AI will play a key part in the future retail media landscape, says a new guide from Particular Audience – but it warns LLMs will take control if retailers don’t

AI will inevitably reshape the commerce media supply chain, with the potential of enormous benefits for advertisers, retailers and consumers. But retailers have only a brief window to establish ownership or influence over the distinct layers of AI architecture – or “risk becoming inventory in someone else’s auction”.

According to a new industry guide, released today by AI-powered retail media personalisation platform Particular Audience, the benefits of AI are close enough to touch, as the industry enters an agentic shift in which agents will orchestrate workflows, generate creative and assist operational decisions.

“For advertisers, AI offers improved efficiency and validated ROAS through predictive targeting and automated campaign management,” says Particular Audience CEO and founder James Taylor. “For retailers, the prize is higher-margin yield and monetisation potential, while customers benefit from greater relevance and utility.”

Marketing Technology News: MarTech Interview with Theresa Pham, Head of Product @ Wayvia

The report, Retail Media AI Architecture: From Prediction to Decisioning, details AI use cases at different stages of maturity, including using predictive models to capture in-session semantic intent; approaches to enable agents to connect to data sources; creating functional ad units within LLM chat; and generating synthetic audiences to pre-evaluate creative and messaging.

But the key question of who ultimately controls how retailers and their brands surface within AI experiences is something that retailers need to actively build for, says Taylor, who warns that retailers urgently need to banish persistent misconceptions about the way AI works in practice.

“As LLM providers gain influence over how products are surfaced and ranked, the strategic question is: who controls the decisioning layer in commerce?” says Taylor. “Retailers that own or meaningfully influence this layer retain commercial control of their monetisation surface. Those that cede it – by exposing catalogues directly to third-party LLMs without an intervening decisioning layer – risk becoming inventory in someone else’s auction.”

The architectural choices made in the next two years will determine which side of that line each retailer sits on, suggests the report, which notes that a lack of understanding of the five distinct disciplines in play is the principal pain point in current vendor and industry discourse.

Retail Media AI Architecture: From Prediction to Decisioning defines the five layers as:

• Causal Measurement – the only valid accountability layer in retail media, capable of proving whether a campaign caused a sale. Not AI, in fact, but applied statistics.
• Predictive Models – machine learning models that predict contextual relevance and drive outcomes.
• Decisioning Systems – ranking and optimisation engines that join model probabilities with business constraints (such as available inventory, margin and others) and sponsored bid modifiers to produce a final ranked output. This is where monetisation happens in this future state for commerce.
• Generative Interfaces – rich experiences within LLM and associated AI interfaces through MCP-enabled tool-calling that maximise the opportunity behind natural-language intent signals.
• Agentic Orchestration – multi-step workflow execution, in which an agent uses tools such as APIs, MCP servers and decisioning systems to complete tasks such as ad-set construction or pacing rebalancing

“AI has the potential to reshape the retail media supply chain, but only if the industry stops conflating layers,” says Taylor. “MCP is a protocol, not a decisioning engine. Raw catalogue exposure to an LLM functions, but is commercially blind and presents gaps in practice. Vector semantic search is a real and valuable improvement, but cannot enforce inventory, margin, or bid constraints without a control centre. A decisioning layer – built or bought – is the architecture required for monetisation, margin and management.”

You can download the whitepaper here:
https://explore.particularaudience.com/whitepapers

Particular Audience is an AI-native retail media and personalisation platform that unifies proprietary large language powered search, hyper-personalisation, and sponsored product ad decisioning into a single ranking system.

Founded in 2017, PA helps retailers maximise profit-per-pixel—balancing relevance, conversion, margin, and ad yield in real time. Its multimodal machine-learning platform replaces legacy keyword and rules-based approaches, enabling automated, intent-aware retail media without degrading the shopper experience.

Headquartered in London, Sydney, and Vancouver, Particular Audience supports global enterprise retailers and contributes to open standards such as the Model Context Protocol (MCP), enabling agentic commerce on retailer-controlled infrastructure.

AI/ML that powers all customer interactions. If you see it, search it, click it, and buy it – that’s PA.

More information at www.particularaudience.com. For more information on the Modular Retail Media offering, visit: https://explore.particularaudience.com/