Normalization Process Theory: why interventions fail to stick

Normalization in NPT means a practice becoming routinely embedded in everyday work — done without deliberation, sustained without a champion, surviving the departure of the person who introduced it.

The theory's central move is to treat that embedding as the outcome of collective work that people do, individually and together. It is not primarily about whether staff have positive attitudes toward the intervention; it is about whether the sense-making, relational, operational and appraisal work required to normalise it actually gets done, and whether the setting makes that work possible.

The practical consequence is that NPT findings tend to be actionable in a way that attitudinal findings are not. "Staff were resistant" supports nothing. "Nobody could distinguish the new pathway from the existing referral route, so both were maintained in parallel and the new one was dropped under pressure" identifies a coherence failure with an obvious remedy.

NPT was developed by Carl May, Tracy Finch and colleagues from work on why telehealth and other complex healthcare interventions repeatedly failed to embed. The foundational statement is Development of a theory of implementation and integration: Normalization Process Theory in Implementation Science.

Its practical use in evaluation is set out in Using Normalization Process Theory in feasibility studies and process evaluations of complex healthcare interventions, which is the single most useful reference for an evaluator applying it. A current overview appears in the Handbook on Implementation Science.

A quantitative instrument, NoMAD, was developed from the constructs — see Improving the normalization of complex interventions: measure development based on normalization process theory — and has been translated and validated in several languages, for example the Swedish version. Most applied work uses the qualitative constructs; NoMAD is the option where a survey across a large staff group is feasible.

A regional health service introduced structured medicines reviews in community pharmacies. Uptake was strong for four months and then fell to near zero in eleven of fourteen sites. A process evaluation used NPT across pharmacists, technicians, practice managers and two GPs per area.

Coherence was strong among pharmacists — they could articulate what the review was for and how it differed from a dispensing check. It was weak everywhere else. Technicians described it as "the pharmacist's paperwork", and GPs largely did not know it existed, which mattered because the review generated recommendations that needed a prescriber to act.

Cognitive participation had been carried by a regional lead who moved post in month five. No site had a designated owner. This coincided exactly with the drop-off.

Collective action revealed the operational obstruction. The review took 25 minutes and required the pharmacist to leave the dispensary; in single-pharmacist sites this stopped dispensing entirely. The three sites where uptake was sustained were the three with a second pharmacist — a workflow fact, not a motivational one.

Reflexive monitoring was almost entirely absent. Pharmacists sent recommendations to GP practices and heard nothing back. Several said they had stopped because they had no evidence any of it was ever acted on.

The recommendations followed directly from the constructs: brief GP practices and establish a response loop (coherence and reflexive monitoring), name a site-level owner (cognitive participation), and either fund locum cover or restrict the service to multi-pharmacist sites (collective action). A satisfaction survey would have reported that pharmacists valued the service, which was true and would have explained nothing.

• Using it as a coding frame and stopping there. Sorting data into four boxes is not analysis. The finding is in how the constructs interact and where they fail.
• Only interviewing adopters. Guarantees a study that cannot explain non-adoption.
• Asking about attitudes. NPT is a theory of work. Questions about beliefs and satisfaction produce data the constructs cannot use.
• Ignoring what does not fit. Hierarchy, funding and professional boundaries frequently determine the outcome and sit outside the framework. Say so.
• Retrospective-only designs. People reconstruct implementation work poorly; where possible, collect during rather than after.
• Treating NoMAD as a substitute. The instrument measures the constructs; it does not explain them.
• Assuming normalization is desirable. Embedding a practice that turns out to be ineffective is a bad outcome, and NPT is silent on whether the intervention was worth normalising.

A strong NPT study shows sampling across the roles implicated in the work, not just the intended users; a topic guide asking about concrete practice rather than the constructs by name; analysis that reports negative cases and material the framework could not accommodate; attention to interactions between constructs; and recommendations that name specific work to be changed rather than "improve engagement".

A weak one is recognisable at a glance: four sections, one per construct, each reporting that the construct was "partially achieved", with no account of what anyone should do differently.

NPT explains embedding. It does not assess whether the intervention works, and it is indifferent to whether the thing being normalised is any good — a well-normalised ineffective practice is a success by the theory's own terms and a failure by any other.

It also under-theorises power and structure. Where a practice fails because of a professional demarcation dispute or because the funding model pays for something else, NPT can describe the resulting collective-action problem but not the cause of it. Framework-guided evaluation using CFIR handles the outer setting more explicitly and is sometimes the better choice.

It carries a healthcare accent. The theory was built from health-services work and its examples, language and validated instruments all sit there; it transfers to education, social care and organisational settings but the fit needs argument rather than assumption.

Framework-guided evaluation using CFIR or RE-AIM covers overlapping ground with more attention to the outer setting and to reach.

Process evaluation more broadly is the family this sits in; NPT supplies the theory that generic process evaluation often lacks.

Fidelity assessment asks the complementary question of whether the practice was delivered as intended, which NPT deliberately does not.

Realist evaluation shares the interest in mechanism and context and is worth pairing where the question is not just whether a practice embedded but why it embedded here and not there.

NPT studies are deductive coding at scale: thirty to eighty interviews across several roles and sites, coded to four constructs and their sub-components, then compared across sites to find where a construct failed and why.

That cross-site comparison by construct is the analytic step that produces the recommendation, and it is exactly the operation a qualitative platform is built for — retrieving everything coded to collective action across fourteen sites and reading it together. Evidano supports Normalization Process Theory studies as a named methodology, so coding follows the construct structure rather than an emergent frame. Deciding which material genuinely belongs to a construct, and which is telling you the framework does not fit, stays the analyst's call.