The Optimization Horizon
All the tactics used to achieve short-term conversions train the future customer. The question that most dashboards never pose is exactly what they are teaching.
The Question the Dashboard Does Not Ask
Most marketing performance reviews end at the same point, by determining which campaign was successful. Metrics such as open rate, click-through rate, conversion rate, and cost per acquisition are used. The meeting concludes when the figures show that one campaign has performed well, and the budget is then allocated to the winning campaign.
That is not an incorrect question, but rather a flawed one, since behind every metric showing the outcome of an action this week lies a second question that the dashboard never raises: what did that action teach the customer to expect from us in the future?
A promotion which increases this month's conversion rate can cause customers to wait for discounts before making a purchase. If a message includes an urgency factor that results in a sale, the customer may come to distrust any subsequent message that fails to include such urgency. When an email used to re-engage customers is based on offering something for free, the relationship can seem transactional, which can, in turn, make a future request more difficult. Such situations are not exceptional; they are ordinary examples of the unrecorded side effects that occur when you optimize for the immediate action without taking into account what the sequence of actions is intended to build, or indeed, to destroy.
The situation is known as the Optimization Horizon problem, since the unit of measurement is too short to show what the optimization really is doing.
The campaign's outputs and the lessons it provided are different. Most organizations only assess one of them.
In October 2026, a study in Discover Artificial Intelligence proposed a deep reinforcement learning-based framework for e-commerce personalization. The technical architecture matters less than what the framework was designed to optimize. Instead of aiming to maximize a single conversion event, the system was designed to handle multiple objectives simultaneously: click-through rate, conversion rate, and long-term customer lifetime value. The reward function took all three into account.
The findings were instructive: when the system was allowed to optimize over time rather than just for the next action, the increase in lifetime value (8.3%) exceeded the increase in conversion rate (7.2%). This shows that if you design with the overall relationship trajectory in mind rather than focusing on the next click, you will perform better in both areas. Meanwhile, the short-term metric improves as the system ceases to regard it as the only important factor.
It is an engineering contribution, not a marketing strategy document. Still, the change it entails is strategic: the unit of optimization is shifting from the message to the relationship, and organizations whose measurement infrastructure remains based on isolated events are making decisions at the wrong level.
How Expectations Get Built
Each time a customer interacts with an organization, there is, alongside other things, a piece of data indicating what to expect on a future occasion. This is not something the customer does consciously. They are not checking out the pattern, but the pattern is registered.
An organization that sends emails twice a week teaches its customers to place less importance on individual emails. One that sends emails only when something significant occurs teaches customers to open every email. An organization that offers a discount if a customer hesitates teaches customers to hesitate. One that keeps its prices steady teaches customers to make up their minds without waiting.
These expectations are not confined. They affect the effectiveness of future campaigns, price elasticity, the response rate to outreach, and the level of customer incentivization required to take action. Over time, the expectations that a customer has accumulated before each interaction become just as important for marketing performance as any single campaign.
An organization that tunes each campaign on its own, without keeping track of what the sequence as a whole is building, is not remaining unchanged; it is actively helping to shape the customer base, though it is not choosing what kind of base to shape.
The expectations that a customer has accumulated before each interaction are just as important as any single campaign.
Specifically for Community Health
Community health organizations aren't using personalization algorithms, but they are using sequences. Each time a patient comes into contact with the organization —whether it's a reminder text, a follow-up call, a message about a missed appointment, or a campaign promoting a new service—that patient is being taught what to expect when the organization gets in touch.
A clinic that uses frequent monthly messaging ends up being ignored by patients, whereas one that sends messages only when something directly relevant occurs teaches patients to pay attention. Outreach that specifies a clear action and provides an easy way for patients to take it encourages a response. Conversely, outreach that arrives and offers nothing, a number that rings busy or a link that requires a login the patient doesn't have teaches patients it's not worth responding.
The issue involves not just whether the upcoming campaign will result in appointments, but also what the patient group has been conditioned to do as a result of communications over the past 12 months, when the next campaign comes around.
The Diagnostic Question
You don't need to set up a reinforcement learning system to extend the optimization horizon; include an additional question in each campaign review.
What expectation does this action establish in the people it affects, and is that expectation we would like them to have in the following interaction?
Patterns worth auditing in most organizations:
Discount Dependency
Where the campaigns that convert best are always those offering the greatest incentives, the company may be teaching its customers to expect them. Although the conversion figures appear good, the actual customer relationship is becoming more costly to maintain.
Urgency Erosion
If urgency language is used in most outreach — limited time, act now, don't miss this — and it appears even when the situation is not genuinely urgent, the signal loses meaning. Open and click rates tend to decline over time, not because the audience has changed, but because the calibration is lost.
Friction Normalization
If patients or customers repeatedly start a process the organization initiated but fail to complete it — because the path is broken, requires a tool they don't have, or asks for information they can't find — some of them have learned that starting is not worth finishing. The next outreach campaign begins at a deficit that the campaign itself did not create.
The Longer Bet
In commercial marketing, reinforcement learning systems aim for a kind of customer responsiveness that, over time, most human marketing teams have not yet identified as a strategic asset—this responsiveness represents a longer-term commitment than a single campaign and is also more lasting.
Most organizations are still unable to develop such systems, but they can start by posing the question that these systems are designed to answer. Instead of merely asking what worked, they should ask what it taught them. Rather than just considering what led to conversions, they should consider what the next interaction is now working against or in favor of.
Organizations that begin doing this now will run better campaigns, experience fewer surprises when their existing tactics cease to perform, and gain a clearer understanding of the underlying reasons without needing to use a reinforcement learning engine.
The metric indicating what the campaign achieved this week is not incorrect; it's merely not the only one. The other metric is what the campaign taught, and that one builds up over time.
Bai, Y. (2026). A deep reinforcement learning framework with hybrid actions and multi-objective rewards optimizes e-commerce personalized marketing. Discover Artificial Intelligence, 6(1), Article 1499. https://doi.org/10.1007/s44163-026-02429-z
Some ideas are worth discussing in the context of your organization.
