Normalization Process Theory: Why Interventions Become Routine (or Don’t)

Normalization Process Theory (NPT) is a sociological theory built specifically to answer one practical question: when a new intervention, technology, or way of working is introduced into a real setting, what actually determines whether it becomes routine practice or quietly fails to take hold? Where frameworks like the Consolidated Framework for Implementation Research (CFIR) catalogue the factors that predict implementation success, NPT does something different: it models the work that people actually have to do to embed a new practice into their everyday routines. That distinction matters for anyone running or evaluating an implementation — it turns “why isn’t this sticking?” from a vague impression into four specific, answerable diagnostic questions.

Where NPT comes from

NPT was developed by Carl May, Tracy Finch, Frances Mair and colleagues over roughly a decade of qualitative healthcare implementation studies, and formalized in “Development of a theory of implementation and integration: Normalization Process Theory,” published in Implementation Science in 2009 (May CR, Mair F, Finch T, et al. Implementation Science. 2009;4:29). The theory synthesizes findings from studies of telehealth, decision-support tools and other complex interventions across health and social care settings, and it has since been applied well beyond healthcare to any setting where a new practice has to become “just what we do” rather than a one-off pilot.

NPT’s central claim is that normalization — a practice becoming embedded and routinely sustained — depends on four kinds of collective work, done by the people actually expected to carry the practice out. Each is a construct with its own sub-components, and each translates directly into a question you can ask about a specific intervention.

The four constructs, as diagnostic questions

1. Coherence — does it make sense?

Coherence is the sense-making work: do the people involved understand what the new practice is, how it differs from what they did before, and what it’s meant to achieve? NPT breaks this into four sub-components: differentiation (can people distinguish the new practice from existing ones?), communal specification (does the group share an understanding of its purpose?), individual specification (does each person understand what it means for their own role?), and internalization (do people see genuine value in it, beyond compliance?).

Diagnostic question: If you asked five people who are supposed to be doing this differently now, would they give the same answer for what changed and why?

2. Cognitive participation — is anyone actually engaging with it?

Cognitive participation is the relational work of getting and staying enrolled: initiation (who is driving this?), enrolment (do others agree to participate?), legitimation (do people accept it’s their right and responsibility to be involved?), and activation (do participants keep the work going once the initial push fades?).

Diagnostic question: Six months after launch, is anyone still actively sustaining this, or did engagement stop the moment the project team’s attention moved elsewhere?

3. Collective action — does the work actually fit?

Collective action is where the practice meets the operational reality of the setting: interactional workability (does it fit smoothly into the existing flow of work?), relational integration (does it sustain trust and accountability between the people involved?), skill set workability (do people have, or get, the skills required?), and contextual integration (do resources, training and organizational support actually exist to sustain it?).

Diagnostic question: When staff are busy, does this practice get done anyway — or is it the first thing dropped?

4. Reflexive monitoring — is anyone checking whether it’s working?

Reflexive monitoring is the appraisal work: systematization (is there a way to know how it’s going, formal or informal?), communal appraisal (does the group collectively evaluate it?), individual appraisal (does each person evaluate its effect on their own work?), and reconfiguration (does that appraisal actually feed back into adjusting the practice?).

Diagnostic question: If this practice were quietly making things worse in some specific way, would anyone currently find out?

Measuring it: the NoMAD instrument

Because NPT’s constructs are abstract, Finch and colleagues developed NoMAD (Normalization MeAsure Development), a short survey instrument that translates each construct into concrete Likert-scale statements respondents can rate — letting a team move from “we think engagement is a problem” to an actual construct-level score, comparable across teams or over time. NoMAD doesn’t replace the diagnostic questions above; it operationalizes them into something you can administer to a staff group and score.

Worked example: a checklist that never became routine

Illustrative composite — not a real organization or a real dataset. The scenario and every number below are constructed for this article, using a seeded simulation script, specifically to show what a construct-level failure looks like in numbers rather than in the abstract. No institution, clinician, or patient described here is real.

Picture an outpatient clinic rolling out a new medication-reconciliation checklist: a one-page form clinicians are meant to complete at every visit where medications changed. Leadership explained the rationale clearly, clinicians largely agreed it was a good idea in principle, and three months in, someone runs a short NoMAD-style survey (a 1–5 agreement scale, three statements per construct) across a panel of 24 clinicians to find out why completion rates are still low.

The scores, computed from a fixed, reproducible random seed rather than picked to illustrate a point (methodology below the table):

Three constructs look healthy. Clinicians understand what the checklist is for and largely buy into it (coherence), and enough people are still engaged with it three months on (cognitive participation) that it hasn’t quietly died. But collective action is failing outright — not one respondent in the panel rated any collective-action statement a 4 or 5. Read against the sub-components, that’s specifically an interactional workability problem: the checklist was designed as a standalone form, not built into the electronic record workflow clinicians already use, so completing it means stepping outside the normal visit sequence during the busiest part of the day. Reflexive monitoring sits in between (65.3% agreement) because some clinicians are individually aware completion is low, but there’s no systematic team-level review catching it as a workflow-fit problem rather than a compliance problem — which is exactly why it hadn’t been fixed by the time this survey ran.

The practical implication follows directly from which construct failed: this is not a “remind people more” problem (that would be a cognitive-participation fix) or a “the training was inadequate” problem (that would be skill-set workability, a different sub-component of collective action). Interactional workability failures are fixed by changing how the practice fits into existing work — embedding the checklist fields directly into the electronic record’s existing medication-review screen, not adding a new one — not by re-explaining the rationale, which coherence already shows people understand.

Reproducibility note: the table above comes from a seeded pseudorandom simulation (Mulberry32 PRNG, fixed seed) generating 72 item-level ratings per construct (24 clinicians × 3 items), built to model this specific illustrative scenario — a smoothly-embedding intervention failing on one construct — rather than to assert a general finding about medication-reconciliation checklists. Anyone can regenerate the identical numbers by rerunning the script; it exists so the worked example is grounded in an actual, checkable computation instead of numbers chosen to look plausible.

Using the four constructs as a diagnostic sequence

When an intervention isn’t embedding and it isn’t obvious why, working through the constructs in order is more useful than guessing: coherence failures look like confusion or inconsistent explanations of purpose; cognitive participation failures look like enthusiasm that never turns into sustained ownership; collective action failures look like “we agree with this but it keeps getting skipped when things get busy”; reflexive monitoring failures look like nobody being able to say, with any specifics, whether it’s actually working. A single intervention can fail on more than one construct at once, but naming which one is failing — rather than treating “implementation” as one undifferentiated problem — is what turns a stalled rollout into a fixable one.

Frequently asked questions

Is Normalization Process Theory the same as the Consolidated Framework for Implementation Research (CFIR)?

No. CFIR is a determinant framework: it enumerates the factors (intervention characteristics, inner and outer setting, individuals, process) that predict whether implementation succeeds. NPT is a process theory: it explains the mechanism — the actual collective work of sense-making, engagement, operational enactment and appraisal — through which those factors translate into an intervention becoming, or not becoming, routine. The two are frequently used together rather than as alternatives: CFIR to identify what’s likely to matter going in, NPT to explain why a specific implementation is or isn’t sticking once it’s underway.

Does NPT apply outside healthcare?

Yes. It was developed from healthcare implementation studies, but the underlying claim — that embedding any new practice requires sense-making, engagement, operational fit and ongoing appraisal by the people doing the work — doesn’t depend on the setting being clinical. It has been applied to technology adoption, organizational change and policy implementation well beyond health services.

What’s the difference between cognitive participation and collective action?

Cognitive participation is about whether people are engaged and invested — whether they’ve enrolled and stay committed. Collective action is about whether the work itself is operationally possible — whether it fits the workflow, whether people have the skills, whether the organization actually supports it. An intervention can score high on cognitive participation (people genuinely want it to work) while still failing on collective action (it doesn’t fit how the work actually gets done) — which is exactly the pattern in the worked example above.

Can you use NPT before an intervention launches, not just to diagnose one that’s already struggling?

Yes, and that’s arguably its more common use: working through the four constructs during design — will this make sense to the people who’ll do it, who’s actually going to drive engagement, does it fit the real workflow, how will anyone know if it’s working — surfaces embedding risks before they show up as a stalled rollout, rather than only explaining one after the fact.

Related CASRAI resources

• The Consolidated Framework for Implementation Research (CFIR) — the determinant framework most often paired with NPT
• Choosing an evidence-based practice model for your organisation
• The Iowa Model of Evidence-Based Practice
• Johns Hopkins Nursing EBP model: PET process and appraisal tools
• Building a programme logic model that survives evaluation
• Research methods hub