Inside Sylva: how clinical evidence drives everything that we do

Mental health research and practice has run on the same playbook for fifty years: label a condition, then assign the prescribed fix. Diagnose depression, prescribe a Selective Serotonin Reuptake Inhibitor (SSRI) or Cognitive Behavioral Therapy (CBT); diagnose anxiety, assign Exposure Therapy; diagnose ADHD, start stimulants. The assumption is that these diagnostic categories represent distinct disorders with distinct causes, and that the right protocol will reliably produce good outcomes.

That assumption has not held up. In 2013, Thomas Insel, then-director of the US National Institute of Mental Health (NIMH), noted that the Diagnostic and Statistical Manual of Mental Disorders (DSM) fundamentally lacks validity because its categories rely on symptom clusters rather than objective biological measures, prompting the NIMH to shift its entire research focus away from DSM classifications (Insel, 2013).

By 2020, Hayes, Hofmann, and Ciarrochi concluded in Clinical Psychology Review that the "protocol-for-syndrome" approach had failed its primary purpose. Decades of packaging treatments around what they termed "hypothesised but unproved latent disease entities" simply did not deliver the clinical utility once promised.

The evidence is clear. Leichsenring et al. (2022), in a landmark World Psychiatry review co-authored with John Ioannidis, analysed 102 meta-analyses spanning 3,782 RCTs and 650,514 patients. The finding: effect sizes across all disorders and treatments were small: SMD 0.34 for psychotherapies, 0.36 for pharmacotherapies. Thousands of trials and enormous research investment have produced modest, stalling returns. The authors called for a paradigm shift.

Sylva was built in response to exactly this problem. Rather than the protocol-for-syndrome model (match a diagnosis to a treatment manual), Sylva is grounded in process-based therapy, the alternative framework developed by Hayes and colleagues. The question it asks isn't "what diagnosis does this person have?" but "what psychological processes of change matter for this person?" Every Sylva mentor is built around that question, which is why the clinical architecture here looks nothing like a conventional digital CBT programme.

A million people waiting. Almost no support while they wait.

Nearly one million people in England are on open referrals for neurodevelopmental assessment. As of December 2025, up to 735,157 were waiting for an ADHD assessment (NHS England Digital, February 2026). As of March 2026, a further 270,701 were waiting for autism assessment, with 90% waiting longer than the NICE-recommended 13 weeks (NHS England Digital, May 2026).

These figures are almost certainly undercounts: the autism data excludes community paediatric services entirely, and many GPs have stopped adding patients to lists they know will take years to clear.

During these months and years of waiting, families receive little or no structured support. 74% of parents and carers of autistic children report clinically significant stress levels (Crane et al. 2016, Autism). NICE-recommended parenting programmes, such as Incredible Years, Triple P and Stepping Stones, reach only 4 to 18% of eligible families.

The same problem runs through the whole system

The gap extends beyond neurodevelopmental services. NHS Talking Therapies received 1.8 million referrals in 2023/24. Only 37% completed treatment. Only 17% of all referrals resulted in reliable recovery (Bagri 2024, Nuffield Trust), a proportion that has barely changed in a decade.

The Nuffield Trust analysis identifies structural reasons for this attrition. 16% of referrals were never assessed. 20% were deemed unsuitable or terminated. 23% dropped out before completing. Treatment is heavily concentrated in CBT, which accounts for a third of all appointments, while psychodynamic psychotherapy accounts for just 0.2%. One in five appointments is spent on signposting and psychoeducation rather than therapy. Lack of treatment choice has been identified as a likely driver of the high dropout rate.

Sylva was designed to sit in these gaps: available from the moment someone is struggling, through assessment, diagnosis and beyond, offering the breadth of support that the current system does not. It is not a replacement for clinical care. It is support for the years when clinical care is unavailable, and a wider range of approaches than most people will ever be offered.

What Sylva does: the three pillars of understanding & support

Sylva is on the pathway to become a regulated UK medical device (standalone software, Class I), with a designated safeguarding lead, a clinical safety officer, and a clinical risk management process informed by DCB0129 principles. It delivers three things.

First, more than 65 scientifically validated screening questionnaires covering autism, ADHD, common co-occurring conditions (anxiety, depression, OCD, sensory processing, PDA and others), and carer and child wellbeing outcomes. Instruments are selected by age band and can be repeated over time, allowing people to track their own progress on a given construct or process of change.

Second, a strengths-and-needs profile that synthesises screening results and the person's own context to personalise the support Sylva provides.

Third, AI-powered conversational support through a team of specialist mentors, each grounded in therapeutic modalities with systematic review or meta-analysis level evidence. This document focuses on that third component: how the mentor system was designed and what science underpins it.

An independent peer-reviewed study by NHS clinical psychologists (Brown et al. 2026, Scientific Reports, Nature) evaluated Sylva's AI mentors and found a mean rating of 4.8/5 for clinical appropriateness, accuracy and safety.

How we designed the clinical content

Every piece of content Sylva delivers traces back to clinical evidence through a four-layer framework, where each layer builds on the one below it.

Layer 1: NICE guidance

The starting point was current NICE guidance for neurodevelopmental conditions: CG128 (autism in adults), CG142 (autism in under-19s, recognition and referral), CG170 (autism in under-19s, support and management), and NG87 (ADHD, diagnosis and management). These guidelines define what approaches are recommended for the target population and establish the clinical standard of care against which any intervention, including a digital one, should be measured.

Layer 2: Systematic reviews and meta analyses

NICE recommendations point to broad categories of intervention. We went deeper. For each therapeutic modality Sylva uses, we conducted a structured literature review to identify the highest-quality evidence, using AMSTAR-2 and ROBIS tools for critical appraisal.

The result is a library of 49 in-house literature reviews covering more than 20 modalities, referencing over 460 systematic reviews and meta-analyses across 1,738 primary sources. These are not marketing collateral. They are controlled regulatory documents maintained in our technical file and updated as new evidence emerges.

Modalities with the strongest evidence base for this population include behavioural parent training (Incredible Years, Triple P), Acceptance and Commitment Therapy (ACT), emotionally focused therapy (EFT), and short-term psychodynamic psychotherapy, each supported by multiple systematic reviews or meta-analyses confirming efficacy in relevant populations.

Layer 3: Behaviour change techniques

Knowing that a programme works is necessary, but not sufficient. We also need to know what within it works. The Behaviour Change Technique Ontology (BCTO; Marques et al. 2024, Wellcome Open Research) and its predecessor BCTTv1 (Michie et al. 2013) provide a standardised vocabulary for coding the active ingredients of interventions.

We mapped the BCTs delivered by the evidence-based programmes cited above. Each Sylva mentor is then designed to deliver the BCTs appropriate to its therapeutic domain.

Layer 4: Mediators of change

The final layer asks: through what psychological mechanisms do these techniques produce change? Hayes et al. (Evolving an idionomic approach to processes of change, 2022) synthesised 55,000 RCTs to identify replicated mediators of therapeutic change across psychological interventions. It is the most systematic mapping of why therapy works that currently exists in the literature.

The design logic runs as follows: NICE recommends the approach; systematic reviews confirm the modality works; BCT coding identifies the active ingredients; mediator research explains why those ingredients produce change. Sylva's mentors deliver the ingredients and target the mechanisms.

Six mentors. More modalities than any single clinician covers.

In traditional practice, therapeutic modalities are strictly siloed. A clinical psychologist trained in CBT rarely delivers emotionally focused therapy, psychodynamic interventions, or behavioral parent training. Similarly, a family therapist is seldom trained to provide Acceptance and Commitment Therapy (ACT) for parental burnout. These approaches belong to separate training pathways, distinct professional identities, and isolated supervision models.

Sylva's mentor team spans all of these modalities through a single access point:

This combination of modalities has never been integrated into a single intervention before. Individual programmes like Incredible Years or Triple P focus on behavioural parenting. ACT-based interventions target psychological flexibility. EFT targets relational processes. Sylva brings them together because families do not arrive with a single, neatly categorised need.

A parent dealing with a child's school refusal may need parenting strategies (Hazel), support for the emotional toll on their relationship (Luna), help managing their own burnout (Fern), and space to process grief about the life they expected versus the life they have (Rowan). In clinical practice, that would require referrals to multiple professionals across multiple waitlists.

In Sylva, it is one app.

Safety, security and governance

Sylva meets the NHS Digital Technology Assessment Criteria (DTAC) across all five domains. For clinicians considering whether to recommend Sylva to families, here is what that means in practice.

Clinical safety

Sylva's clinical safety framework is informed by DCB0129, the NHS standard for clinical risk management of health IT. A designated clinical safety officer holds sign-off authority over all clinical content and safety decisions. A clinical hazard log identifies, scores and mitigates hazards specific to AI-delivered psychoeducational content, including inappropriate advice, safeguarding detection failures and over-reliance.

Safeguarding

Sylva includes dedicated safeguarding policies for both adults and children and young people, with a designated safeguarding lead. The AI safeguarding subsystem monitors conversations for safeguarding concerns and signposts to appropriate crisis and support services. Each mentor operates within defined scope boundaries: it knows what it is permitted to discuss and what it must decline or escalate. Conversations containing safety-related content are flagged for human review and, where appropriate, escalation.

Data protection

Sylva is registered with the ICO and compliant with UK GDPR. A data protection impact assessment covers all modules. Data storage follows a privacy-by-design model, with primary storage on the user's own device. A published privacy policy, data subject access request procedure and data breach notification policy are all maintained.

Technical security

Sylva holds ISO 27001:2022 certification (externally audited), DSPT and current Cyber Essentials certification. Penetration testing is conducted annually by an independent provider, with findings remediated.

Usability & accessibility

Usability testing has been conducted with the target user population per IEC 62366-1. A WCAG 2.1 AA accessibility audit has been completed by an external assessor. The product is designed for adult lay users with no clinical training, using plain language throughout and avoiding clinical jargon in all user-facing content.

What this means for clinicians recommending Sylva

Sylva does not diagnose. It does not prescribe. It does not replace clinical care. These are not marketing disclaimers. They are hard-coded into the intended purpose statement of the medical device and enforced through the technical architecture of every mentor. When you recommend Sylva, you are recommending a product that has been through the same governance framework the NHS applies to any digital health technology it would consider adopting.

There for families when the system isn't

The strongest clinical intervention in the world is useless if a family cannot access it. Sylva is available 24 hours a day, 365 days a year, with no waitlist, no referral, and no geographical restriction within the UK. A parent at 2am, exhausted after a three-hour meltdown, can open Sylva and receive research-informed support grounded in the same evidence base as the face-to-face programmes they cannot access.

This is not a replacement for those programmes. It is support for the reality that most families will never reach them. When 4 to 18% of eligible families access a NICE-recommended parenting programme, and 83% of NHS Talking Therapies referrals do not result in reliable recovery, the question is not whether digital delivery is ideal. The question is whether leaving the majority with nothing is acceptable.

What we do not claim

Honesty about the limits of our evidence is part of our clinical governance.

We do not claim that Sylva produces equivalent outcomes to face-to-face therapy. Demonstrating that requires a powered clinical investigation with validated outcome measures, which is in progress. Our post-market clinical follow-up programme targets 134 completers with six validated instruments measured at baseline, four, eight and 12 weeks. Until that investigation reports, our outcome claims are limited.

We do not claim that AI-delivered BCTs are identical in quality to facilitator-delivered BCTs. Technique detection is not technique quality. A trained therapist brings clinical judgement, timing and relational warmth that an AI mentor cannot fully replicate. What Sylva can do is deliver the right techniques, at scale, to people who would otherwise receive no structured support at all.

We do not claim to replace multidisciplinary teams. When clinical input is available, Sylva is designed to work alongside it, not instead of it.

What we do claim

Sylva was designed with the same rigour you would expect from a clinical intervention: starting from NICE guidance, grounded in systematic reviews, structured around behaviour change techniques with demonstrated efficacy, and targeting the psychological mediators that drive real outcomes.

That is the standard we hold ourselves to, and the standard against which we invite scrutiny.

Key references

Bagri S (2024). Does the NHS Talking Therapies service have an attrition problem? QualityWatch, Nuffield Trust and Health Foundation.
Brown E, Patel S, Williams R et al. (2026). Independent clinical evaluation of an AI parenting mentor for neurodevelopmental conditions. Scientific Reports, Nature. Peer-reviewed.
Crane L, Chester JW, Goddard L et al. (2016). Experiences of autism diagnosis: a survey of over 1000 parents in the United Kingdom. Autism, 20(2), 153-162.
Darzi A (2024). Independent investigation of the National Health Service in England.
Hayes, Ciarrochi, Hofmann, Chin & Sahdra (2022), Behaviour Research and Therapy, 156, 104155. Evolving an idionomic approach to processes of change.
Hayes SC, Hofmann SG, Ciarrochi J (2020). A process-based approach to psychological diagnosis and treatment: the conceptual and treatment utility of an extended evolutionary meta model. Clinical Psychology Review, 82, 101908.
Insel T (2013). Transforming diagnosis. NIMH Director's Blog, 29 April 2013.
Leichsenring F, Steinert C, Rabung S, Ioannidis JPA (2022). The efficacy of psychotherapies and pharmacotherapies for mental disorders in adults: an umbrella review and meta-analytic evaluation of recent meta-analyses. World Psychiatry, 21(1), 133-145.
Marques MM, Carey RN, Norris E et al. (2024). The behaviour change technique ontology. Wellcome Open Research, 9:298.
Michie S, Richardson M, Johnston M et al. (2013). The behavior change technique taxonomy (v1) of 93 hierarchically clustered techniques. Annals of Behavioral Medicine, 46(1), 81-95.
Sigurdardottir S, Leijten P et al. (2025). Mechanisms of change in parenting programs: an individual participant data meta-analysis. N=3,252.
NHS England Digital (2026). ADHD management information, February 2026.
NHS England Digital (2026). Autism waiting time statistics: referral to care contact, April 2025 to March 2026.