep what they like and return the rest. The agent gets a reward signal from every decision: what was kept, what was returned, what fit feedback was given. Over time, it builds a multidimensional profile that goes way beyond demographics. Luna: And they layer human stylists on top. The AI agent generates a candidate set, the stylist curates it, and that human touch actually trains the agent further. It's a hybrid model. Lucas: Which is exactly the kind of approach more CMOs are adopting. Pure AI personalization can feel cold. Pure human personalization doesn't scale. The hybrid is where the magic happens. According to a 2025 survey by Gartner, 43 percent of CMOs at companies with over 500 million in revenue have deployed some form of AI agent for personalization. That's up from 22 percent two years prior. Luna: Forty-three percent — that's a serious adoption curve. But I wonder how many of those are just using glorified segmentation rules and calling it an AI agent. Lucas: That's the critical distinction. A true AI agent doesn't just bucket people into segments. It maintains a dynamic model of each individual. Amazon is the obvious comparison here — their recommendation engine processes billions of interactions daily, but it's largely transactional based on purchase history and viewing patterns. It's powerful, but it's not learning from, say, how you felt about a product after using it. Luna: Stitch Fix gets that emotional data through the return mechanism. They know not just what you bought, but what you rejected and why. That's the kind of signal that builds real preference models. Lucas: And that's exactly what's driving the next wave. CMOs at companies like Sephora and Nike are experimenting with AI agents that engage in natural language dialogue with customers — asking questions like 'what's the occasion?' or 'what's your budget?' and then building a personal profile in real time. It's like having a personal shopper in every channel. Luna: But that raises a huge question about brand voice. If you've got millions of conversations happening simultaneously, how do you maintain consistency? One agent might come across as too pushy, another too passive. Lucas: That's the brand governance challenge. The smart CMOs are defining a 'brand personality matrix' — a set of guardrails for tone, vocabulary, and response patterns. They then train the AI agent within those boundaries. It's not unlike training a new employee on the brand bible, but at machine scale. Luna: So the CMO becomes part product manager, part brand cop. They have to ensure the agent doesn't veer off message while still being adaptive enough to be useful. Lucas: Exactly. And the best ones are building feedback loops directly into the agent's interface. When a customer rates a recommendation as 'not helpful', that doesn't just get logged — it becomes a training example that updates the model within hours, not weeks. That's the difference between batch personalization and real-time personalization. Luna: Let's talk about the privacy side. These agents are collecting vast amounts of behavioral data. With GDPR, CCPA, and now the European Union's AI Act taking full effect this year, how do CMOs navigate that? Lucas: It's the number one concern I hear from marketing leaders. The approach that's gaining traction is 'federated learning' — the AI agent trains on user data locally on the device, and only sends anonymized model updates to the central server. So the brand still gets the intelligence, but raw personal data never leaves the customer's phone. Luna: Apple has been pushing that with their on-device machine learning. For a brand that's already in the Apple ecosystem, that could be a huge advantage. Lucas: Absolutely. But regardless of the tech stack, the CMO has to be transparent. The brands that win trust are the ones that say, 'here's what we know about you, here's how we use it, and here's how you can control it.' That's not just compliance — it's a competitive differentiator. Luna: So if I'm a CMO listening to this and I haven't started with AI agents, what's the first step? Do I need to build a Stitch Fix-level system from day one? Lucas: No, and that would be a mistake. Start with one use case where personalization has high leverage and low risk. For a retailer, that might be email subject lines. Train an agent to A/B test subject lines dynamically at the individual level — not just segments — based on open rates. That's a low-stakes way to prove the concept and build internal capability. Luna: And then expand to product recommendations, then to content personalization on the website, then to conversational agents. A crawl walk run approach. Lucas: Right. And each step builds data that feeds the next. The companies that do this well treat personalization as a compounding asset — every interaction makes the next one better. I think we're going to see a world where every brand has a 'personalization maturity curve', and CMOs will be measured on how quickly they can move up that curve. Luna: One more thing I want to touch on: the emerging trend of agent to agent marketing. Where your AI agent interacts with another brand's AI agent on your behalf. Lucas: That's fascinating. Imagine a consumer authorizing a personal shopping agent to negotiate the best price on a new laptop. That agent then dialogues with multiple brand agents — or retailer agents — to find the best deal, and it can even haggle in real time. Some early implementations are already happening in B2B procurement. Luna: For CMOs, that means you're not just marketing to humans anymore. You're marketing to other AI agents. The creative and strategy have to account for how an algorithm will evaluate your offer. Lucas: Exactly. It flips the funnel. Instead of top of funnel awareness, you need bottom of funnel precision. The agent doesn't care about your brand story — it cares about price, availability, delivery time, and return policy. So your marketing has to feed those signals into a machine-readable format. Structured data, APIs, maybe even a dedicated 'agent channel'. Luna: This is a whole new competency for marketing teams. They'll need to collaborate with engineering and data science in ways they never have before. Lucas: And the CMO has to lead that integration. If today's episode sparked something you've actually used in your planning or just made you think differently about your personalization roadmap, one way to keep these conversations ad-free and independent is listener support. You can head to buy me a coffee dot com slash fexingo — that's buy me a coffee dot com slash fexingo. Luna: Yeah, it's a small gesture that helps us keep bringing on the best guests and digging into the topics that matter most to marketing leaders. No pressure, just appreciated. Lucas: So back to the agent to agent future — I think the CMOs who start experimenting with these autonomous agents now, even in small ways, will have a massive advantage in three to five years. The early movers in personalization are already pulling ahead. Luna: And the ones who wait for the perfect tech stack or the perfect data set? They'll be playing catch-up. It's about starting, not waiting. Lucas: That's the key takeaway. AI agents for personalization aren't a futuristic concept — they're here, they're working, and the only question is how fast you can integrate them into your marketing operations. I'm Lucas. Luna: And I'm Luna. We'll be back next time with another conversation that'll make you a smarter marketing leader. Until then, keep testing, keep learning.