Can You Make an Algorithm Loyal?

Maya has bought the same coffee beans for seven years. Same roaster, same blend, same little ritual when the bag arrives. By any marketing definition, she is a loyal customer.

Last week she asked her AI assistant to reorder coffee. It came back with a suggestion: a different roaster, similar tasting notes, rated higher, 18% cheaper, arriving a day sooner. “Want me to switch?”

Seven years of loyalty, challenged in four seconds.

What Maya does next is the most important question in commerce right now. If she says “sure,” a lot of what we called loyalty turns out to have been habit and friction. If she says “No, always my usual roaster,” loyalty survives, but in a new form: an instruction she gave to a machine.

Both answers are happening millions of times a day. And the brands that understand the difference will keep their customers through the shift to agents.

A Lot of “Loyalty” was Really Friction

Let’s be honest about how loyalty has worked for decades. Some of it is real affection. But a big part of it is:

• Inertia. Comparing options is tiring, so we don’t.

• Switching costs. New accounts, new passwords, re-entering card details.

• Sunk points. “I’m 200 points from a free drink.”

• Visibility. The brand we saw last is the brand we bought.

Shopping agents remove almost all of that. Comparing costs nothing, switching is one tap, and the agent doesn’t care about your points balance unless it can read it. Industry research suggests around three in four customers would switch for a better discount. Until now, the effort of finding that discount protected incumbents. Agents take that protection away.

No wonder a widely cited survey found that 81% of retail executives expect generative AI to weaken brand loyalty by 2027.

But “weaker” isn’t quite right. Loyalty isn’t disappearing. It’s splitting in two.

Loyalty, Split into Two

1. Declared loyalty: What humans tell their agents

When Maya says “always my usual roaster,” she creates a standing instruction, a line in her agent’s memory that says pin this brand.

That kind of loyalty has unusual properties:

• It’s explicit. It isn’t a probability in a churn model. It’s a rule.

• It’s sticky. Once set, the agent follows it without re-deciding every time.

• It’s hard to win. Customers only pin brands they genuinely care about. You can’t buy your way into someone’s preferences with a banner ad.

The new top of the loyalty funnel isn’t “top of mind.” It’s “in the instructions.”

2. Earned loyalty: What agents learn from evidence

When there’s no standing instruction, the agent decides. And it decides the way a good procurement manager would: on track record.

• Did the price at checkout match the listed price?

• Did it arrive when promised?

• Was the return painless?

• Did the product match the description and the reviews?

An agent never forgets a broken promise, and it can explain its choice: “I picked this seller because 19 of your last 20 orders arrived on time and their return process took one step.”

That’s the shift behind the title: from loyalty based on emotion, which is hard to measure, to loyalty based on evidence, which can be explained. Humans forgive a late delivery if they love the brand. Agents just quietly move you down the list.

The Loyal-Customer Paradox: Your Best Prices are Invisible

Here’s a problem most loyalty teams haven’t noticed yet.

Shopping agents usually shop logged out. They compare public list prices. But for many retailers, the price that matters is the member price:

• Ulta Beauty says roughly 95% of its sales come from loyalty members.

• Kroger says over 95% of transactions are tied to a loyalty card.

So when an agent compares your anonymous shelf price against a competitor’s, you look more expensive to the agent working for your most loyal customer. Your loyalty program is actively working against you.

The fix is making loyalty machine-readable, and the infrastructure is arriving:

• Identity Linking in Google’s Universal Commerce Protocol (UCP), added in early 2026, lets a shopper connect their retailer account to an agent via OAuth. The agent then shops as a member, seeing tier pricing, rewards and free-shipping thresholds.

• Loyalty platforms are publishing incentives in formats agents can read (Talon.One launched a Unified Incentives Protocol in January 2026, for example).

• Payment players are carrying benefits into agent checkouts (PayPal’s integration with Perplexity is an early example).

The rule of thumb: If a benefit isn’t exposed through an API, it doesn’t exist to an agent.

What an Agent can’t Optimize away

Not all loyalty is equally at risk. A useful way to think about it, which Merkle describes as a three-tier hierarchy:

1. Transactional benefits: points, discounts, free shipping. Agents will apply these automatically. They become table stakes, not differentiators.

2. Experiential benefits: early access, exclusive drops, events, service perks. Harder to compare, and more valuable.

3. Emotional connection: identity, values, community, taste. The one thing an agent can’t compute away.

That maps neatly onto categories:

• Agent territory: groceries, household staples, printer ink, batteries, commodity electronics, subscriptions. Here, earned loyalty (price + reliability) wins, and switching will be brutal.

• Human territory: fashion, beauty, luxury, hobbies, fandoms, values-driven brands. Here, declared loyalty is strong, and customers will tell their agents what to buy.

• The contested middle: travel, consumer tech, restaurants, mid-market apparel. Brands here need to win both. The honest takeaway: if your brand has never given people a reason to care, agents will expose that very quickly.

New Loyalty Metrics for an Agent World

App opens and email click rates won’t tell you much when agents do the shopping. Start tracking:

• Pin rate: the share of customers whose agents have you set as a preferred or “always” brand.

• Agent recommendation rate: how often agents choose you when no preference is set.

• Promise-keeping score: price accuracy, on-time delivery, return friction and review honesty. This is effectively your credit score with agents.

• Linked-member share: the share of your loyalty members who have connected their accounts to an agent.

• Delegated share of wallet: of the purchases a customer lets agents handle, how much comes to you?

A Playbook for Explainable Loyalty

Make every benefit machine-readable.

Member prices, tier discounts, shipping thresholds, credits and subscriptions should all be structured data. Publish structural offers openly (subscribe-and-save, free shipping over $X). Signal that member pricing exists even if the exact number stays behind login.

2. Support identity linking.

Let customers connect their account to their agent in one step. Every linked member is one more customer the agent sees your real price for.

3. Treat operations as marketing.

Your on-time rate, return speed and price accuracy are now your brand promise, read by machines. The logistics team is suddenly part of the loyalty team.

4. Ask to be pinned.

Create moments where customers can easily tell their agent “always use this brand,” right after a great experience, not in a pop-up on the first visit. Think of it as the “follow” button for the agent era.

5. Move emotional budget into experiences agents can’t compare.

Early access, community, events, craft, a point of view. Points programs get commoditized. Belonging doesn’t.

6. Never try to trick an agent.

Hidden fees, bait pricing and fake scarcity might have fooled a hurried human. Agents record the gap between what you promised and what you delivered, and they don’t give second chances easily.

7. Prepare for agents that negotiate.

Agents will soon ask on their user’s behalf: “Is there a better rate before I cancel?” Decide your retention rules now, and make them consistent, because agents will compare notes across millions of customers.

The Twist: Loyalty might Move to the Agent itself

One uncomfortable possibility to end on.

If people trust their agent more than any individual brand, the real loyalty relationship shifts to the agent platform: Meta’s Muse, ChatGPT, Gemini, Amazon’s Rufus. Customers stay loyal to the assistant, and brands become interchangeable suppliers behind it.

And agents have owners with commercial interests. Will “the best option for you” quietly include sponsored placements? Will platforms charge brands for the right to be considered? This is where regulation, transparency standards and the banks’ new agentic commerce principles will matter a lot.

For brands, the defense is the same as before: be the name customers say out loud, and be the supplier whose track record an agent can’t argue with.

The Bottom Line

Loyalty isn’t dying. It’s getting honest.

The friction that used to pass for loyalty is disappearing. What’s left is two things that were always the real deal: people who genuinely love you, and a track record that proves you deliver. The first gets you into your customers’ instructions. The second gets you chosen when no one’s watching.

Maya, by the way, told her agent to stick with her usual roaster. But she also asked it to let her know if the price goes up again.

That’s the new deal: love, with receipts.

Sources and further reading: Merkle, PYMNTS, Alhena, Voucherify, Eagle Eye, The Wise Marketer, Forbes and company disclosures from Ulta Beauty and Kroger.

Author: Mohamed Haseel, Chief Architect and Head of Technology Practices

This blog is part 2 of the series on agentic commerce, first published on Medium.com. Watch this space for the next one in the series.