Services need a feed standard

Google has already solved this problem for physical products. The question is whether anyone will do the same for services before it's too late.

What Google figured out for retail

In March 2026, Google announced updates to its Universal Commerce Protocol, an open standard designed to let AI agents interact directly with retailer catalogues. The UCP lets agents retrieve real-time pricing and inventory, add items to a cart, and carry loyalty benefits across platforms.

The mechanism is straightforward. Retailers expose structured data. AI systems consume it. Transactions happen without the user ever visiting a product page.

That shift is already underway for physical goods. For services, it hasn't started. There is no equivalent standard. No structured interface. No way for an AI agent to query an insurance policy, a mortgage product, or an energy tariff and return a meaningful result.

That gap will not stay open for long.

Why services are harder

A product feed is a catalogue. It describes what exists: title, price, availability, image. Static data, refreshed periodically.

A services feed can't work the same way. Services aren't static. Their outputs depend entirely on who is asking and in what context. A car insurance premium for a 28-year-old in London driving a Tesla is a completely different product from the same policy for a 19-year-old in Manchester driving a modified hatchback. The policy name is identical. The price, eligibility, and terms are not.

This is why no one has built the standard yet. It looks like a data problem. It isn't. It's a decision problem.

What a services feed actually needs to expose is a logic interface, not a catalogue. Something that accepts structured inputs, applies eligibility and pricing rules, and returns a structured output. A price, a coverage set, an excess figure, a recommendation. Given this person, in this context, with this asset: here is the result.

That structure works across almost every service category. Insurance, mortgages, utilities, broadband, financial products. The inputs change. The architecture does not.

The problem with letting the platforms build it

Google, ChatGPT, and Cloudflare will all reach for this eventually. Google's UCP expansion makes the direction obvious. Once agentic commerce is normalised for products, services are the next surface.

But there is a structural problem with letting any of those three define the standard.

We have seen this pattern before. Platforms take the data, build the interface, and capture the value. Once the interface is established, the only way to exist within it is to pay for the privilege. The brands that fed the system end up renting visibility inside it.

A services feed built by Google would be a Google product. The standard would serve Google's commercial interests. Providers would hand over their pricing logic, their eligibility rules, their decision architecture, and Google would sit between them and their customers permanently.

The standard needs to be built by someone with a different set of incentives.

Who could actually do this

Three candidates come to mind. Each has something the others don't.

MoneySupermarket has existing relationships with every major UK insurance and financial services provider. It already aggregates service outputs at scale. It understands the data structures. It has the trust of both providers and consumers. The gap is technical ambition. MoneySupermarket is a comparison engine. Defining an open decision standard is a different kind of project. But it is the organisation with the most to gain from owning this layer before someone else does.

Profound is building in exactly the right space: AI visibility for brands and services. It understands how AI systems consume and surface information. It could define the standard from the consumption side, specifying what AI agents need to receive rather than what providers want to publish. The risk is that Profound lacks the provider relationships to achieve adoption at scale.

An agency network has the client relationships across the whole service economy. A network like WPP or Publicis touches the marketing and digital infrastructure of most major service brands. It could convene the right people and frame this as an industry initiative rather than a proprietary platform. The risk is execution. Agency networks are not known for building technical standards.

The most realistic answer is probably a coalition: a comparison platform with the distribution, an AI-native firm with the technical framing, and a body like the IAB or an industry trade group to give it legitimacy.

What the standard would actually look like

The architecture is not complicated. Three layers.

Inputs define the situation. User attributes, asset details, context variables. Standardised fields that any platform can pass in.

Rules define the logic. Eligibility conditions, pricing calculations, feature inclusions. This layer can be either declarative, expressed as structured if/then logic, or API-based, where the provider's pricing engine operates as a black box and returns outputs against defined inputs.

Outputs define what the AI agent receives. A structured response: price, terms, coverage, validity window. Standardised enough to be compared across providers. Specific enough to be actionable.

Three extensions make the model complete. A time dimension handles quote expiry and dynamic pricing. A confidence layer handles probabilistic outputs and indicative pricing. A state field tracks where the interaction sits in an underwriting or application journey.

That is the whole architecture. The complexity is not technical. It is commercial and political.

Why this matters now

AI agents are already simulating product outcomes before the user clicks. That is what UCP is built for. Services are next.

When that happens, the brands that have invested in structured, machine-readable decision logic will surface. The brands that haven't will not. Visibility in AI-mediated discovery will depend on whether your service can accept a structured query and return a structured answer.

The window to define how that works, and who controls the standard, is open. It will not stay open. Google will move, or ChatGPT will, or Cloudflare will.

The question is whether anyone in the service economy moves first.