The Next Big Problem in AI Shopping Isn't Acquisition. It's Retention.
For most of ecommerce history, brands have spent enormous amounts of time and money trying to get people onto their websites. We optimized landing pages, tested product detail pages, and removed friction from checkout. After someone purchased, CRM and lifecycle marketing took over. We welcomed them, explained the product, encouraged a second order, and tried to build a relationship that lasted beyond the first transaction.
Now imagine a customer who never visits your storefront. They ask an AI assistant to find a product, compare the alternatives, and help them check out. The brand gets an order, but the customer may have missed the education, membership invitation, and other touchpoints that traditionally introduce the relationship.
That is the part of agentic commerce I think retention leaders should be talking about. The industry is focused on how AI can deliver the first transaction. I'm more interested in what happens when the customer is ready to buy again.
What is agentic commerce, and why does it matter now?
Agentic commerce describes shopping journeys where AI assistants help people find products, evaluate alternatives, manage carts, and sometimes complete purchases. The degree of autonomy and the merchant's role vary by platform and integration. It is not a single checkout experience or a promise that every customer will suddenly abandon retailer websites.
In January 2026, Google and Shopify announced the Universal Commerce Protocol (UCP), an open standard designed to connect AI shopping experiences with existing commerce systems. Google later expanded UCP's capabilities for carts, current product details, and identity linking for loyalty benefits. Shopify opened its agentic commerce tooling to developers and made participating merchants discoverable across AI channels.
The underlying changes are documented in Google's technical introduction to UCP and Shopify's spring developer update.
Shopify's June 2026 developer announcement describes access to tools for product search through checkout. The availability of this infrastructure is meaningful, but it does not establish that agent-assisted checkout has become the dominant purchasing behavior.
The 2026 holiday season makes the question harder to ignore
This is no longer just a futuristic prediction. In October 2026, Shopify described the coming holiday season as a milestone for AI-assisted shopping. Its Global Holiday Retail Report found that 65% of shoppers said they intended to use AI for at least one holiday shopping task. That is an intention measure, not proof of checkout adoption, and using AI for gift ideas is different from completing an order through an agent.
Still, the timing deserves attention. Shopify's October 2026 holiday analysis shows why this discussion is becoming urgent for merchants heading into their busiest acquisition period.
There is a major difference between discovering a brand through an assistant and becoming a repeat customer of that brand. If we don't distinguish those outcomes, we risk celebrating a new source of first orders without understanding the longer-term economics.
Making a sale is not the same as acquiring a loyal customer
Ecommerce organizations sometimes treat customer acquisition and order acquisition as interchangeable. They aren't. An agent-assisted transaction can generate revenue while creating fewer opportunities for a customer to learn about the brand, opt into marketing, or understand its subscriptions and loyalty program.
None of that means the relationship is lost. Shopify says merchants remain the merchant of record on supported experiences, and customer data can still flow into the commerce system. But having an order in the database and earning future customer preference are two different achievements. The specifics of consent, identity, and communications need to be checked for each integration.
Can the merchant recognize the customer when they return? Can it provide useful post-purchase education? Was promotional marketing permission actually granted? Will the second purchase happen directly, or will the customer ask their AI assistant to find the best option again?
These are the questions I'd put on the table before calling a new AI shopping channel a retention success.
The second purchase is where agentic commerce becomes a retention problem
Consider a fictional premium coffee brand. A shopper asks an assistant to find a medium-roast coffee under $25, compares the recommendations, and buys a 12-ounce bag from a brand they've never heard of. The order arrives, the coffee is good, and the initial purchase appears to be a win.
Three weeks later, the shopper needs more coffee. Do they search for that same brand by name? Visit its website? Enroll in a subscription? Or ask the assistant to find the best medium-roast coffee under $25 again?
If the customer asks the assistant to start fresh, the brand may have to compete for someone it already paid to acquire. That isn't evidence that AI-shopping customers are less loyal. We do not have the mature cohort data to claim that. It is a concrete risk worth measuring.
I've spent much of my career working on CRM, customer acquisition, and subscription retention. The first order tells you much less about long-term value than many acquisition reports suggest. Some channels bring customers who develop a habit and come back. Others bring people who responded to an offer but never form a lasting preference.
The same distinction will matter here. The important question is not just whether an AI assistant sends orders. It's whether those orders develop into profitable, ongoing customer relationships.
What changes for lifecycle marketing and CRM?
A retention program depends on more than receiving an email address. It needs accurate customer identity, order history, product expectations, marketing permission, service information, and the ability to recognize the customer across transactions.
Agentic shopping introduces an additional environment where product information is discovered and transactions begin. Brands need to know how that environment connects back to the systems responsible for customer communications and support.
For subscription merchants, this matters especially. A customer who buys through an AI experience still needs to know when their order ships, what happens next, how to get support, and whether a recurring purchase was explicitly authorized. The underlying records must match what the assistant, merchant, and customer see.
This is where the lifecycle marketing technology stack matters. Better copy cannot correct a broken identity or transaction record.
Loyalty and identity linking may become major differentiators
Google's UCP updates include identity-linking capabilities that can allow participating shoppers to access loyalty benefits. That is a promising technical direction, not a guarantee that every brand's loyalty program works across every assistant.
Google's March UCP update discusses identity linking alongside carts and product detail access.
The strategic question is straightforward: if customers have already chosen to build a relationship with a brand, can that relationship carry value into their next shopping interaction? Brands should evaluate whether a customer can be recognized, whether earned benefits are applied properly, and whether consent and privacy expectations remain intact.
A loyalty program that provides value only when customers remember to visit a particular website may need to evolve. But I would prioritize dependable customer records and clear benefits before building another layer of loyalty features.
The attribution problem: AI shopping may look better than it actually is
Imagine someone who first hears about a product in a social advertisement, asks an AI assistant for comparisons days later, checks out through an agent-assisted experience, and reorders directly from the merchant a month afterward.
Which channel acquired that customer? The answer depends on what the business is measuring. The checkout platform may accurately report the immediate referral while missing the earlier exposure. First-order revenue is measurable, but it does not capture every source of influence or the entire customer relationship.
I would evaluate AI-originated orders by new-versus-returning status, contribution margin, customer identity match rate, marketing permission rate, support friction, and 30-, 60-, and 90-day repurchase behavior. Cohorts should be compared carefully, because different channels attract different buying intentions.
I've written about the larger issue of moving from campaign automation to customer decisioning. The same principle applies here: optimize for the decisions and outcomes behind the transaction, not only for a channel's reported conversions.
Six changes I would make before scaling agentic commerce
1. Map customer identity, consent, and transaction ownership
For each integration, document which data is received, which identifiers connect to existing records, how consent is captured, and who handles support and refunds. A successful order is not automatic permission to enroll someone in promotional marketing.
2. Audit product and subscription information
Product attributes, prices, inventory, shipping promises, discount restrictions, and renewal terms need to be accurate in every surface that may supply answers to AI shopping tools. Product feeds and commerce records should have clear owners and QA procedures.
3. Rebuild post-purchase entry criteria around actual events
Customers from agent-assisted transactions should receive the confirmations and service messages they need. Do not assume those orders entered a standard web signup journey, and do not mistake transactional contact details for broad promotional consent.
4. Create a dedicated agentic-commerce cohort view
Track order source, first-order margin, identity resolution, repeat purchase, support contacts, returns, and subscription adoption where applicable. Use consistent observation windows and avoid declaring early cohort differences causal.
5. Test the path to the second order
Pay attention to reordering, loyalty enrollment, product education, and how customers seek assistance after delivery. Where practical, test whether clearer product guidance or post-purchase experiences lead to better customer outcomes.
6. Give retention a seat at the integration meeting
Agentic commerce affects acquisition, engineering, product, legal, customer service, and CRM. If the integration changes which customer events exist or which permissions are available, the lifecycle team needs to help define requirements before launch.
What I would put on the agentic-commerce dashboard
My starting scorecard would include first-order contribution margin, percentage of orders from new customers, customer identity match rate, explicit marketing permission rate, 30/60/90-day repeat purchase, second-order channel, refund and support rates, and repeat-customer contribution margin.
These metrics won't all be available on day one. That is not a reason to ignore them. It is a reason to agree on the measurement plan before assigning ambitious revenue targets to a channel whose customer behavior is still emerging.
Agentic commerce FAQ
What is agentic commerce?
It refers to shopping journeys in which AI agents help customers discover products, compare options, manage carts, or buy products, depending on the platform and available merchant integrations.
Does agentic commerce replace ecommerce websites?
Not necessarily. Websites still support product education, customer accounts, direct shopping, loyalty, and service. Agent-assisted commerce adds a buying route rather than demonstrating that every other route will disappear.
Do merchants still own their customer relationships?
In supported integrations, merchants can remain the seller of record. But identity matching, marketing permission, customer records, and communications should be validated for each implementation rather than assumed.
Why does agentic commerce matter for CRM?
It can change how customers are introduced to the brand, which data enters the customer record, and how the business measures returning customers. CRM teams should design the post-purchase journey and reporting alongside the commerce integration.
What should retention leaders do first?
Map one complete agent-assisted purchase: discovery, payment, confirmation, delivery, service, and repurchase. Identify where data or customer experience breaks down, and make those issues the initial requirements.
The first transaction is only the beginning
AI shopping could make discovery and checkout easier for both customers and merchants. I'm optimistic about that opportunity. But for businesses that depend on repeat purchases, the question cannot stop at whether the technology delivered an order.
A customer acquired through an assistant still needs a good product, reliable information, meaningful service, and a reason to return. The buying interface may change quickly, but earning the second purchase remains the work that determines whether acquisition becomes lasting growth.
About the author
John Brown is a director-level CRM and retention leader with experience across direct-to-consumer ecommerce, subscriptions, customer acquisition, and lifecycle marketing. He writes about marketing technology, customer experience, and measurable long-term growth.
More on CRM and retention: The Future of CRM: From Campaign Automation to Customer Decisioning.
Comments