AI and Commerce Media - A Q&A with Members of Our Retail & Commerce Media Committee

AI is rapidly reshaping Commerce Media. This Q&A with experts from our Retail & Commerce Media Committee explores how it is already delivering measurable value - from real‑time intent understanding and dynamic bidding to predictive audience building and multi‑agent workflow automation. Retailers and advertisers are beginning to overcome long‑standing challenges around data fragmentation and operational complexity, with AI enabling smarter planning, faster execution, and more relevant customer experiences. The shift from static segments to real‑time signals, alongside the rise of conversational interfaces and synthetic audiences, is setting a new standard for efficiency and performance.

Looking ahead, the next 12 months will bring even more transformative change. Advances such as multi‑agent orchestration, natural language “talk to your data” platforms, Bayesian measurement, and interoperability standards like MCP promise to eliminate operational bloat and unlock new strategic capabilities. Yet our experts stress that sustainable growth depends on strong governance, clean data, and user trust. As AI accelerates creative production, optimisation, and measurement, the industry must balance automation with oversight to ensure AI enhances - rather than overwhelms - the teams and brands that rely on it.

A big thank you to the following contributors for sharing their thoughts:

Diana Abebrese, Global Retail Media Lead, Empathy Lab by EPAM

James Taylor, Founder & CEO, Particular Audience

Q1: How do you see AI currently being used across commerce media today, and which use cases (e.g., personalisation, targeting, real-time bidding) are already proving most effective?

James: "Right now, AI in commerce media is driving what I call the Tripartite Win: it delivers efficiency and ROAS for advertisers, drives yield and monetisation for retailers, and most importantly, ensures absolute relevance and a better experience for the customer. But to understand its effectiveness, we have to stop grouping all AI into one bucket and define the spectrum: Statistical, Machine Learning (ML), Predictive, and Generative.

Where we are seeing the most immediate and effective value today is in the transition from static segments to real-time signals.

• Real-Time Semantic Intent: I heavily argue for prioritising onsite intent capture over broad omnichannel assumptions. By using ML and Predictive AI to understand semantic, in-the-moment shopper intent, we can move beyond batch-processed audience segments.
• Dynamic Bidding & Personalisation: We are seeing practical applications of this where human intelligence guides AI. Tools like algorithm cascades, gating (filters, boost/bury), and business rules allow operators to shape the AI's output, feeding much more effective real-time bidding engines and recommendation models."

Diana: "This is a pivotal moment in commerce media - retailers and advertisers can see how AI can resolve the challenges they have been facing for many years around operational complexity, data and technical fragmentation and lack of standardisation. AI promises to deliver outcomes that are smarter, faster and easier, and we are seeing early adoption and delivery of real business value in 3 key areas:

• omni-channel media planning
• workflow orchestration enabled by agentic workflows and conversational interfaces
• transforming customer data into tangible applications such as predictive audience building and synthetic audiences.

Q. What are the key opportunities for AI to optimise workflows within commerce media - from campaign planning to reporting - and what impact will this have on team efficiency over the next year?

Diana: "Orchestration is one of those overused words in retail media. What we really mean is how we can reduce the number of processes and platforms involved in a typical commerce media campaign lifecycle, and how we can standardise both data inputs and data outputs in order to reduce cognitive workload and duplication of efforts. - cross platform; cross-channel; cross-retailer; cross-team. Optimisation opportunities to streamline and accelerate processes come with multi-agentic workflows powered by LLMs and API connectors. Pollen by Sainsbury's is a great example of how this can work well. Getting this right requires future-facing planning, data hygiene and process optimisation. Not to mention user-friendly and intuitive user interfaces!"

James: "The biggest opportunity over the next year is eradicating operational bloat. AdOps has historically been weighed down by manual campaign setup and fragmented platforms, but we are entering an era of true, agentic orchestration.

The massive catalyst here is MCP (Model Context Protocol). For the audience less familiar, MCP is an open standard that securely connects generative AI models to local data sources. It effectively shortcuts what used to take months of dedicated, brittle API integrations into a process that takes mere minutes.

• The Interoperability Explosion: MCP unlocks a new level of system-to-system communication.
• Automated Ad Set Builders: We are moving toward interfaces where fewer clicks achieve the same or better results with AI-driven targeting (signal based), and conversational access where agentic workflows handle the heavy lifting of campaign planning, setup, and pacing.

Functional Ad Units: Technologies like Retail-MCP are enabling the creation of dynamic, functional apps, and we believe functional ad units will follow, on the fly, for retailers and their brands. This shift will drastically reduce the cognitive load on AdOps teams, freeing them from manual data entry to focus on high-level strategy."

Q. Looking ahead to the next 12 months, which emerging AI technologies or trends (e.g., generative models, agentic AI assistants, automated optimisation) do you expect to transform commerce media strategies the most?

Diana: "I expect to see the industry embracing AI in order to make things faster, smarter and easier. Multi-agentic solutions, natural language models and generative AI will start to power commerce media capabilities cross-retailer and cross-channel. We will see a huge rise in Talk to your Data platforms - informing media planning decision-making, creative processes, real-time optimisation strategies and measurement and insight capabilities - satisfying many of the demands coming from the market At Empathy Lab we are seeing growing demand for our Synthetic Audience solutions - creating “customer” panels for testing of new products, promotions and creative concepts at scale, with Mars reporting over 75% accuracy of results, when compared with scores from human panels."

James: "Measurement and predictive modelling are about to undergo a radical transformation. As an industry, retail faces a massive multi-objective optimisation problem. "Incrementality" is the buzzword, but we have to ask: incrementality of what?

Over the next 12 months, the most transformative trends will be:

• A Continued Shift to Bayesian Measurement: Given the wide variation in data density and noise across different retail metrics, traditional p-values for statistical significance are not appropriate and have always been insufficient. We need a necessary shift toward Bayesian measurement to prove true incrementality across disparate goals.
• Synthetic Audiences: We will see a surge in the use of AI-generated customer panels (like the incredible work Diana and Empathy Lab is doing, hitting 75% predictive accuracy for brands like Mars). This allows us to model and predict incrementality before a brand even spends a dollar.
• The Human Bottleneck in Gen AI: Generative AI for creative is unequivocally "here." However, it introduces a new trend: solving the oversight limiter. AI can generate hyper-personalised creative at an infinite scale, but humans physically cannot approve it at that same speed. The next year will be defined by how we solve for automated oversight without strangling scale."

Q. What are the biggest challenges or risks associated with AI in commerce media (e.g., data quality, transparency, measurement, ethical concerns), and how should the industry address these to unlock sustainable growth?

James: "When it comes to unleashing AI, my favourite word internally at Particular Audience is Governance. The fear of losing control over generative creative at hyper-scale is a massive barrier to AI adoption, which is exactly why we haven't all just handed the keys over to an autonomous agent to run our businesses.

To unlock sustainable growth, the industry must address governance head-on by splitting our approaches to Predictive versus Generative AI:

• Predictive Guardrails: Ensure that models are optimising for the right incrementality metrics across the retail org, and delivering incrementality based on advertiser goals, not just gaming attribution windows.
• Generative Guardrails: Establish strict parameters around brand safety, tone, and visual identity so that AI-generated functional ad units don't go rogue at a scale humans can't manually monitor.
• Product Data & Software Readiness: The absolute prerequisite for any of this is clean data and SaaS style gated control interfaces. 

Ultimately, governance ensures we are using AI to solve actual problems, rather than just rapidly automating our mistakes.

Diana: "At a micro level,  we see a high level of cynicism within companies. Whilst users recognise personal benefits when using AI at work and at home, they need to feel empowered and in control of their interactions with AI business tools before fully embracing them. User research and consultation to define use cases alongside a measured roll-out process and feedback loop will be instrumental in overcoming these challenges."