Who Keeps the Customer When the Agent Does the Shopping?
Agent-first shopping tripled in a year. As shoppers hand more of the buying journey to AI agents, the Shopper Context Protocol is the industry's answer to keeping the relationship, but nine months in, adoption is still unmeasured.
Let’s start with some stats:
- 15% of shoppers now start their buying journey with an LLM or an AI agent, up from 5% a year ago, according to Salesforce's latest State of Commerce data
- According to Salesforce’s Holiday 2026 forecast, about 1 in 5 ecommerce visits this season will come from a crawler or an agent, not a person
- And roughly 1/3 of B2C brands already have some kind of agent deployed, and more than half plan to have one live within six months
We are witnessing a clear shift in traffic, but what happens to the relationship underneath it—the loyalty status, the order history, the things a retailer knows about a customer that used to travel with every visit—is becoming less clear. It’s the part of commerce not solved by any of the protocols currently racing to define agentic shopping as a whole. It's the part a newer, less-discussed standard is trying to fix. And it's worth asking, nine months in, whether it's working.
The gap in what "agentic commerce" actually covers
Most of the infrastructure built for AI shopping so far solves for transaction, not recognition. OpenAI and Stripe's Agentic Commerce Protocol (ACP), which underpins ChatGPT's shopping experience, defines how an agent completes a purchase. It doesn't define who the shopper is beyond that single session. An agent working through ACP can find a product and check out, but it has no native way to know that the person on the other end has bought from that brand for eight years, or that they're a loyalty-tier customer who never pays full price for anything.
That gap isn’t staying empty, though. OpenAI's ChatGPT keeps a running record of a shopper's payment methods and order history across every merchant they buy from inside the app— a cross-merchant memory that lives with OpenAI, not with any individual retailer. However that memory gets used, its existence marks a real shift in whose systems hold the customer relationship.
The clearest evidence this is already playing out
Walmart was one of the first major retailers into ChatGPT's Instant Checkout, the feature OpenAI launched last September that allowed shoppers to complete a purchase without leaving the chat window. But then, in March of this year, OpenAI walked it back. "We've found that the initial version of Instant Checkout did not offer the level of flexibility that we aspire to provide, so we're allowing merchants to use their own checkout experiences while we focus our efforts on product discovery," the company wrote, shifting the deeper, transactional experiences into ChatGPT "apps" that individual merchants control.
Walmart's response is the interesting part. Rather than stepping back from ChatGPT, it built forward: a full in-ChatGPT app experience powered by Sparky, its own AI agent, that carries Walmart account linking, loyalty status, and Walmart-processed payments into the conversation. "Shopping can start anywhere now, whether that's Walmart or a question in ChatGPT," said Daniel Danker, Walmart's EVP of AI acceleration, product and design. "Our approach is simple: wherever it begins, customers still get the same personalized Walmart experience, including our assortment, value and speed. With Sparky, we're bringing our trusted experience into more places."
That's a retailer explicitly refusing to let a third-party agent become the system of record for its own customers, while still showing up wherever the shopper starts. Target, Sephora, Nordstrom, Lowe's, Best Buy, Home Depot, and Wayfair have all integrated with the retooled ACP for product discovery since, a sign that the "be present in the agent, but keep the relationship" posture is becoming the default rather than the exception.
The industry's answer to the gap: the Shopper Context Protocol
Not every retailer has Walmart's leverage to build its own branded agent. That's the problem the Shopper Context Protocol (SCP) is built for. First published as an open spec in October 2025—backed early by the marketing platform Cordial—and organized since December under a Retail AI Council working group chaired by Matt Howland, SCP is designed to sit between discovery and transaction. It carries the parts ACP leaves out: loyalty tier, preferences, purchase history, portable across whatever agent or platform the shopper happens to be using.
Mechanically, it's built to be easy to adopt: a retailer publishes SCP support the way it would publish an MX record for email, and an agent queries that shopper context through OAuth-style permissions with the customer's consent. The pitch is 2-4 days to implement, no licensing fee.
Cordial's own research, published alongside the spec, found the problem it's solving is real: 40% of marketers said they'd lost revenue from misreading a customer's intent, and nearly half of consumers said they lost trust in a brand that got it wrong. Almost half of the marketers surveyed—47%—said they had no strategy at all for how their brand should show up inside a personal AI agent like ChatGPT or Perplexity.
What we don't know yet
Nine months after launch, there's no public number for how many retailers have actually implemented SCP. The named participants so far are a marketing-technology vendor (Cordial) and a list of Retail AI Council members—global brands like Adidas, Tesla, Amazon, and Apple among them—with no confirmation that any of them have shipped SCP support specifically. Compare that to UCP, where recent research shows only 22.5% of top retailers have a live UCP endpoint. SCP, the standard built specifically to protect the retailer's side of the relationship, doesn't yet have that kind of visibility into its own uptake.
That's not necessarily a knock on the protocol. Standards bodies are always ahead of implementation, and a 2–4 day integration is a low enough bar that adoption could show up quickly once a few large retailers commit publicly. But it's worth naming plainly rather than assuming momentum that hasn't been measured yet.
The part that doesn't show up in any protocol spec
Whichever standard wins, none of it matters if a retailer's underlying product, loyalty, and order data isn't structured well enough to expose in the first place. Speaking about agentic systems, Forrester analysts Joe Cicman and Chuck Gahun made this point saying that composability remains valuable precisely because "agentic systems depend on the APIs, integrations, and services that composable architectures expose." An agent—anyone's agent—can only carry context that a retailer's systems can actually produce.
That's the piece most brands are actually behind on. It's not that they haven't picked a side in the ACP-versus-UCP-versus-SCP question (they’ll likely need to be able to leverage all of these and more). It's that most don't have clean, structured, real-time customer and product data to plug into any of them. Cordial found that 64% of marketers said their customer data was used only for campaign planning, not accessed in real time, which means even a retailer that adopts SCP tomorrow may not have anything current to publish through it.
None of this is waiting on the standards to mature. Agent-first shopping tripled in a year, and Salesforce expects a fifth of holiday traffic to come from something other than a person. SCP is too new to judge by results, and so is everything else in this space. What isn't too new to judge is the prerequisite underneath all of it: a retailer's product, loyalty, and order data has to be structured and current before any protocol—SCP, UCP, or whatever comes next—has anything real to carry. That's the work due now, regardless of which standard wins.
Leigh Bryant
Editorial Director, Composable.com
Leigh Bryant is a seasoned content and brand strategist with over a decade of experience in digital storytelling. Starting in retail before shifting to the technology space, she has spent the past ten years crafting compelling narratives as a writer, editor, and strategist.