Topic 8.3 — Behavioural Science & Adoption

Deploying a system changes what is possible; only behaviour change makes it real, so treat adoption as a designed, measured outcome — not a hopeful by-product of go-live.

Why this matters in health and care#

A system that is installed but unused is not a neutral outcome. In health and care it is a live risk: half-adopted tools breed workarounds, dual recording, and gaps where information should flow, and each of those is a route to a clinical incident. When a hospital buys electronic prescribing but a proportion of prescribers keep scribbling on paper, the record is now split across two media and the safety case built on the digital system quietly fails. Deployment and adoption are different achievements, and the distance between them is measured in behaviour.

On the patient and citizen side the stakes are about equity and outcomes. An app that manages a long-term condition only helps the people who download it, activate it, and keep using it week after week — and those people are disproportionately younger, wealthier, more digitally confident, and healthier to begin with. If you measure success by "the service is available", you can report a triumph while widening the very inequalities the NHS exists to narrow. The people who most need a diabetes or hypertension pathway are often the least likely to adopt it unaided.

Behavioural science is the discipline that takes adoption seriously as an object of design and evidence. It sits alongside — not inside — change management (see Topic 8.1 — Change Management), which leads the organizational transition, and communications and engagement (see Topic 8.2 — Communications & Engagement), which builds understanding and trust. Behavioural science is narrower and more specific: it asks what makes a single target behaviour more or less likely for a real person in a real context, and it uses that understanding to make the desired action the easy, obvious one. Because you are influencing people's health decisions with public money, doing this well — and doing it ethically — is part of your duty of care, not an optional refinement.

Core concepts#

Behaviour change and behavioural science. Behavioral economics and the wider behaviour change field start from a realistic model of people: attention is scarce, effort is costly, defaults are sticky, and decisions are shaped by context far more than by information alone. Giving clinicians a memo, or citizens a leaflet, rarely changes what they do; changing the surrounding environment usually does. The practical question is never "do people know they should?" but "what makes the right action easy, and the wrong action harder, here?"

The COM-B model and the Behaviour Change Wheel. COM-B, developed by Susan Michie and colleagues at University College London, states that any behaviour (B) requires three conditions: Capability (the physical and psychological ability to do it), Opportunity (the social and physical context that permits and prompts it), and Motivation (the reflective and automatic drives toward it). The Behaviour Change Wheel wraps COM-B in a systematic method: diagnose which of Capability, Opportunity, or Motivation is the binding constraint, then select intervention functions (education, training, enablement, environmental restructuring, incentivization, and others) that address that specific gap. Its discipline is that it forces you to diagnose before you intervene, rather than defaulting to yet more training or yet more communications.

Nudge theory and choice architecture. Nudge theory, from Richard Thaler and Cass Sunstein, holds that small changes to the way choices are presented — the choice architecture — can steer behaviour without removing options or changing economic incentives. Defaults, framing, ordering, and salience are the levers. The UK's Behavioural Insights Team (the "Nudge Unit"), founded inside the Cabinet Office in 2010, made this operational in public services, testing interventions such as social-norm messages and SMS reminders through controlled trials.

Adoption, activation, and sustained use. These are three distinct behaviours, not one. Adoption is starting — logging in, downloading, enrolling. Activation is reaching the point where the user experiences the core value (the "aha" moment: sending a first secure message, completing a first remote reading). Sustained use is continued, embedded use weeks and months later. Many services are strong on adoption, collapse at activation, and are graveyards by month three — so measuring only the first tells you almost nothing.

Friction, defaults, reminders, and feedback. Friction is anything that makes a behaviour harder — a login, a form field, a step, a wait. Reducing friction on the path you want, and adding it to paths you don't, is the most reliable behavioural lever you have. Defaults set the outcome for people who do nothing and are powerful precisely because so many people do nothing. Reminders solve forgetting; feedback (progress, comparison, confirmation) sustains motivation once the behaviour has begun.

Habit formation. A habit is a behaviour that becomes automatic through a cue–routine–reward loop repeated in a stable context. Sustained use of a clinical system is really a habit-formation problem: you want opening the system to become the unthinking first step of a ward round, not a conscious decision each time. Habits form through repetition tied to existing cues, which is why embedding a new tool into an established workflow beats bolting it on beside one.

Incentives and their limits. Incentives — financial, professional, or reputational — can shift behaviour, but crudely applied they backfire. They can crowd out intrinsic motivation, produce gaming to the metric, and stop working the moment they are withdrawn. Self-determination theory explains why autonomy, competence, and relatedness usually sustain behaviour more durably than external reward alone.

Diffusion and adoption curves. How an innovation spreads through a population is described by diffusion of innovations and the technology adoption life cycle — innovators and early adopters first, then the majorities, then laggards. This connects behavioural science to the wider question of phasing in innovation (see Topic 8.0 — Phasing in Innovation): behavioural levers must be tuned to where a group sits on that curve.

Best practices#

1. Define the target behaviour precisely, for a named group, in a named context. "Improve engagement" is not a behaviour; "district nurses record a wound assessment on the mobile app during the home visit, not afterwards on paper" is. Behavioural science only works on specifics: who, does what, when, where, instead of what. Write the behaviour down at that resolution before you design anything, because a vague target guarantees a vague intervention.

2. Diagnose the barrier with COM-B before choosing a fix. Ask whether the binding constraint is Capability (they can't — skills, understanding, or the device fails), Opportunity (the context blocks it — no time, no signal, no prompt), or Motivation (they won't — no perceived benefit, competing habits, or distrust). The intervention follows from the diagnosis: a Capability gap needs training or simplification, an Opportunity gap needs environmental change, a Motivation gap needs a different lever entirely. Skipping diagnosis is why organizations answer an Opportunity problem with more emails and wonder why nothing moves (see Topic 8.1 — Change Management on diagnosing resistance).

3. Reduce friction on the desired path relentlessly. Every extra click, login, free-text field, and context-switch loses a fraction of your users, and the fractions compound. Count the steps in the current journey, then delete, pre-fill, and default your way to fewer. Single sign-on, sensible defaults, and forms that arrive pre-populated from the record often move adoption more than any amount of persuasion, because they change Opportunity rather than merely appealing to Motivation.

4. Set defaults deliberately and ethically. Because most people accept the default, the default is a design decision with real consequences — make it the safe, recommended option, and make opting out easy and transparent. Opt-out structures (for example, a default appointment slot the patient can change) raise uptake far more than opt-in equivalents. The ethical line is that the default must serve the person, the alternative must remain genuinely available, and you must be open about what you have done.

5. Use reminders and feedback to convert intention into action and sustain it. Well-timed prompts close the gap between wanting to act and acting — the classic behavioural finding behind SMS appointment reminders that cut did-not-attend rates. Feedback (a progress bar, a streak, a "your reading is in range" confirmation, a peer comparison) sustains a behaviour once begun. Keep prompts specific, timely, and sparing; a channel that over-notifies is soon muted or uninstalled.

6. Design for habit by embedding into existing workflow and cues. Do not ask people to remember to use a new tool; attach it to something they already do at a fixed point — the login that starts the shift, the moment before discharge, the existing daily huddle. Stable cues plus repetition build automaticity. A tool that lives inside the workflow becomes a habit; one that sits beside it stays a chore and is abandoned under pressure.

7. Use incentives cautiously, and watch for gaming and crowding-out. If you attach a payment or a performance target to a behaviour, expect people to optimize for exactly what you measure — sometimes at the expense of what you meant. Prefer intrinsic and professional motivators (autonomy, mastery, peer norms, visible patient benefit) where you can, and treat financial incentives as a short-term catalyst rather than a permanent prop. Always ask what breaks when the incentive is removed, because a behaviour that only exists while you pay for it was never adopted.

8. Test interventions rather than assuming them. Behavioural effects are context-dependent and frequently surprise even experts, so pilot, A/B test, or run a small randomized controlled trial where you can — the working method of the Behavioural Insights Team. A two-arm comparison of two reminder wordings costs little and can double effectiveness. Testing also protects you from confidently rolling out a nudge that does nothing, or worse, backfires for a subgroup.

9. Check the equity effect of every nudge, and design for the disadvantaged first. Nudges and digital defaults tend to work best for people who are already advantaged — literate, connected, confident, well — which means an intervention can lift the average while widening the gap. Segment your adoption data by age, deprivation, language, and disability, and design assisted routes, human fallbacks, and proxy access for those the digital path leaves behind (see Topic 1.9 — Digital Inclusion & Accessibility and Topic 3.3 — Patient & Citizen-Facing Digital Services). An adoption strategy that only reaches the easy-to-reach is a failure dressed as a success.

10. Measure behaviour and adoption, not deployment. Instrument the actual behaviours — logins, activation events, feature use, sustained-use cohorts, workaround and reversion rates, did-not-attend and adherence rates — and report them against a baseline, segmented by group. Deployment metrics ("system is live in 40 sites") flatter you; behavioural metrics tell you the truth. Make sustained use at ninety days, not go-live, the definition of done (see Topic 10.1 — Key Performance Indicators).

Questions to discuss with your team#

1. For the change we are about to make, what is the single precise behaviour we need, from whom, and which of Capability, Opportunity, or Motivation is actually blocking it today? This question forces the team past the comfortable abstraction of "driving adoption" into the specific behaviour that has to happen — and by whom, when, and where. In a health and care setting the honest answer is often uncomfortable: the barrier is not that clinicians lack awareness or motivation, but Opportunity, because the system adds ninety seconds to a consultation they do not have, or Capability, because it fails on the ward's ageing hardware. Debate should surface the difference between what leadership assumes the barrier is and what frontline observation and data actually show, because the two frequently diverge. Consider whether you have watched real users attempt the behaviour, or are inferring from complaints and hope. A good answer names one behaviour at high resolution, cites evidence for which COM-B component is binding, and commits to an intervention that matches that diagnosis rather than defaulting to more training or more communications. The best version identifies a case where the team was about to fix the wrong component and redirects the effort.

2. Where are we relying on nudges and defaults that may work better for our already-advantaged users than for those who need the service most — and how would we even know? This is the equity question that behavioural science makes unavoidable, because the same techniques that lift overall uptake can quietly widen inequalities. The tension to debate is real: an opt-out default or a slick digital-first journey can raise your headline numbers while leaving older, poorer, less digitally confident, or non-English-speaking patients further behind, and an average that improves can hide a gap that grows. The discussion should confront whether your adoption data is segmented at all — by deprivation, age, language, disability — or whether you are reporting a single flattering figure. It should also examine what assisted and non-digital routes exist for people the nudge does not reach, and whether those routes are resourced or merely notional. A serious answer accepts that "available to all" is not the same as "adopted by all", commits to measuring the distribution and not just the mean, and treats a widening gap as a failure condition that would stop or change the rollout. The weakest answer is the one that has never looked.

3. Where is the ethical line between helping people make good health decisions and manipulating them — and are we on the right side of it, transparently? Influencing behaviour in health carries a duty that ordinary marketing does not, because the power asymmetry is greater and the subject is people's bodies and lives. The debate should distinguish legitimate choice architecture — making the healthy or recommended option easy, while keeping alternatives genuinely available and being open about what you have done — from covert manipulation, dark patterns, exploiting fear, or hiding the opt-out. Consider concrete cases: a default screening appointment the patient can freely change is defensible; a design that makes declining deliberately difficult, or that uses shame, is not. The team should ask whether they would be comfortable explaining each nudge to the person it targets, and to a regulator or the public, because transparency is the practical test of ethical influence. Autonomy, informed consent, and trust are not obstacles to adoption but its foundation (see Topic 1.8 — Data Ethics, Consent & Public Trust), and a manipulative win today spends the trust you need for every future change. A good answer sets a clear standard, applies it to your actual designs, and is willing to remove a nudge that works but cannot be defended openly.

4. Have we distinguished adoption from activation from sustained use — and does anyone here actually own the number that says people are still using this at ninety days? Teams routinely collapse three different behaviours into one word, "uptake", and then celebrate the wrong one. Adoption (signing up) is the easiest to achieve and the most flattering to report; activation (reaching the first moment of real value) is where most services quietly haemorrhage users; sustained use months later is what actually delivers the clinical benefit in the business case. The discussion should confront where your own data stops — many programmes can quote sign-ups but have never instrumented the activation moment or plotted a retention curve, so they cannot even see the week-two cliff, let alone explain it. It should also name an accountable owner: if "adoption" belongs to everyone in general it belongs to no one in particular, and the ninety-day figure becomes an orphan that no dashboard carries. Debate what your definition of "done" really is, because if go-live counts as success the whole system is optimized to stop caring exactly when the hard part begins. An honest answer separates the three behaviours, shows the funnel with real numbers, and points to a named person answerable for sustained use — not a benefits case that assumed it.

5. Are we embedding this behaviour into an existing workflow and cue, or bolting a new tool on beside one and hoping people will remember it? Habits form when a behaviour is tied to a stable, pre-existing cue and repeated, which is why the integration question is a behavioural question and not merely a technical one. The tension to surface is that the easiest thing to deliver — a standalone app or a separate portal with its own login — is often the thing most certain to be abandoned under pressure, because it competes for attention and memory rather than riding on something the user already does. The team should examine, concretely, what existing moment the desired behaviour will attach to: the login that starts the shift, the step before discharge, the morning medication round. It should also be honest about the cost of the alternative, because embedding into the electronic record or the daily workflow is harder, slower, and more political than shipping a separate tool, and that difficulty is exactly why organizations skip it. A good answer identifies the real cue and commits to in-context integration even when a standalone build would be quicker; the weak answer ships beside the workflow and blames the users when the tool is quietly dropped the first busy week.

6. What have we actually tested here rather than assumed — and are our incentives quietly training people to game the metric? Behavioural effects are notoriously context-dependent and frequently wrong-foot even experts, so a plan built on confident assumption rather than a small comparison is a plan carrying hidden risk. The discussion should ask what could be piloted or A/B tested cheaply before scaling — two reminder wordings, two default settings, an assisted route against none — because a two-arm test costs little and routinely changes the answer, and because scaling an untested nudge can entrench something that does nothing or backfires for a subgroup. The conversation must also turn to incentives, financial or performance-target, and ask what behaviour they are really rewarding: attach a payment or a target to a number and people will optimize for that number, sometimes at the expense of the outcome you meant, and the behaviour may collapse the moment the incentive is withdrawn. Consider whether any current target could be met while the underlying clinical intent is missed, which is the signature of gaming. A strong answer names what will be tested and how, treats financial incentives as a short-term catalyst to be monitored for side-effects rather than a permanent prop, and asks of every incentive what breaks when it stops.

In practice: a health & care example#

An integrated care system is rolling out a remote monitoring service for people with heart failure: patients take daily weight and symptom readings on a supplied device, and a virtual ward team reviews them. The technology works in the pilot. Six months in, the programme lead notices the pattern that behavioural science predicts — strong sign-up, a sharp drop after week two, and a small, demographically skewed group still using it by month three.

Rather than commission more patient leaflets, the team runs a COM-B diagnosis. Sign-up (adoption) is fine. The collapse is at activation and sustained use, and the barriers differ by group. For many older patients the constraint is Capability — the device setup has too many steps — and Opportunity — no one prompts them and the daily reading has no anchor in their routine. For the virtual ward nurses, the constraint is Opportunity too: reviewing readings means a second login to a separate portal that is not part of their existing workflow.

The interventions follow the diagnosis. Device setup is de-frictioned to a pre-configured, ready-to-use kit delivered to the home, with a brief in-person or telephone activation call so the first reading — the "aha" moment — happens with support. A daily reminder is timed to an existing cue (morning medication), and a simple feedback message confirms "your reading is in range" or escalates. For staff, the readings are surfaced inside the existing clinical system rather than a separate portal, turning review into a habit attached to the start of the shift. Crucially, the team segments adoption data by age and deprivation, finds the digital path is failing the most deprived quartile, and adds a telephone-based assisted route and carer/proxy access so uptake does not simply track privilege.

They test the reminder wording in a small two-arm comparison before scaling. Success is redefined: not "device issued to 300 patients" but sustained daily use at ninety days, tracked and reported by subgroup, alongside heart-failure admission rates against baseline. The board dashboard now shows behaviour, not deployment — and the equity gap is a headline metric, not a footnote.

Sector lenses#

Startup#

A digital-health startup lives or dies by activation and retention, so behavioural design is not a nicety but the core of the product. Founders should instrument the funnel obsessively — sign-up, activation, day-7 and day-30 retention — and run cheap A/B tests on onboarding, defaults, and prompts, because a small lift in activation compounds into survival. The trap is optimizing for vanity metrics (downloads, registered users) that impress investors while real sustained use quietly leaks away. Startups also carry an outsized ethical risk: growth pressure tempts dark patterns and manipulative retention tricks, which in health can cost you the clinical trust and evidence base you need to sell to the NHS at all. Build honest choice architecture and equity-aware design in from the start, because retrofitting them after a safety or trust incident is far more expensive.

Small business#

An established small provider — a GP practice, a community pharmacy, a care home or domiciliary-care agency, a single-site clinic — carries the same duty to make a new tool genuinely used, but with none of the enterprise's analytics, super-users, or spare capacity to run trials. Here the behavioural work has to be lean and physical rather than instrumented: pick one precise behaviour, strip its friction to the bone, and tie it to a cue that already exists in a small team's day, because there is no one whose job is to chase adoption. Defaults and embedding matter even more than in a large trust, since a five-person team cannot absorb a parallel workflow or a second login and will revert to paper the first busy morning if the tool sits beside their routine rather than inside it. Watch the equity effect with the modest means you have — a care home's residents or a practice's older, more deprived patients are exactly the group a digital-first default can leave behind — and lean on assisted and proxy routes and on the trusted personal relationship a small provider already has. Where capability or evidence is beyond you, adopt the ready-made playbooks, defaults, and assisted-route patterns published nationally rather than inventing your own.

Enterprise#

A large trust or ICS has the scale to do behavioural science properly — real adoption analytics, subgroup segmentation, embedded super-users, and controlled pilots before wide rollout — but also the inertia that lets "deployed" masquerade as "adopted" across thousands of staff and sites. The enterprise challenge is to make adoption an owned, measured outcome rather than an assumption in a benefits case, with someone accountable for sustained use at ninety days and beyond. Integration is the key lever at scale: embedding a behaviour into the existing electronic record and workflow, via single sign-on and in-context prompts, beats any standalone tool or training push. Beware change fatigue and notification overload — with many simultaneous programmes, every extra prompt competes for the same scarce attention, and poorly governed nudges cancel each other out. The enterprise advantage, used well, is the ability to embed behaviour durably through workflow, policy, and audit.

Government#

A national or local public body rarely designs the individual click; it shapes the conditions — standards, funding, defaults, and national services — under which behaviour happens at population scale, which is exactly where nudges bite hardest and equity risks are largest. Government is where the "Nudge Unit" model belongs: rigorous testing (randomized trials), published evidence, and reusable behavioural playbooks that local organizations can adopt. Accountability to ministers, Parliament, and the public raises the ethical bar sharply — a national default or nudge is highly visible, politically sensitive, and, if it is manipulative or widens inequality, corrosive to public trust for years (see Topic 1.8 — Data Ethics, Consent & Public Trust). Government's distinctive duty is to watch the distribution, not just the average: a national digital default that lifts overall uptake while leaving the most deprived behind is a policy failure even if the headline number improves. Its proper role is to make the healthy, equitable behaviour the easy one across a federated system it does not directly line-manage.

Common failure modes#

• Counting deployment as success. Reporting "live in 40 sites" while behaviour never changed. Avoid it by defining done as sustained use at ninety days and measuring it.
• One intervention for every barrier. Answering an Opportunity or Capability gap with more communications. Avoid it by diagnosing with COM-B first.
• Ignoring the activation cliff. Optimizing sign-up while use collapses in week two. Avoid it by designing and measuring the activation moment, not just adoption.
• Nudging the advantaged. A default or slick journey that lifts the average and widens the gap. Avoid it by segmenting adoption data and building assisted routes for those left behind.
• Incentives that game or crowd out. Behaviour that exists only while you pay for it, or is gamed to the metric. Avoid it by favouring intrinsic motivators and testing for side-effects.
• Manipulative design. Dark patterns, hidden opt-outs, exploiting fear. Avoid it by applying a transparency test: could you defend this nudge openly to the person it targets?
• Bolting on, not embedding. A standalone tool beside the workflow, abandoned under pressure. Avoid it by attaching the behaviour to existing cues and systems.
• Assuming instead of testing. Rolling out a nudge no one checked. Avoid it by piloting and A/B testing before scaling.

Maturity model#

Checklist#

• The target behaviour is written down precisely: who, does what, when, where.
• The binding barrier is diagnosed with COM-B (Capability / Opportunity / Motivation) before any intervention is chosen.
• The desired path has been stripped of friction (steps, logins, fields), and defaults are set to the safe, recommended option with an easy, transparent opt-out.
• Reminders and feedback are timed to existing cues and kept sparing.
• The behaviour is embedded into existing workflow and systems to build habit, not bolted on beside them.
• Incentives, if used, are checked for gaming and crowding-out, and for what breaks when they are withdrawn.
• Adoption, activation, and sustained use are measured separately, against a baseline.
• Adoption data is segmented by age, deprivation, language, and disability, and assisted / non-digital routes exist for those the nudge leaves behind.
• Interventions are piloted or A/B tested before scaling.
• Every nudge passes a transparency test: it could be defended openly to the person it targets, a regulator, and the public.
• A named owner is accountable for sustained use past go-live, and the board dashboard reports behaviour, not deployment.

Key sources#

• Susan Michie, Lou Atkins & Robert West, The Behaviour Change Wheel: A Guide to Designing Interventions — the COM-B model and BCW method.
• Richard H. Thaler & Cass R. Sunstein, Nudge: Improving Decisions About Health, Wealth, and Happiness — nudge theory and choice architecture.
• The Behavioural Insights Team ("Nudge Unit") — behavioural insights applied to public policy, including EAST framework (bi.team).
• NICE Guideline NG155, Behaviour change: individual approaches, and PH49, Behaviour change: general approaches (nice.org.uk).
• Everett M. Rogers, Diffusion of Innovations — how innovations spread through a population.
• NHS England — digital inclusion and the framework for reducing digital exclusion (england.nhs.uk).
• Government Communication Service — behavioural science and the OASIS / EAST approaches to public behaviour change (gcs.civilservice.gov.uk).
• See also Topic 8.0 — Phasing in Innovation, Topic 8.1 — Change Management, Topic 8.2 — Communications & Engagement, and Topic 3.3 — Patient & Citizen-Facing Digital Services.

References#

1. Behavioral economics — Wikipedia — https://en.wikipedia.org/wiki/Behavioral_economics
2. Behavior change (public health) — Wikipedia — https://en.wikipedia.org/wiki/Behavior_change_(public_health)
3. Susan Michie — Wikipedia — https://en.wikipedia.org/wiki/Susan_Michie
4. Behavior change method (including the Behaviour Change Wheel) — Wikipedia — https://en.wikipedia.org/wiki/Behavior_change_method
5. Nudge theory — Wikipedia — https://en.wikipedia.org/wiki/Nudge_theory
6. Choice architecture — Wikipedia — https://en.wikipedia.org/wiki/Choice_architecture
7. Behavioural Insights Team — Wikipedia — https://en.wikipedia.org/wiki/Behavioural_Insights_Team
8. Habit — Wikipedia — https://en.wikipedia.org/wiki/Habit
9. Self-determination theory — Wikipedia — https://en.wikipedia.org/wiki/Self-determination_theory
10. Randomized controlled trial — Wikipedia — https://en.wikipedia.org/wiki/Randomized_controlled_trial
11. Diffusion of innovations — Wikipedia — https://en.wikipedia.org/wiki/Diffusion_of_innovations
12. Technology adoption life cycle — Wikipedia — https://en.wikipedia.org/wiki/Technology_adoption_life_cycle
13. The Behaviour Change Wheel: A Guide to Designing Interventions — Susan Michie, Lou Atkins & Robert West, Silverback Publishing (2014) — http://www.behaviourchangewheel.com/
14. Nudge: Improving Decisions About Health, Wealth, and Happiness — Richard H. Thaler & Cass R. Sunstein, Yale University Press (2008) — https://en.wikipedia.org/wiki/Nudge_(book)
15. EAST: Four Simple Ways to Apply Behavioural Insights — The Behavioural Insights Team — https://www.bi.team/publications/east-four-simple-ways-to-apply-behavioural-insights/
16. Behaviour change: individual approaches (NG155) — National Institute for Health and Care Excellence — https://www.nice.org.uk/guidance/ng155
17. Diffusion of Innovations, 5th edition — Everett M. Rogers, Free Press (2003) — ISBN 978-0743222099
18. Inclusive digital healthcare: a framework for NHS action on digital inclusion — NHS England — https://www.england.nhs.uk/long-read/inclusive-digital-healthcare-a-framework-for-nhs-action-on-digital-inclusion/