Customers became less responsive to repeated promotional messages in an analysis of randomized marketing tests at a telecommunications provider, suggesting that a campaign’s timing and a customer’s previous exposure can matter alongside the offer itself.
The 2026 working paper by Peter Ebbes, Eva Ascarza and Oded Netzer combines results from repeated experiments at one provider in the Middle East. It is preliminary research, not a guarantee that the same pattern applies to other companies or communication channels.
The researchers found evidence of intervention fatigue: the effect of an additional message tended to decline with repeated exposure and recover slowly. A model using previous experimental results also identified opportunities to reach fewer customers while improving estimated revenue in a held-out evaluation.
What the experiments measured
The company sent promotional text messages to eligible prepaid mobile customers. For each campaign intervention, eligible people were randomly assigned to receive the message or to a comparison group that did not receive it.
The working paper describes roughly two years of data and 140 distinct campaign designs. Its detailed data section says the dataset included about 580,000 customers, while the abstract describes nearly half a million. The authors distinguish a campaign design from each weekly occasion on which that campaign was tested.
Randomized holdout groups help estimate what might have happened without a message. They do not mean that every customer would have reacted identically, or that the same offer would work as well after several previous contacts.
Why a single campaign average can mislead
A campaign may increase revenue on average while being valuable for some customers and ineffective for others. Differences in the offer and its timing can contribute, but the authors’ model found that variation in customer responsiveness accounted for much of the dispersion in estimated effects.
Some of those differences were not explained by the customer characteristics available to the researchers. A person’s response history may therefore contain useful information that is not captured by a simple demographic segment.
That finding does not justify assuming that more personal data will always improve a campaign. Data collection, consent and retention rules still apply, and a business needs a reliable way to assess whether targeting adds value.
Measure the effect of another message
A business considering another reminder can compare similar eligible customers who receive it with those held out. It can then examine incremental revenue, costs and customer complaints over a defined period rather than counting only clicks or redemptions.
For example, an offer that produces purchases from people who would have bought anyway may look successful in a campaign dashboard without adding much revenue. The practical distinction is explained in MBN’s guide to measuring extra sales from marketing.
Repeated tests can also reveal whether a result changes after several contacts. The study’s model found that response did not simply return immediately when messages stopped, although the speed and shape of recovery will depend on the business and its audience.
What the results do not promise
The authors report revenue gains above 10% in some especially informative scenarios in their held-out policy evaluation. That is not a prediction for an arbitrary marketing campaign or a measured return from installing a particular software tool.
The data came from one telecommunications provider and text offers to prepaid customers. Other products, email campaigns and paid advertisements may behave differently. The analysis is also a working paper, and its statistical model depends on assumptions about how responses vary.
For managers, the useful starting point is simpler than the model: record what customers were offered, when they were contacted and what happened to a comparable group that was not. That record can show when another message is useful and when it may be adding little.
