By Derek Slager, Founder and Co-CEO, Amperity
A customer returns a jacket. A week later, her refund still hasn’t posted. That morning, the same brand emails her a set of picks “chosen just for you.” By lunch, a social ad offers her 20% off her first order.
Every one of those messages was tailored to some version of that customer. None of them was relevant.
The AI era demands smarter data. Moving from generating content to making decisions raises that requirement. A model can write a convincing offer without knowing much about the customer. Deciding whether to send it requires knowing who that person is, what just happened and whether the offer makes sense. Those are different problems.
The stakes are rising. In Amperity’s 2026 Consumer Priorities Report, 60% of U.S. consumers said AI had led them to choose a brand they hadn’t previously considered. When AI can introduce a customer to another brand in a single question, an irrelevant message gives them another reason to look elsewhere.
In 2027, the hardest part of AI won’t be creating. It will be deciding.
1. AI Will Make the Cost of Fragmented Identity Harder to Ignore
Most brands still run loyalty, ecommerce and marketing on separate systems. When one person shows up under a loyalty ID, two email addresses and a guest checkout, those systems can see multiple people. An existing customer can look exactly like a new prospect.
The result is acquisition spend aimed at people the brand already serves. Suppression lists are supposed to prevent this, but they only work when a brand knows who its customers are. AI-driven targeting adds speed and scale to the problem. A model built to find likely buyers can select current customers unless it knows to exclude them.
If the same person looks like three customers in your data, AI can optimize against all three. The targeting may work exactly as designed while the business pays to acquire someone it already serves. In 2027, I expect more brands to examine how much of their acquisition performance depends on that mistake. Resolving identity changes who they target, what they offer and how they measure the result. Reaching existing customers isn’t the mistake. Treating them like strangers is.
2. Relevance Will Beat Volume
Generating a tailored message is no longer the hard part. Deciding whether to send it is.
Consider a customer with an unresolved return. A model that works only from purchase history identifies a strong candidate for a follow-up offer. The customer sees a brand that isn’t paying attention. With the full picture, the right move changes from sending another promotion to fixing the problem first.
The cost of producing a message used to limit how many a brand could send. AI is removing that constraint. The customer’s attention is still limited, and generating another message doesn’t give them another reason to care.
In 2027, I expect investment to shift toward trusted customer data and the context it provides to determine whether a brand should engage at all. That includes service history, returns, recent interactions and consent preferences. A customer who agrees to one kind of message hasn’t agreed to every kind. I also expect more brands to measure whether an interaction was useful, including whether the better decision was to stay quiet. For the customer waiting on a refund, resolving the return matters more than improving the next offer.
3. Leaders Will Decide What AI Can Handle Alone and What Needs a Human Sign-Off
As AI moves from recommending to acting, the question changes. It’s no longer “Can the model do this?” It’s “Should it do this without anyone checking?”
Using AI to analyze an audience generally carries less risk than acting on the results. Changing how a brand treats a group of customers, through pricing, eligibility or a service response, carries much more. Those decisions affect real relationships, and some are hard to undo.
In 2027, I expect companies to get more explicit about what an agent is allowed to do. How much could the action affect the customer? How hard would it be to undo? How reliable is the underlying data? Can the team reconstruct what happened and why? Those answers should determine the permissions, approval requirements and escalation paths built into the workflow.
Clear sign-off points can give companies the confidence to give AI more responsibility. Without those decisions, every new use case becomes another debate about what the system can be trusted to do.
4. Knowing How to Question AI Will Matter More Than Knowing How to Prompt It
The first wave of AI training taught people how to use the tools. The next wave needs to teach them how to question what the tools produce. A confident answer still rests on assumptions.
Say AI recommends targeting a high-value audience. A marketer should be able to ask how value was defined. Are these distinct customers or duplicate records? What information is missing? How will the team know whether acting on the recommendation improved anything?
Those questions sound basic. They’re also easy to skip when the output looks polished. I expect them to become a more routine part of campaign reviews and training in 2027. The marketers who stand out will be the ones who can explain whether an answer is worth acting on, what would prove it wrong and how they will measure the result. Accountability for the decision still belongs to the team.
5. Loyalty Programs Will Have to Prove They Grow Customer Value, Not Just Membership
Enrollment numbers are easy to report and easy to misread. With 63% of consumers willing to switch brands for a better offer, a signup alone doesn’t prove loyalty. The same AI that helped consumers discover a new brand can help them find the next one.
In 2027, I expect loyalty teams to face a more demanding test: what changed because the program existed? A member’s lifetime value is useful context, but a high-value member is not proof that the program created that value. Teams will need to separate incremental retention and spending from purchases they would have received anyway, then account for the cost of rewards.
Answering those questions takes a connected view of each customer across channels, not a series of isolated transactions. It also brings us back to the first point. A brand that can’t tell its loyalty members from its acquisition targets can’t accurately measure the value of either program.
What Separates the Brands That Scale AI
Moving AI into customer decisions exposes problems that a convincing demo can hide. Does the system know who the customer is? Does it know what just happened? Are the customer’s permissions connected to the decision? Is it clear what the agent can do and when a person needs to step in?
In 2027, I expect those questions to determine which companies can give AI more responsibility. The work is in connecting customer identity, history, current signals and permissions, then making that context usable at the point of decision.
The customer should experience the result as less work: a problem resolved, an irrelevant offer avoided, a situation they don’t have to explain again. That is a more useful test of AI progress than how much content a company can produce.
##
ABOUT THE AUTHOR

Derek Slager is Founder and co-CEO of Amperity, where he leads the company’s AI-first transformation across product and operations. He co-founded Amperity to help organizations build trusted customer data and previously held engineering leadership roles focused on large-scale distributed systems and security.






