The Same Intervention Does Not Produce the Same Person

Two people join the same six-week programme for protecting focused work. They receive the same instructions, use the same planning technique and aim for the same result.

Six weeks later, one has established a reliable routine. The other has made little durable progress.

It is tempting to conclude that the intervention worked for one person but not the other. That may be true—but their different outcomes do not prove it.

Three questions must first be answered. Did they receive the same intervention in practice? What personal and contextual conditions did it enter? What would have happened to each person without it?

Different outcomes are visible. Different causal effects must be inferred.

The title does not mean that an intervention creates a person from scratch. It means that people who receive what appears to be the same intervention may emerge in different states, changed in different ways or to different degrees.

Understanding why requires us to distinguish what was assigned from what was received, what happened from what caused it, and what is true on average from what can be predicted for one person.

A shared label can conceal different interventions

Two people can be assigned the same programme without receiving the same intervention in practice.

An intervention has several stages. It must be delivered as intended. The person must be exposed to its relevant components, understand them and have an opportunity to use them. The intervention must then influence the process through which it is expected to create change—if that proposed process is correct.

These stages can separate.

A planning exercise may be explained clearly to one participant and ambiguously to another. Both may understand it, but only one may have a stable opportunity to enact it. Both may attend every session, yet practise at different intensities between them.

Even engagement can conceal several different things. Attendance does not guarantee attention. Attention does not guarantee understanding. Understanding does not guarantee enactment. Enactment does not establish that the proposed mechanism changed.

Implementation research therefore distinguishes the intervention described in a protocol from the intervention actually delivered, received and used. That distinction matters whenever different outcomes are attributed to differences between people.

It also prevents blame. Limited exposure may reflect caregiving demands, disability access, financial pressure, unsafe conditions or institutional restrictions. Describing these constraints as deficient motivation turns unequal opportunity into a personal defect.

Before asking why two people responded differently, we should ask whether the relevant intervention was functionally equivalent.

Every intervention meets a prior state

Even equivalent delivery would not make their starting conditions equivalent.

An intervention does not enter an empty person. It meets an existing biological, developmental, psychological and social organisation.

People begin with different levels of difficulty, skills, expectations and histories of learning. They have different habits, relationships and material resources. They encounter the intervention at different developmental moments and under different concurrent pressures. Some have substantial room for measurable improvement; others begin near the upper limit of what the outcome measure can detect.

Biological and genetic differences may influence relevant pathways, depending on the intervention and outcome being studied. They might affect processes such as metabolism, attention, learning or sensitivity in some domains. But influence is not destiny. A biological contribution does not determine what will happen to one person or identify the correct intervention by itself. Biology operates within development, experience and context.

Temporary states matter too. Sleep, illness, stress and recent events can change what is available at a particular moment. Someone measured during an unusually difficult period may later improve partly because conditions return toward their typical level. Someone beginning near a measurement ceiling may have little detectable room for further improvement.

Neither pattern, by itself, proves a different intervention effect.

The same technique can therefore encounter different possibilities. A plan that depends on a predictable cue will operate differently in a stable environment than in one where demands change by the hour. A reflective exercise may require a degree of safety or attention that is not always accessible.

The intervention is the same on paper. What it can do depends partly on what it meets.

The effect arises through an interaction

We often speak as though an intervention contains a fixed amount of effectiveness that it transfers to each recipient. But an effect is not stored inside a technique.

It arises from the difference the intervention makes relative to an alternative, under particular personal and contextual conditions.

The intervention brings specific components, demands and possible mechanisms. The person brings a prior state, history, capacities and expectations. The context enables some actions and obstructs others. Timing, dose, delivery and engagement influence whether the proposed pathway becomes available.

Consider the focused-work programme. Its planning technique may require a stable cue, memory for the plan, a feasible response and some control over when it can be enacted. If those conditions are absent, the technique cannot operate in the same way.

That does not mean the person belongs to a fixed category of non-responder. The intervention may need a different delivery format. Its proposed mechanism may not have changed. That process may have changed without affecting the selected outcome. The measurement may have missed a relevant benefit. Or the intervention may genuinely offer little benefit under those conditions.

Observed outcomes cannot select among these explanations on their own.

This interaction does not remove agency or imply that context determines everything. It means that action occurs within conditions that make some responses easier, harder or temporarily unavailable.

The intervention’s effect is therefore conditional—not because every person possesses a hidden response type, but because interventions operate through pathways situated in people, contexts and time.

An average is evidence, not destiny

Research commonly estimates what an intervention does on average.

Suppose a study compares a programme with an appropriate alternative and finds a better average outcome in the programme group. Under the study’s design and assumptions, that difference estimates an average intervention effect in the studied population.

That result matters. It tells us that the intervention was more promising, on average, than the specified comparison. But it is not a promise about every participant.

The average may combine people who improved substantially, people who improved slightly, people whose measured outcomes did not change and people whose outcomes worsened. It does not follow that everyone received the average benefit.

This does not make averages useless. Population evidence is one of our strongest resources for deciding whether an intervention is generally more promising than its alternatives. The mistake is not using the average. The mistake is asking it to answer a question it was not designed to answer.

For one person, causal benefit is the difference between what happened with the intervention and what would have happened to that same person, at the same time, without it.

Only one outcome can ordinarily be observed.

A person who improves might have improved anyway. Someone who shows no improvement might have deteriorated without the intervention. Measurement error, unrelated events and natural fluctuation can further complicate the comparison. Observed change is therefore not identical to causal benefit.

This does not make individual guidance impossible. Repeated measurements, crossover designs, adaptive experiments and validated prediction models can sometimes improve decisions for individuals. None removes uncertainty in every setting, and not every intervention can be safely or meaningfully tested within one person.

Population evidence guides individual decisions. It does not dictate individual futures.

Predictors, moderators and mechanisms

Researchers can investigate why outcomes and effects vary, but several questions are often collapsed into the phrase “what works for whom.”

An outcome predictor indicates who is likely to have a particular outcome. A moderator indicates whether the comparative effect of an intervention differs across people or conditions. A mediator is a measured process proposed to lie along the route from intervention to outcome. A mechanism is the causal process claimed to produce the effect.

These are not interchangeable.

Suppose participants with greater schedule control tend to establish stronger focused-work routines. Schedule control predicts a better outcome. But those participants might also have done better without the programme. Schedule control would then predict success without necessarily predicting greater benefit from the intervention.

To investigate whether it moderates the effect, researchers must compare the programme with an alternative at different levels of schedule control. Looking only at programme participants is not enough.

Finding evidence of benefit in one subgroup and inconclusive evidence in another is also not, by itself, evidence that the groups respond differently. The difference between their estimated effects must be examined directly.

Subgroup findings require restraint for other reasons. When researchers search across many ages, traits, histories and contexts, some apparent differences emerge by chance. Small subgroups produce unstable estimates. A pattern discovered and evaluated in the same data may disappear in a new sample.

A moderation claim becomes stronger when it was justified in advance, tested through an appropriate comparison, estimated with sufficient data and validated elsewhere. Even then, it describes an average difference between groups. It does not determine the effect for every group member.

A statistical moderator is not automatically an established mechanism. Nor does subgroup membership create a permanent human type.

The same person can respond differently later

Variation does not occur only between people. The same person may respond differently at different times.

A technique attempted during severe sleep deprivation may become usable after recovery. Support delivered during a calm period may have little immediate value but matter during a moment of high risk. An intervention requiring focused attention may work in one setting and become inaccessible in another.

This is why some adaptive interventions account for changing states such as stress, opportunity, vulnerability and receptivity. Researchers can also test support repeatedly across different moments to examine whether its immediate effect changes with time and context.

Earlier responses may alter later ones.

An initial success can increase skill, confidence and willingness to continue. Repeated prompts can instead create fatigue or irritation. Early practice may reduce the amount of later support required. The environment may react to changed behaviour, creating new opportunities or resistance.

The intervention is therefore not always acting on the same system twice. Previous intervention, behaviour and feedback may have changed the conditions of the next encounter.

Stable individual differences can still matter. But it is unsafe to infer a permanent response type while state, timing, context and learning remain plausible explanations.

The same person may not be equally receptive today and tomorrow. Personalisation must account for that movement.

Personalisation without prophecy

Personalisation is sometimes imagined as perfect matching: collect enough information, identify the person’s type and assign the intervention guaranteed to work.

The evidence supports a more disciplined ambition.

Personalisation can mean choosing among defensible options, adapting delivery to access needs and measuring whether the intended outcome changes. It can involve checking whether the intervention was actually delivered, understood and enacted before concluding that its proposed mechanism failed. It can incorporate informed preference, monitor benefit and harm, and revise the working explanation when the expected result does not appear.

This is structured learning under uncertainty—not unrestricted trial and error.

Responsible adaptation requires an appropriate outcome, enough observation to interpret it and defensible reasons to continue, modify or stop. Some interventions cannot safely be self-tested, rapidly switched or evaluated through informal observation. Personalisation does not remove the need for evidence, safeguards or professional judgment where these are required.

Applicability matters too. An effect estimated in one population or delivery setting may not transfer unchanged to another. The people studied, the people receiving the intervention and the contexts in which it is delivered may differ in ways that affect both implementation and outcomes.

A predictive model may also forecast outcomes without accurately predicting intervention benefit. A subgroup pattern may fail to replicate. Data may represent some populations better than others, producing less reliable recommendations for people already facing unequal access.

Responsible personalisation must therefore remain empirical, revisable and equitable. It should not convert biology into destiny, demographics into response types or constrained engagement into personal blame. It should not withdraw useful support because an uncertain model assigns someone to the wrong category.

The purpose is not to discover a permanent label hidden inside the person. It is to determine which option is most defensible under the present conditions, observe what follows and remain willing to update the judgment.

The same intervention does not produce the same person because neither interventions nor people operate outside delivery, history, biology, context and time. Different outcomes may reflect meaningful differences in causal effects, but they may also arise from unequal exposure, different starting states, measurement, chance or natural fluctuation.

Different outcomes are visible. What produced them remains a question for careful inference.

The responsible question is not simply whether an intervention works, but for whom, under which conditions, through what pathway—and how confidently we can know.