Stop Asking
Which Protocol.
Start Asking Which Layer.

If you've sat through a single strategy meeting on agentic commerce in the last six months, you've heard some version of this question: Should we prioritise UCP or ACP? And if you've heard it enough times, you may have noticed something: nobody giving confident answers to it is actually deploying anything at scale.

The framing is broken. UCP, ACP, and MCP are not competitors. They are layers — distinct parts of a single infrastructure stack, each solving a different problem at a different altitude. Choosing between them is like asking whether a building needs a foundation, walls, or a roof. The answer, obviously, is all three. In the right order. For the right reasons.

What follows is the strategic framework that cuts through the noise — grounded in what is actually live in production today, what has already failed at scale, and what the early movers are doing differently.

The Stack, Not the Protocol

Think of agentic commerce infrastructure as a three-floor building. Each floor serves a distinct function. None of them works without the others. And critically, you cannot skip from the ground floor to the top without the middle holding your weight.

The dependency flows upward. ACP and UCP can only transact what MCP can surface. MCP can only surface what your underlying product data actually contains. The protocols are not the bottleneck — your data is.

The Reality Check Nobody Is Talking About

Before committing engineering resources to protocol integrations, every e-commerce leader needs to absorb one data point that has been quietly circulating since March 2026.

Walmart — arguably the most sophisticated early mover in agentic commerce — disclosed that ChatGPT's Instant Checkout, ACP's flagship implementation, converted at three times worse than walmart.com. Three times worse. Walmart pulled the plug and pivoted to embedding its own Sparky AI assistant inside ChatGPT instead.

This is not an argument against protocols. It is an argument against treating the protocol as the strategy. Protocols are rails. What runs on those rails — your product data quality, your brand presence in AI surfaces, your owned agent experience — is what actually converts.

What Each Protocol Is Actually For

MCP: Start Here, Always

MCP is not a commerce protocol. It is a data access protocol that makes commerce protocols possible. Without clean, structured, machine-readable product data flowing through MCP-compatible endpoints, neither UCP nor ACP can do meaningful work on your behalf.

The practical implication: before configuring a single agentic commerce integration, audit your product catalog. Complete schema markup, accurate pricing and availability, detailed product attributes, return policies in formats agents can parse, fulfillment timelines — all of it needs to exist in a form an AI can reason about. A single shopping query from an AI agent can trigger 10 to 15 rapid-fire API requests as it explores variants, shipping options, and compatibility. If your data infrastructure cannot respond reliably at that volume, the agent moves on.

UCP: The Wider Surface, the Longer Game

UCP is the more architecturally ambitious of the two commerce protocols — and the one with the stronger long-term position. Its modular design means retailers declare which capabilities they support, and any UCP-compliant agent discovers them dynamically. Build once; distribute everywhere the standard is adopted.

For merchants already on Google Merchant Center, UCP adoption is largely an extension of existing infrastructure. The March 2026 update added multi-item cart support, real-time catalog queries, and loyalty program integration. Dual-protocol merchants running both UCP and ACP see up to 40% more agentic traffic than single-protocol stores.

ACP: The ChatGPT Ecosystem, Nothing More

ACP is narrower in scope than its launch positioning suggested. It is, today, primarily an OpenAI and ChatGPT protocol. Its managed approach reduces integration complexity but extracts control in exchange — OpenAI decides what gets surfaced, Stripe is the only payment option, and the combined fee structure runs approximately double what UCP costs. For brands where ChatGPT is a meaningful discovery channel, ACP is worth the tradeoff. If you are already on Stripe, activation can require as little as one line of code.

Strategy by Platform

The 5-Step Implementation Sequence

The Protocol Comparison at a Glance

The Moat That Actually Matters

There is a version of this story that ends with: implement the protocols, get the traffic, win. That version is incomplete.

AI agents optimise for utility. They prioritise price, delivery speed, ratings, and inventory availability. They do not optimise for brand affinity, loyalty history, or the customer relationship you have spent years building. Every purchase completed through a third-party agent without touching your owned experience is a purchase where your loyalty economics, upsell logic, and customer data disappear into someone else's platform.

The organisations that will compound their advantage in agentic commerce are not those with the most protocol integrations. They are those that combine protocol readiness with owned agent experiences that preserve brand value, customer relationships, and data — the things that make a second purchase more likely than the first.

The window for early-mover advantage in this infrastructure cycle is not infinite. The protocols are consolidating. The consumer habits are forming. The organisations establishing protocol presence and owned agent experiences now will capture traffic, data, and customer relationships before mass adoption forces everyone in at once — at which point their compounding advantages in measurement, optimisation, and brand-agent personalisation will be difficult for latecomers to replicate.

The question for every e-commerce executive is not which protocol to choose. It is whether you are building a deliberate agentic strategy — or waiting to react to one built by platforms that do not share your interests.