Agentic Commerce in 2027: Will Your Brand Make Your Customer’s AI Shortlist?

Imagine a shopper in early 2027 asking an AI assistant for a rain jacket under $150 that will arrive by Friday. The assistant compares jackets from a dozen retailers, shows three side by side and waits for approval. The shopper chooses one without even opening a brand’s website.

If your jacket isn’t among those three, it never gets considered in that comparison. That, in a nutshell, is the challenge agentic commerce presents to marketers: your product may need to satisfy the shopper’s agent before the shopper ever sees it.

Shoppers are already open to handing over the purchase. In SAP’s Global Customer Loyalty Index 2026, a survey of 10,002 consumers in the US, UK, Germany and Hong Kong, 33% of consumers say they trust AI assistants to complete purchases on their behalf.

So, what is agentic commerce? It’s shopping in which an AI agent acts for a person, within limits that person sets. The agent can search for products, compare them and fill a cart. If the shopper allows it, it can also pay.

The difference between an AI that helps you shop and one that shops for you comes down to permission. Asking an assistant to compare three rain jackets is asking for advice. Telling it to buy the blue one if the price drops below $150 is handing over a decision. Shoppers can do either, purchase by purchase. Someone might happily let an agent reorder dog food every month but insist on approving a new sofa themselves.

For marketers, that adds a new kind of customer between your brand and the person who buys from you. The agent reads your product information, checks it against what the shopper asked for and decides whether you make the cut. It doesn’t see your brand campaign or feel the warmth of your homepage. It sees what you’ve told it about price, stock, delivery and returns.

AI shopping assistants can make shopping feel like a single conversation. Behind that conversation, five things happen.

1. The shopper says what they want

It starts the way a conversation with a good sales assistant does. Maya has a hiking trip next weekend and no rain jacket. She tells her AI assistant what she needs: waterproof, under $150, in a medium, at her door by Friday, and ideally from the outdoor store where she bought her boots last spring.

Many shoppers would let the agent go a step further and buy, especially younger ones: among Gen Z, 44% would trust an AI assistant to purchase an item for them. Whether she approves the final pick or lets the agent buy, Maya has set the limits herself, and she expects the agent to stay inside them.

2. The agent goes looking

Maya’s agent doesn’t wander the aisles or scroll category pages the way she would. It reads what retailers publish about their products on their websites, in schema markup and in product feeds: descriptions, sizes, prices, stock levels and delivery times. AI companies, retailers and payment providers are building shared standards so agents can pull that information straight from merchants, and the retailers who share it clearly are the ones the agent can actually consider.

If the outdoor store lists its jackets with a price but no delivery date, the agent can’t tell whether they’ll arrive by Friday. It moves on to a store that gives clear delivery dates. Consequently, Maya never finds out her favorite store was in the running.

3. The agent compares and shortlists

Now the agent weighs up what it found. One jacket is $139 but won’t arrive until Monday. Another arrives Thursday, costs $145 and comes with free returns. A third, from a store she’s never shopped at, only comes in a large. The agent narrows a dozen options down to three and lines them up side by side, with the reasons each one made the list.

This is where a brand wins or loses the sale. The agent compares what it can read and calculate: price, delivery, returns, stock and any offer it can put a value on.

4. Maya approves, and the agent pays

Maya picks the $145 jacket and taps approve. That approval matters more than it looks. It gives the agent a clear record of exactly what Maya agreed to buy and at what price, so what she sees is what she pays for. If she had told the agent to buy on its own, the same record would hold the rules she set in advance: the budget, the deadline, the size.

Her card details stay protected along the way. Payment providers can give the agent a token that works only for that store and that basket total, so it can pay without ever seeing Maya’s card number.

5. The store delivers

The order lands with the retailer like any other. The store takes the payment, packs the jacket, ships it and handles the return if it doesn’t fit. The agent found the store, but the store still has to keep the promise the agent picked it for.

If the jacket arrives Thursday as promised, Maya is more likely to tell her agent to check that store first next time, or to add that store’s connector to her personal agent so she can shop it from there. If it turns up the following Tuesday, she might tell it the opposite. Every delivery feeds into the next shortlist.

A year ago, the story was about AI assistants that could check out inside a chat window. In 2026 it became something bigger: personal agents that connect to a shopper’s email, calendar and the apps they use to shop and travel, and act for them wherever they buy. Through all of it, the store still owns the order.

Meta’s Muse takes personal agents mainstream

Meta’s Muse is the clearest sign of where this is heading. Muse can search retail sites, surface products and prepare a purchase, then show the shopper an “approval card” with the order details to confirm, as CNBC explains.

Meta has announced partnerships with Walmart, Gap, Sephora, Expedia, Wayfair, Best Buy and others, and a Meta spokesperson told CNBC that shopping has quickly become one of Muse’s biggest usage drivers.

In late October it launched in Singapore, connecting to email, calendars and local shopping, dining and travel services. Not every retailer is on board, though: Amazon has blocked Muse from buying on its site, saying it violates its terms of service.

Instinct shows how fast delegation is growing

Muse has plenty of competition. Instinct, a personal agent that people text, call or email, launched its invite-only service in August and raised $1 billion at a $10 billion valuation in September. Its founder, Noah Shinn, said on a podcast that the platform is approaching $1 billion in annual transactions and that more than half of them are travel, though he didn’t say how the annual figure is calculated, TechCrunch reported.

Instinct is one of many startups vying to be everyone’s personal agent.

The Personal Agent Protocol sets the rules of engagement

As agents multiply, businesses need to know who they’re dealing with. In October, Meta and Sierra announced the Personal Agent Protocol, or Poppy, an open standard for how personal agents interact with businesses. The principle is that consumers decide what access to give their agents, and companies set the parameters for what those agents can do.

It’s still in draft, but the direction is set: agents will identify themselves, and brands will decide what they can do.

Walmart gives agents their own profile

Walmart shows what this looks like from the retailer’s side. Agent Profiles, the Walmart Agent Gateway and terms for AI agents give each authorized agent a dedicated profile within the customer’s account, so its purchases show up separately from the customer’s own. Customers can require approval for every purchase or let the agent buy within a spending limit, and switch an agent off at any time.

Talk to marketers about agentic commerce and you hear worry and hope at once: worry that an AI choosing for the customer will undo years of brand building, and hope that an agent reordering your product every month could be the most loyal customer you’ve ever had. 2027 is when many brands will find out which one they’ll see more of, and three things are already in motion:

• Shoppers expect to hand over more of their buying. In a March 2026 survey of 12,005 consumers in six markets, carried out by Censuswide for Checkout.com, a third expected AI to drive at least 10% of their purchases within a year.
• Personal agents are going global and across categories. Meta’s Muse is officially available in the United States, Canada, Japan and Singapore. Expect more personal agents to handle transactions across categories.
• Retailers will set the terms for agents. Walmart’s new terms require customer approval for agent purchases by default and block agents that break the rules. Expect other retailers to follow.

What does this mean? Agents won’t replace customer engagement entirely but they’ll certainly raise the bar for it. Every promise you make, from a delivery date to a member price, now gets checked by an assistant that compares you with your competitors every single time.

Maya was happy to let her agent shortlist jackets. Letting it spend her money is a different conversation. Before she says “go ahead and buy,” she wants to know three things: how much her agent can spend, how she can stop its transactions and how she can undo a purchase it made by mistake.

She isn’t alone. When the Checkout.com survey asked consumers which controls they’d need before letting an agent buy for them, those same three topped the list.

Retailers are starting to build those controls in, and that matches what most shoppers want: in the Customer Loyalty Index 2026, 60% of consumers want shared control with AI, and just 4% want a fully autonomous experience.

Shoppers are also a little contradictory, and brands need to design for that. In the Customer Loyalty Index 2026, 50% would rather a brand “just gets it right” without needing to engage, yet 34% of customers change their opinion on AI regularly. In other words, people want things to happen without being asked, right up until the moment they want to take back control. The brands that earn their trust will make both easy. For more on how much control shoppers want to keep, see our guide to the AI control spectrum in personalization.

For shoppers

Think about the jobs people put off: comparing six nearly identical vacuum cleaners, hunting for a discount code, remembering when the laundry detergent is running low. Agents are good at exactly that kind of work. In the Customer Loyalty Index 2026, 39% of consumers trust AI to find deals, and 27% use AI because they dislike shopping.

For Maya, the benefit was simple. She described what she needed once and got her evening back. For a parent restocking the same groceries every week, it might be never having to think about the list again.

For brands

An agent can bring you a customer who has never heard of you. If your products are the best match for what someone asked for, the agent can put them in front of that person, even if they’ve never visited your website. The agent learns from the shopper’s behavior, so if they keep choosing your products you can become its top pick over time, or the shopper may add your connector to their personal assistant.

It’s too early to say how widely shoppers will use third-party personal agents, or how much they’ll spend through them. Walmart and Amazon have said shoppers using their own assistants, Sparky and Alexa for Shopping, convert at higher rates and spend more.

• Being left out of the comparison. Maya’s favorite store lost her order because the agent couldn’t see a delivery date.
• Paying to be seen. Comparison sites and paid search both started by sending retailers traffic and ended up charging for the top spots. BCG expects agent-routed demand to follow the same path. Budget for it now.
• Not knowing why the agent chose what it chose. Advertising is coming into agent shopping experiences, with Google announcing checkout from some of its ads. Shoppers will want to know whether a recommendation was the best match or a paid placement, and brands that are open about it will be easier to trust.
• Wrong orders and unintended purchases. An agent that buys the wrong size or ignores a budget damages trust in the brand as well as the agent. Check what happens in every purchase route you support when something goes wrong.
• Fewer direct conversations. If Maya never visits your website, you have fewer chances to learn what she likes and to earn her consent to stay in touch. Every interaction you do get matters more.

Let’s use the example of Maya’s jacket to demonstrate the steps you’ll need to take:

1. Make sure the agent can find you

Maya’s agent shortlisted three jackets, specifically because it could see everything it needed in their listings: the price, the size, whether it was in stock and a delivery date that beat Friday.

Your products need to be just as easy to read. That means:

• Descriptions that answer the questions shoppers actually ask
• Prices and stock levels that update in real time
• Delivery and returns promises built into the product data rather than left on a homepage banner.
• Sending a product feed to personal assistants, and using schema markup AI agents can read.

2. Focus on the messages a human still sees

Even when an agent places the order, people still open the box, read the confirmation message and call customer service. Put coupons in box inserts. Add marketing messages to order confirmations and tracking updates. Train customer service agents to upsell and cross-sell when people call in.

Invest in physical experiences, too: pop-ups, in-store events and experiential rewards in your loyalty program. Mind share and brand love carry over into what the agent picks.

3. Choose how agents can buy from you

Decide which routes you’ll support: your own checkout, a shared cart on a platform like Google’s, a deeper integration inside an AI assistant, or a direct connection for personal agents through a standard like Poppy. Then walk through each one as Maya would. How does she approve the purchase? Where does she set a spending limit? How does she cancel an order or withdraw permission? If you can’t answer those questions in a minute, neither can she.

4. Connect your messages to what actually happened

Maya’s jacket was promised for Thursday. If a warehouse delay pushes it to Monday, the worst thing your brand can do is send her a cheerful “Gear up for your next adventure!” email the same morning. The shortlist rewarded your delivery promise. Your follow-up needs to know whether you kept it.

Maya’s agent may soon hear about a delay before your marketing team does: the Personal Agent Protocol’s authors are exploring push notifications that could tell a personal agent the moment an order ships or a flight is delayed.

Our Customer Loyalty Index 2026 recommends building agentic workflows that draw on ERP data, such as purchases, order orchestration and return management. SAP Engagement Cloud is built for that connection. Its automated workflows respond to real-time business events, such as orders, inventory, service tickets and loyalty milestones, and it works natively with SAP Commerce Cloud, Sales Cloud, Service Cloud and core SAP systems. Bring your customer data together with those events, and a delayed parcel can trigger an apology and a new delivery date instead of a sales pitch.

5. Decide what your own agents can do alone

If you’re answering the shopper’s agent with agents of your own, set their limits before they go live. A simple way to do it is to agree, for each workflow, whether the agent should only observe, suggest a response, act once someone approves it or act on its own. An agent that flags a delayed order is very different from one that issues refunds without asking, and your team should decide which is which before a customer finds out.

The two terms sound alike, but the agents work for different people. As we put it in our guide, agentic marketing is “the application of AI agents that can interpret goals, reason about what needs to happen, and take action across marketing workflows with a defined level of autonomy.”

The two will increasingly meet in the middle, with the shopper’s agent asking questions and the brand’s agents answering them. Our guide to what agentic marketing is covers the brand’s side in more depth.

Picture Sam, who has bought the same coffee pods every three weeks for two years. Sam tells an agent to keep the cupboard stocked. The agent will look for those pods, wherever they’re cheapest and quickest to arrive. Sam is loyal to the coffee. The store that sold it isn’t automatically part of the deal.

That’s a pattern our research already shows. In the Customer Loyalty Index 2026, 56% of consumers are more loyal to a specific product than the brand. When an agent does the buying, that gap gets wider, unless the store gives Sam a reason to name it in the instructions: a member price, a delivery promise it always keeps, a returns process that never causes a headache. As Ekta Chopra, who writes the AI-First Board newsletter, points out, a shopper who tells an agent to “find any moisturizer under $25” is expressing very different intent from one who says “reorder the brand I love.” Your job is to be the brand in the second sentence. Feeling valued still gets you there: 60% of consumers are more loyal to brands that make them feel personally valued.

Sara Richter, Chief Marketing Officer, SAP Engagement Cloud, puts it this way: “Now, loyalty is increasingly shaped by trust, convenience, and the decisions customers choose to delegate.” The brands that earn those delegated decisions will be the ones the agents keep coming back to.