Just-in-Time Does Not Mean Just-Right

The central question about a JITAI is not whether it can act quickly, collect more data, or produce individualized messages. It is whether the intervention makes a defensible match among a changing state, an appropriate option, a person’s ability to use that option, and an outcome that matters. This final synthesis brings those conditions and the available evidence together so the reader can judge when JITAI complexity is warranted, what current results actually support, and where the promise remains a hypothesis.

The just-right test

A JITAI is best understood as a chain of decisions rather than as a feature of an app. A person’s internal state or context changes. A sensor, device, or self-report captures some part of that change. The intervention interprets the resulting signal through a rule and decides whether to deliver an option. The person then has to engage with that option, and the engagement must influence a proximal mechanism before repeated changes can contribute to a meaningful distal outcome.

That chain gives “just-right” an operational meaning. The sensed state must be relevant enough to the mechanism the intervention is trying to change. The tailoring variable must represent that state well enough for the decision being made. The selected option must have a plausible route to a proximal response. The person must be able to use the option under the conditions in which it arrives. The burden and possible harm of delivery must be justified by the expected benefit. Finally, the study must measure an outcome that corresponds to the claim.

flowchart TD A["Changing state or context"] --> B["Meaningful and sufficiently valid tailoring signal"] B --> C{"Does the option match the target mechanism?"} C -->|No or unclear| X["Reject or redesign the rule"] C -->|Yes enough| D{"Can the person use it at acceptable burden?"} D -->|No or unclear| Y["Withhold, simplify, or change option or timescale"] D -->|Yes enough| E["Engagement and proximal response"] E --> F{"Does a meaningful distal outcome improve?"} F -->|No or unknown| G["Inspect measurement, adherence, comparator, and missing data"] F -->|Yes| H["Make a bounded evidence claim"] G --> I["Revise the rule or the study question"] H --> I I -.-> B

The loop is deliberately stricter than the phrase “personalized intervention.” It allows a JITAI to be useful without requiring perfect knowledge of the person. A location can be a useful risk signal even though it does not explain every reason someone is there. A recent behavior can identify an opportunity even though it does not prove why that behavior occurred. A self-report can reveal a private state even though it may arrive after the best decision point. The standard is not complete visibility; it is an honest relationship between what is observed and what the rule claims to do.

The loop also permits no delivery. If the relevant state is absent, withholding the option may conserve attention. If the state is present but the person cannot use the option, withholding or changing the option may be more fitted than interrupting automatically. The absence of a prompt is not evidence of a failed system unless the intervention was supposed to deliver one under that condition. A rule should be judged by the decision it was designed to make, not by the number of messages it sends.

This is why timing is only one dimension of fit. A prompt can arrive at the right moment according to a clock and still be wrong for the situation. It can arrive when a risk signal is present but when the person is driving, in a meeting, preoccupied, or too stressed to process it. It can reach a real vulnerability while offering an option that does not address the relevant mechanism. It can be low-burden for the device and high-burden for the person. “Now” is not a mechanism.

The same logic protects the identity of the subject. JITAIs are not synonymous with digital health, artificial intelligence, ecological momentary interventions, or generic personalization. Their defining contribution is a protocolized way of using dynamic information to decide whether and how to intervene on a fast timescale. A momentary intervention may offer support in daily life without tailoring the option to the person’s momentary state. A stable demographic profile may help customize a program without making it a JITAI. An algorithm may vary content without showing that its variation improves a meaningful outcome.

The distinction is important because the evidence base contains interventions at different levels of adaptation. The mental-health synthesis included JITAIs and EMIs because their terminology and operational descriptions overlap. It found that most included interventions were classified as EMIs and only eight as JITAIs. That does not invalidate the pooled evidence, but it means the result should not be described as a definitive test of one sharply bounded design.