Why customer AI agents raise the bar for marketing relevance

src-10a6d446

As customers begin using AI agents to review brand messages, marketers may face a tougher test of whether each notification is timely and relevant. Paul Meinshausen, MoEngage’s Head of Agentic Decisioning and co-founder of Aampe, says the shift makes it more important for marketing systems to learn from how people respond—not simply to send messages across more channels.

Meinshausen described asking Meta’s Muse to assess the marketing notifications he had received. The agent’s review was more critical than he expected, he told CX Today. The experiment prompted a broader question for marketers: what happens when software examines a customer’s communications on their behalf?

More attention to each message

Meinshausen sees personal agents as a change in how customers can allocate attention, rather than merely a new communications channel. He compared them with human assistants who closely manage another person’s affairs, noting that AI agents could make that kind of support available more widely.

People divide their attention among responsibilities such as work, family and relationships. An agent, by contrast, can spend more time on an individual task or decision. In Meinshausen’s view, that raises expectations for the detail, relevance and personalization of brand communications. It also increases the amount of work required to meet those expectations.

For customer experience and CRM teams, the implication is not simply to tailor more messages. If a customer’s agent is evaluating what arrives, sending frequently without accounting for the recipient’s interests may make a brand’s communications less useful. Relevance has to be demonstrated through the message and its timing, rather than assumed from the fact that a campaign was delivered.

Learning from responses, including negative ones

Meinshausen distinguishes intelligent activity from automation that merely performs more actions. He describes intelligent activity as taking an action, observing the result and updating what happens next. Applied to marketing, that means a system needs to learn from customer responses, not just select a message, channel or send time.

Those responses can take several forms. A retailer may track whether customers read material or click through, while also considering direct feedback about relevance. Meinshausen said businesses should make their agents correctable, both through explicit feedback and through signals inferred from behavior.

Not every useful signal indicates approval. Meinshausen cited unsubscribing, saying that a message is unwanted, and failing to respond when a response was expected as forms of negative feedback that systems should take into account. If an agent continues sending without learning from such signals, it risks becoming a source of over-messaging rather than a tool for more relevant communication.

This puts suppression and message selection within a broader feedback loop. A system that accounts for customer reactions can adjust subsequent activity; one that treats delivery as the end of the process has little basis for knowing whether its approach worked. The distinction is particularly relevant as marketers consider how decisioning tools should respond to behavior across customer journeys.

Adoption requires prioritization

Meinshausen expects AI decisioning to enter lifecycle and CRM marketing quickly, with applications also extending to orchestration and content production. He acknowledged that teams have more potential work than they can take on at once, making it necessary to prioritize and calibrate adoption decisions.

That does not mean waiting until every possible use case is settled. Meinshausen’s message to marketers was that there is “no time to do nothing.” A practical starting point is to assess what signals a marketing system uses, whether it can incorporate negative feedback, and how it changes future actions in response.

As customer-side agents become part of how people manage incoming communications, brands will have to earn attention through useful interactions. The measure of progress is not simply the number of automated messages sent, but whether systems can observe the consequences and improve what they do next.