Whether to send: an order of decisions for marketing messages

1The problem

Most marketing automation is configured from the message outward. Someone writes a campaign, picks a segment, schedules a time and presses send; frequency caps and unsubscribe handling are settings applied afterward, if at all. The result is that the question that matters most, whether this person should receive this message now, is answered last or not at all.

We propose the opposite order. Every send, whether a one-off campaign or a triggered flow, passes five decisions in sequence, and a message that fails one does not reach the next.

2The five decisions

Order	Decision	Question it asks	What it protects
1	Consent	Has this person agreed to this kind of message on this channel, under the rules of their market?	The recipient and the sender's legal position
2	Suppression	Has this person opted out, bounced, complained, or is there another reason not to contact them?	The recipient and the sender's legal position
3	Fatigue	Has this person received enough from us recently?	The relationship that makes future messages possible
4	Local time	Is it a reasonable hour where this person is?	The response to this message
5	Uplift	Is this message likely to change what this person does?	The response to this message, and the budget

Source: our framework. Steps 1 and 2 follow the US and EU rules in^{1, 2}; step 5 follows^{3, 4}.

2.1Consent

The rules differ by market, which is why consent comes first and is read per recipient. In the United States, commercial email does not require prior consent, but it must not be deceptive, must be identified as an advertisement, must include a valid physical postal address and must offer a way to opt out that the sender honors within 10 business days¹. Under the GDPR, where consent is the legal basis for processing, it must be freely given, specific, informed and unambiguous, and the person must be able to withdraw it as easily as they gave it². A system that stores only a single subscribed flag cannot answer the consent question for a recipient in a market with stricter rules. It needs to record what the person agreed to, on which channel, when and how.

2.2Suppression

Suppression is the list of people who must not be contacted regardless of what a campaign wants: those who opted out, addresses that bounced, people who marked a message as spam. It is checked second because it overrides any consent recorded earlier. The opt-out deadline in US rules¹ and the right to withdraw consent at any time² both mean suppression has to apply to every send, including flows already in progress.

2.3Fatigue

People who receive too much leave. A study of a firm's permission-based email program modeled how long customers took to opt in and how long they stayed before opting out, and found that marketing intensity affects both, and that a firm can extend the time before opt-out by how it allocates its contacts⁵. The sender's own frequency is one of the forces that ends the permission it depends on. A fatigue rule counts all marketing messages a person received across campaigns and flows over a recent window and holds a send that would exceed the limit. It comes before timing and targeting because a well-timed, well-targeted message that pushes someone to unsubscribe costs every future message to that person.

2.4Local time

A message that arrives at three in the morning is read with everything else the next day, or not at all. The send time should be computed in the recipient's own time zone, taken from their address or recent activity, with a fallback when it is unknown. This is an optimization, which is why it comes after the three protective steps.

2.5Uplift

The last question is whether the message will change behavior. The intuitive answer, send to the customers most likely to leave or most likely to buy, turns out to be wrong in a useful way. In field experiments on retention campaigns, targeting the customers at highest risk of leaving was less effective than targeting those whose behavior the campaign was most likely to change, because many high-risk customers did not respond to the intervention at all³. A broader review of retention management reaches the same point: the decision should rest on the effect of the action, not only on the risk⁶.

Estimating that effect is the task of uplift modeling, which predicts the difference in outcome with and without the message for each person. A review of the literature describes the main approaches, from training separate models on treated and control groups to transforming the outcome so a single model estimates the difference, and stresses that they need a randomized control group to learn from and to evaluate on⁴. A store that keeps a small random holdout from every campaign builds the data for this step at no extra cost.

3Limitations

This paper proposes an order, not a set of thresholds; the right fatigue limit and holdout size depend on the channel, the audience and the store. We cover US and EU rules only, and only at the level needed to explain the order; other markets have their own rules, and none of this is legal advice. The uplift evidence comes mainly from retention campaigns at larger firms; its size at small volumes is not established.

4Conclusion

Whether to send is a sequence of five decisions, and their order matters: consent and suppression protect the recipient and the sender, fatigue protects the relationship, and only then do timing and uplift optimize the response. A system that enforces this order on every send, and keeps a random holdout, can answer the question that matters most before it answers the ones that are easier to optimize.