A phone detects a stressful moment and offers a breathing exercise. The prompt may arrive at just the right time or become another demand when the person cannot engage. How can a system decide when to offer help and when to wait?

01

The message before the usual routine

A person wants to brush their teeth regularly but sometimes forgets in a rushed morning. An app connected to an electric toothbrush could check whether they have brushed recently and, before their usual brushing time, decide whether to send a reminder. If they already brushed, it stays quiet. Inbal Nahum-Shani and Susan A. Murphy use examples like this to explain an intervention that adapts to a person's changing situation during everyday life.

The approach is called a just-in-time adaptive intervention, or JITAI. Its defining feature is a planned rule that repeatedly uses recent information to decide whether, when and how to offer support. Merely sending the same daily notification to everyone does not make an app adaptive. Nor does putting a long course on a phone. The decision needs to depend on what is known about the person's current state and context.

The aspiration is easy to like: help when it might matter and spare the person when it will not. But the authors insist that timing is a scientific question, not a slogan. A reminder can arrive too late, during a meeting or after someone has already done the task. The right comparison is not between a clever app and doing nothing in theory, but between specific decision rules and what they actually change.

02

An adaptive rule has parts that can be inspected

A JITAI specifies decision points: moments at which a choice about support will be made. It also specifies what information guides that choice, such as a recent activity report, and which options are possible: a reminder, a different kind of help or no message. The rule links the information to the option. In the toothbrush example, the recent absence of brushing is the trigger; the scheduled morning or evening window is the moment to decide.

A decision point need not produce a message. This matters because silence can be the best intervention when a prompt is unnecessary or burdensome. It also makes the protocol reviewable: a researcher can describe exactly when someone should have received a message and compare that plan with what the technology delivered. Adaptation here is not an app inventing advice without oversight; it is a specified way to use changing information.

The review distinguishes fast changes within a JITAI from slower changes to a treatment program. A care team might decide after several weeks whether to add coaching. The phone may decide several times a day whether one prompt is appropriate. These decisions can work together but answer different questions. A good design says which decisions are made by whom, at what intervals and with what evidence.

03

Need, opportunity and willingness are not the same

A vulnerable moment is not always one when help can be used. A stressful period may increase the risk of returning to a behavior someone wants to avoid. But the same stress may make it hard to read and use a long message. An opportunity for change might arise after someone completes part of a new routine. Receptivity asks whether the person can engage with a particular kind of help in that moment.

A person driving is not receptive to reading a phone, even if the app detects a difficult moment. Someone in a work meeting may be able to use a one-line reminder later but not a five-minute exercise now. Receptivity is therefore specific to the offered action, not a permanent trait of the person. Designing for it may mean shortening the support, offering it at another time or deciding not to interrupt.

This is the review's first major challenge: people may be least available to engage precisely when they need support most. That is not solved by collecting more sensor data alone. Developers need to test whether vulnerability and receptivity actually overlap and whether an alternative form of support would work better. Otherwise the app can optimize for sending messages in moments that look important while the person receives none of their benefit.

04

The immediate effect is not the long-term goal

A brushing reminder aims to change behavior in the next several hours. Better oral health lies much farther away. The review calls the first a proximal outcome and the second a distal outcome. The path between them cannot be assumed: a notification might lead to one extra brushing without establishing a lasting routine. It could even make someone less attentive to later notices if repeated too often.

To learn whether a prompt works, researchers can randomize whether it is sent at many decision points for the same participants. Such micro-randomized trials compare short-term outcomes after different options and investigate which states change the effect. They can ask whether a reminder helps more before an established routine, whether a shorter message works in a busy moment, and when no prompt is preferable.

The method provides stronger evidence about a particular short-term comparison than merely observing that people who opened the app did better. People who open a message may already be more motivated or less busy. Randomization helps separate the message's effect from those differences. Even then, a useful immediate effect does not establish that the entire program improves health. Longer-term evaluation remains necessary.

05

Opening a message is not the same as using it

Digital interventions often struggle to keep people engaged. A device can record whether someone tapped a notification, but not necessarily whether they read, understood or applied the advice. For a breathing exercise, even measuring the behavior can require another question that increases the burden. The authors argue that engagement includes physical, cognitive and emotional effort; app usage captures only part of that picture.

That is why designing an app only around click rates can mislead. A message might be opened out of irritation and quickly dismissed; another might not be opened because the person already used the skill. The review recommends learning from the people the intervention is meant to support, including their routines, limits and preferences. Designing with the people who will use the intervention can reveal why a protocol that looks efficient in a spreadsheet fails during an ordinary day.

Human support may also help. Coaching or therapy can build understanding and a working relationship that a phone notification lacks, while digital messages can offer brief support between meetings. Combining them requires attention to cost and timing. A team should test which part adds benefit and for whom, rather than assuming that a more elaborate package is automatically better.

06

Relationships are still missing from many designs

The review's third challenge is social. A person trying to change a routine does not live alone with an app. Family members, friends, coworkers and care providers may support the change or make it harder. A reminder timed perfectly for one individual's schedule may fail if it overlooks who else is present or whose help is needed. The authors call for more research on how relationships can be included without treating other people as mere delivery channels.

The presence of another person changes both opportunity and privacy. A prompt about a sensitive behavior might be unwelcome on a shared screen. A suggestion to ask a friend for support may be useful only if that relationship is trusted and the request is appropriate. These details cannot be inferred from location data alone. They require a theory of the behavior and input from the people who will live with the intervention.

A social design also changes what success means. The outcome might include whether support was actually available, whether responsibilities were shared fairly and whether the intervention added pressure to a relationship. The review does not claim that every JITAI should involve another person. It points out that ignoring relationships can make an apparently personalized system less responsive to the real conditions of change.

07

Precision means knowing when to stay quiet

Return to the hurried morning. A useful app needs more than a clock and a motivational sentence. It needs a reason to think the routine was missed, evidence that a prompt can help at this moment and a way to avoid adding another interruption when it will not. The same logic applies to more serious health behaviors, but the evidence must be stronger as the consequences become greater.

The review offers a discipline for building these systems: specify the decision rule, the immediate outcome and the long-term aim; test different options at the times they would be used; check who is receptive; and revise the design when engagement declines. More adaptive algorithms may learn from accumulating data, but their decisions still require evaluation. ‘Personalized’ is not evidence of effectiveness by itself.

The most honest promise is therefore modest. A digital intervention may offer timely, efficient support for a defined behavior under defined conditions. It can also over-notify, miss a vulnerable moment or ask for effort the person cannot give. The success of the approach depends as much on choosing silence and respecting daily life as on choosing a well-worded message.

What to keep

  1. Timing is part of treatment.
  2. Adaptation needs explicit decision rules.
  3. Engagement is not guaranteed by technology.
?

The nuance

What this story does not let us claim

The moment of greatest need may also be the moment when a person has least capacity to engage.

Go deeper: concepts and optional reading practice

Reversible cards

Test the concept

1 / 1

Topic assessment

Check whether you can explain the idea, not merely recognize it.

Every answer is grounded in the explanation above. Some questions introduce new cases to test whether you can apply and analyze the ideas, not merely recall them.

7 questions4 options per questionOne correct answer
01Which statement best captures the review’s central thesis?
02Which is the most appropriate operational definition of “Decision point”?
03Which case correctly applies the central concept?
04If two studies obtain different results, which interpretation best preserves the logic of this guide?
05Which complementary conclusion belongs to the paper’s core argument?
06Which formulation best distinguishes “Decision point” from a superficial label?
07Which limit should accompany any application of the paper to a new case?
Answer all seven questions.

Source & scope

Just-in-Time Adaptive Interventions: Where Are We Now and What Is Next?. Inbal Nahum-Shani & Susan A. Murphy. Annual Review of Psychology, vol. 77 (2026).

This editorial guide is an independent explanation. Consult the original publication for the studies, references and full context.

Open the original review ↗