Why Your Users Are Ghosting You (And Why It’s Not Their Fault)

The digital therapeutics attrition problem isn’t a motivation problem. It’s a design, measurement, and system problem.

By Mat Rawsthorne

I recently hosted a coffee chat for the One HealthTech Mental Health Network on what I’ve started calling the “ghosting problem” in digital therapeutics. The room included founders, NHS consultants, user researchers, a health psychologist, and people building everything from VR mental health tools to ADHD screening platforms. A common theme quickly emerged – even if you build something that works, many people try it and then silently disappear.

That vanishing act isn’t new. The editor of JMIR flagged digital attrition over twenty years ago. But what struck me about this conversation, and the recent wave of academic research on the topic, is that we’re still largely asking the wrong question. We keep asking why don’t users stick around? when we should be asking what are we measuring, what are we building, and who are we actually designing for?

We’re counting the wrong things

One of the most energising threads in our coffee chat was a shared frustration with behavioural metrics. Clicks, logins, streaks, time-on-app, these are the industry’s default measures of success. But as one participant put it, they don’t reflect the emotional or psychological shifts that someone experiences when using a mental health tool. You can track every tap and still have no idea whether someone’s life is getting better.

The academic literature backs this up forcefully. Cipriani and colleagues (2025), writing in the British Journal of Psychiatry, argue that fewer than 5% of mental health app users open the app after two weeks, but they also point out that this statistic is almost uninterpretable, because the field has no consensus on what “engagement” even means. Their proposed solution is to reframe engagement as a process rather than an outcome, drawing an analogy to pharmacology: just as a drug’s efficacy depends on how it binds to a receptor, a digital tool’s impact depends on how it engages specific psychological mechanisms. Counting clicks tells you about as much as weighing a pill.

Gupta, Bharti, and Singhal (2026) take this further with their Dose-Content-Disposition framework, arguing that digital exposure isn’t one-dimensional. The same amount of screen time can be beneficial or harmful depending on what someone is doing and who they are. Active, goal-oriented engagement looks nothing like passive scrolling, and individual vulnerability moderates everything. This maps directly onto what we heard in the coffee chat: lumping all users into a single attrition curve hides the people who got what they needed in one session alongside those who were failed by the design.

The system is the barrier, not the user

One of the most striking moments in our session came from a contributor working with an NHS remote monitoring platform for severe mental illness. Their research showed that when a digital intervention was proactively offered to eligible patients, with a genuine option to decline, about 50% said no. But of those who said yes, adherence sat at 75% over twelve months. That’s an extraordinary number. And it tells us something important: the problem often isn’t that people won’t use digital tools. It’s that the systems around those tools aren’t set up to support them.

Linardon (2026), writing in World Psychiatry, makes a compelling case that clinicians are central to this equation. Premature dropout typically occurs when patients have unclear expectations, limited understanding of how a digital tool fits into their care plan, or no early support to troubleshoot when things feel confusing. He advocates for front-loading clinician support through structured early check-ins and intentional intake dialogue, essentially, treating digital onboarding with the same care we’d give to starting a new medication.

Our coffee chat echoed this. Multiple participants pointed to the role of digital navigators, dedicated staff who support onboarding and monitoring, as the single most effective lever for sustained engagement. Frontline clinical staff simply don’t have the capacity to learn and champion a new platform on top of their existing caseload. But having a navigator layer means the intervention reaches patients without requiring clinicians to change their workflow overnight. And crucially, the data from a meta-analysis of 117 trials (Zainal et al., 2025) shows that guided interventions consistently outperform self-guided ones on engagement, with the therapeutic relationship between user and coach being a significant predictor of completion.

What people actually value keeps surprising us

Perhaps my favourite story from the session came from a participant who’d built a digital platform for people transitioning out of secondary mental health services in the NHS. They integrated clinical data, mood diaries, and a range of tools. The most popular feature? A simple document storage function where users could keep digital copies of their benefits and housing letters. Not the clinical content. Not the outcome measures. The admin tool that made their life a bit less chaotic.

This story captures something the literature increasingly confirms. Herpertz and colleagues (2026) analysed Germany’s prescribed digital therapeutics (DiGAs) and found that mental health apps had the poorest follow-up adherence of any category, with re-uptake rates falling below 15%. Their comparison with commercial wellness apps revealed a stark functional gap: prescribed apps leaned heavily on rigid psychoeducational content while commercial apps used gamification, wearable integration, and native mobile design, features that meet users where they are rather than demanding they come to the content.

But gamification isn’t a silver bullet either. Ngabo-Woods and colleagues (2026) identify what they call the Engagement-Efficacy-Ethics Trilemma: the uncomfortable reality that maximising engagement metrics doesn’t consistently translate to better therapeutic outcomes, and can introduce ethical risks around manipulation and autonomy. Designing for dopamine hits and designing for psychological growth are sometimes in direct tension.

Rethinking what success looks like

One of the most useful reframes from our conversation was the question: are we supporting meaningful change in people’s lives, or are we just tracking whether they’re engaging with our tech? That distinction matters enormously. Smith, Cipriani, and colleagues (2025) note that attrition can itself be a marker of success, a concept they call “e-attainment,” where users stop engaging because they’ve internalised what they needed. If someone completes a single VR session and carries the insight forward, that’s not a failure of retention. It’s the intervention working.

Our session surfaced a parallel anecdote: a GP practice needing to bring out a shredder every time they trim the hedgerow, because it’s stuffed with prescriptions patients accepted to be polite and then immediately discarded. People accept things to please their doctor. They download apps because they were told to. Compliance isn’t the same as engagement, and engagement isn’t the same as benefit.

So, what should we be measuring instead? Our group landed on a few principles. First, track whether the user’s intent was met at each touchpoint, not just at sign-up and study-end. Second, take self-reported outcomes seriously: if a hundred users consistently report fewer A&E visits since using a platform, that’s meaningful evidence, even without system-level data. Third, design for flexibility. Let people use the subset of features that fits their life. The most robust engagement isn’t built through prescriptive pathways; it’s built by offering genuine value at every possible entry point.

Where do we go from here?

If I had to distil our session and the current evidence into a handful of suggestions for the field, they’d be these:

Stop treating engagement as a single number. A login count tells you almost nothing. We need to define what engagement means for each specific tool, set thresholds that are clinically justified, and link those thresholds to outcomes that matter to users, not just to funders.

Invest in the human layer. Digital navigators, trusted referrals, early check-ins, these are not nice-to-haves. They are the infrastructure that makes digital interventions work. The evidence is clear - accountability to another person drives behaviour change more reliably than any feature on a screen.

Design for the person, not the pathway. NHS funding is siloed by diagnosis. Commissioner priorities don’t always match user needs. But the tools that sustain engagement are the ones that solve problems people actually have, sometimes in ways the designers never anticipated.

Accept that some people are done, and that’s okay. Not every dropout is a failure. Some of them are successes we haven’t learned to recognise yet. Our measurement frameworks need to catch up with that reality.

The ghosting problem isn’t going to be solved by making apps stickier. It’s going to be solved by being more honest about what we’re building, who we’re building it for, and what we’re willing to measure. That conversation has started. It’s time to keep it going.

---

References cited:

Cipriani, A. et al. (2025). Beyond counting clicks: rethinking engagement in digital mental health. The British Journal of Psychiatry.

Gupta, L., Bharti, D., & Singhal, S. (2026). Not all clicks are equal: digital dose, content, and user disposition in mental health. Academia Mental Health and Well-Being.

Herpertz, J. et al. (2026). From prescription to practice: Uncovering missing ingredients for user engagement in prescription digital therapeutics. Current Treatment Options in Psychiatry.

Linardon, J. (2026). Rethinking the prediction and prevention of dropout in digital mental health. World Psychiatry, 25(2).

Ngabo-Woods, H. et al. (2026). Gamification in digital mental health interventions: A systematic review of the engagement–efficacy–ethics trilemma. Information, 17(2).

Smith, K.A. et al. (2025). Engagement and attrition in digital mental health: current challenges and potential solutions. npj Digital Medicine.

Zainal, N.H. et al. (2025). What factors are related to engagement with digital mental health interventions? A meta-analysis of 117 trials. Health Psychology Review.