Enterprise & Work

Why the Smartest AI Decision Might Be to Send Nothing

One-third of consumers said holiday email volume alone was enough to make them unsubscribe. AI systems optimized for clicks need measures for retention, margin and customer effort.

The Smartest AI Decision Might Be Not Sending the Message
The Smartest AI Decision Might Be Not Sending the Messagegeralt / pixabay
By Elena Vasquez7 min read

Why it matters

  • Validity’s 2026 holiday consumer survey found one-third of respondents said holiday email volume alone was enough to make them unsubscribe.
  • The essay’s proposed test compares existing messaging with a policy that holds promotions back under defined conditions.
  • That test tracks repeat purchases, retention, complaints and margin over a period that fits the buying cycle.
  • The essay reserves human review for ambiguous context, conflicting rules and actions that exceed the system’s authority.

By Elena Vasquez

One-third of respondents in Validity’s 2026 holiday consumer survey said holiday email volume alone was enough to make them unsubscribe, according to a Unite.AI Thought Leaders essay.

The essay turns to a customer who returned an order last week and still waits for a refund. A marketing system sees a recent purchaser with a high likelihood of responding. It queues a promotional email. From the customer’s perspective, the brand wants more money before returning what it owes.

Generative AI has made personalized messages much cheaper to produce. It has not made the decision to send one any easier. The source warns that a system can produce a relevant offer and predict a click while still making the wrong decision.

The Unite.AI essay’s author separates two problems: "A poorly written email is a content problem. A well-written email sent while the customer is waiting for their money is a decision problem." Better content cannot fix the second problem.

The stakes sit with repeat business, not just campaign performance, according to the essay. A system can improve click volume while damaging the relationship it was meant to support. A customer may also have an unresolved complaint or have bought the same product under a different email address.

What should an AI system optimize?

Sends, opens, and clicks are easy to count. The essay calls them useful operating metrics. It also describes them as tempting substitutes for what a business actually wants: a customer who sees value in the relationship and keeps coming back.

An AI system optimized for total clicks, with few constraints on contact frequency, may find that sending more messages improves its score. Each additional send can look reasonable in isolation. Across campaigns and channels, the same customer can end up hearing from the brand far more often than anyone intended.

The Validity survey does not establish that AI caused the overload. The essay uses it to show another risk: increasing activity can cost the relationship even when campaign metrics improve.

An unsubscribe is visible. The frustration that preceded it often is not. If the system gets credit for a click today but faces no penalty for making a customer less likely to return next month, it has little reason to account for that tradeoff.

Why does an unresolved refund change the decision?

The return may live in the order system, the refund status in a service tool, and the purchase in a marketing profile. Those records may not even connect to the same person. The model can work exactly as designed while deciding from an incomplete account of the customer.

Connecting those records changes what the system can know. It does not automatically change what the system will do. The decision process also needs to account for an unresolved refund and allow holding back a promotion as a valid outcome.

Otherwise, teams give the system more information without changing the behavior it receives reward for.

Context and objectives have to be designed together, according to the essay. Context explains what is happening. The objective and decision rules determine what should happen next. A return alone does not mean all communication should stop. The customer may need a refund update.

The distinction between a useful service message and another promotion matters. One serves the customer’s unresolved issue. The other asks for another purchase while the first remains unresolved.

Is a loyalty signup the same as loyalty?

Loyalty programs offer another version of the same problem. If success means more enrollments, a system may learn to recruit people with a discount. Enrollment rises. Whether those customers return without another discount is a different question.

The essay recommends three harder questions:

  • Are members coming back?
  • Is their spending profitable after the cost of the incentives?
  • Does the program change their behavior, or mostly reward purchases they would have made anyway?

Answering them requires connecting loyalty activity with purchase and service history over time. The source also calls for a comparison that separates the program’s effect from behavior that was already likely. A larger membership count cannot tell the team that on its own.

Where does human review belong?

People remain responsible for deciding what the system should optimize. The essay names retention, margin, and customer effort as tradeoffs that leadership has to own. A model can help evaluate those tradeoffs. It cannot take responsibility for choosing them.

The source proposes a practical starting question for any action: how hard would this be to undo? Holding a promotional message for review has different consequences from changing a price, issuing a credit, or canceling a reservation.

According to the essay, those consequences, the customer’s permissions, and the reliability of available context should determine what can run automatically and what needs approval.

An open refund does not need to become a manual review task every time. If the policy is clear, a rule can hold the promotion while service communication continues. Human review belongs where context is ambiguous, rules conflict, or consequences exceed the system’s authority.

If the information needed to make that distinction is missing, the system needs a defined fallback. It does not have permission to guess.

Reviewers also need more than a confidence score. The essay says they need to see:

  • Which customer records informed the recommendation
  • How current those records are
  • What the system is trying to achieve
  • Which constraints apply

A confident recommendation based on stale or mismatched records is still a bad recommendation, the source warns.

Overrides should become evidence. If people repeatedly stop the same kind of promotion, the source says, teams should record why. The pattern may reveal missing service data, an identity problem, or an objective that rewards the wrong behavior. Fixing that pattern is more valuable than asking a person to catch the same mistake indefinitely.

How can teams measure whether restraint works?

Teams can start with one decision rather than trying to redesign every customer interaction at once. Take a promotional send, identify the context it requires, and define the conditions under which it should be held. Keep necessary service messages outside that suppression rule.

Then compare the existing approach with a policy that holds back under defined conditions. The essay recommends tracking repeat purchases, retention, complaints, and margin over a period that fits the buying cycle. Sends and clicks still help explain what happened. They should not be the final verdict.

The test may show that fewer promotions protect the relationship. It may show that a particular restriction is too broad. The source treats both results as useful. Silence should not get a free pass any more than sending should. Both are decisions whose consequences can be measured.

The engineering work does not disappear, according to the essay. Teams need to:

  • Resolve customer records to the right person
  • Deliver recent events to the decision process while they can still affect the outcome
  • Apply permissions that constrain the action
  • Keep a record of what happened and why

A better objective cannot compensate for missing context. Better context cannot rescue the wrong objective.

What should leaders change now?

Leaders should ask what their systems count as a good decision, the essay says. If the answer is a click, the model may do its job while making the customer experience worse.

Trusted customer context gives people and AI a more dependable basis for deciding what to do. Clear objectives and controls then determine how that understanding gets used. The source wants a system that can explain why a message is useful now, recognize when it is not, and learn from what happens next.

The essay ends with a deliberate challenge to volume-driven thinking: "Sometimes the smartest decision an AI system can make is to send nothing at all."

The forward-looking test is measurable. Teams must determine whether holding back improves repeat purchases, retention, complaints, or margin over the buying cycle. Evidence can make restraint a managed decision rather than a missing message.

Source: unite.ai

Share this article:

More from Elena Vasquez

Elena Vasquez

Show full bio

Market editor covering technology and product news in artificial intelligence and machine learning at AI In Context.

708 articles

Related articles

  1. Personal AI Could Break Customer Loyalty Built on Friction
  2. Spara Launches Scout, an AI Sales Assistant That Works Before You Ask
  3. Bloomreach acquires AthenaHQ to connect AI search and personalization
  4. Meta's Muse Agent Takes Aim at the Subscription Economy's Debt to Forgetting
  5. Google Says Gemini Doesn't Train on Your Emails — With Exceptions

« Previous articleNext article »