01
It treats behaviour, not information alone, as the thing that drives outcomes.
Behavioural AI consulting and product partner for GLP-1 and digital health
Health AI that patients actually trust, use, and stick with.
Most health AI struggles on behaviour, not technology: people don't engage with it, trust it, or keep using it. Behavioural AI is the layer that changes that, the tone, timing, and support that make AI shift real patient behaviour. We help GLP-1 and digital health teams design, evaluate, and safely deploy it to improve adherence, retention, and outcomes.
26 years working across healthcare, academia, and global corporate settings
27 publications and presentations at 49 national and international conferences
Built patient-facing AI systems for behaviour change across digital health and obesity care
Clients and partners
Proud to work with leading teams building and scaling in digital health and obesity management.
The GLP-1 problem
GLP-1 medicines work, but only while patients stay on them.
More than half of patients stop within the first year, mostly for behavioural reasons: difficult side effects, unrealistic expectations, waning motivation, cost, and the everyday barriers that make lasting change hard. Improving adherence, persistence, and long-term maintenance is where much of the remaining value in GLP-1 care now sits.
That is a behaviour problem, not just a clinical one, and it is exactly what behavioural AI is built to solve.
The signature idea
What is behavioural AI?
Behavioural AI is the layer that helps AI systems influence behaviour safely and effectively in real-world health settings. It is the decision logic that determines what support someone needs, when they need it, and how that support should be delivered to fit the moment.
02
It adapts delivery to the patient: tone, timing, framing, intensity, and support style.
03
It links AI design to adoption, retention, adherence, safety, and operational performance.
Explored in our peer-reviewed paper, The missing discipline in AI: a call for behavioural science (Wellcome Open Research, 2026).
Built on real-world experience, not theory. The behavioural methods we design into AI were developed and evidenced with real patients: in a study of the text-based coaching we led at Allurion, 77% of patients said it helped them reach their weight-loss goals, with improvements in eating habits, activity, and wellbeing at six months (Canadian Obesity Summit, 2023).
Mission
Built for teams where behaviour, trust, and safety are part of the product problem.
Our mission is to design, evaluate, and deliver behavioural AI solutions that drive meaningful behaviour change, support clinical and wellbeing outcomes, and hold up in real-world deployment.
Impact at scale
22
countries with patient-facing AI deployed
500k+
GLP-1 patients in programmes we have supported
80
obesity clinics reached internationally
2
of the largest patient-facing AI systems in obesity
Independent by design
Advice shaped by what will work, not by what we need to sell next.
We do not own a patient-facing platform or care app, so our advice on your product is based on what gives it the best chance of adoption, workflow fit, safety, and long-term value. Our own product, PromptSafe, is an AI testing tool, not a patient product.
Our point of view
Better models help. Better implementation decides what happens next.
We treat health AI primarily as an implementation challenge, not just a prediction challenge. The real questions are whether teams adopt it, whether it fits live workflows, and whether it improves persistence, trust, and outcomes in practice.
Outcomes we improve
The outcomes that matter in digital health.
01
Retention and lifetime value
Reduce early churn, keep patients engaged for longer, and convert programme completers into longer-term users.
02
Patient engagement and adherence
Increase engagement, support more consistent adherence to treatment, and strengthen habit formation across populations.
03
Clinical and behavioural safety
Reduce risk through proactive monitoring, stronger triage support, and systems built to clinical and behavioural safety standards.
04
Patient profiling and personalisation
Understand what each patient actually needs and adapt support to barriers, behaviours, motivation, and progress.
05
Operational efficiency and support performance
Automate high-volume support work, reduce manual burden, and improve consistency without scaling headcount linearly.
Selected case studies
Proof that connects strategy, delivery, and outcomes.
GLP-1 at scale
Reducing churn and operational pressure in digital obesity care
Worked on AI and behavioural support models for a large UK GLP-1 provider facing high early churn, fragmented systems, and support demand rising into the hundreds of thousands of contacts per month.
Focused on retention as a major growth lever, alongside proactive support, operational efficiency, and stronger clinical consistency.
Learn more →AI agent build
Behaviour change agent delivered in eight weeks
Designed and delivered a behaviour change and intuitive eating coach for Holly Health, grounded in COM-B, advanced behaviour change techniques, and structured PromptSafe testing.
The MVP was delivered on time, ready for real user testing, and led directly to continued client engagement.
Learn more →Research and evidence
From evaluation frameworks to publishable behavioural research
Built formal evaluation and research capability including FAST, the MBQ phenotyping work, and broader clinical and behavioural evidence programmes with academic and commercial partners.
Strengthened regulatory credibility, investor confidence, and external proof through frameworks, abstracts, and peer-reviewed publication.
Learn more →When teams bring us in
The moments where Sacher AI tends to create the most value.
01
You have AI ideas, but no clear priority
We help leadership teams turn competing ideas into a focused direction with clear use cases, sequencing, ownership decisions, and decision points.
02
You are building a patient-facing AI system and want it to work in the real world
We add the behavioural, clinical, and workflow thinking that makes a patient-facing AI system more useful, safer, and more likely to be adopted in practice.
03
You need stronger evidence, oversight, or launch confidence
We design testing, evaluation, and research that surfaces failure modes early and gives you clearer proof for governance, investors, partners, and future deployment decisions.
Client feedback
What our clients say
Paul played a foundational role in shaping strategy across AI, behavioural medicine, and research. His impact at Numan has been profound and lasting.
Jamie Smith Webb
CTO, Numan
Sacher AI quickly felt like an extension of our team. User safety was prioritised and the operational tests were really thorough.
Grace Gimson
Co-founder and CEO, Holly Health
Paul's transformative work in creating our behaviour change programme has had astonishing results. Our patients have been able to sustain 95% of their weight loss even 12 months after the programme's end.
Benoit Chardon
Chief Commercial Officer, Allurion Technologies
Our product
PromptSafe
Find out where your AI agent fails users, before your users do.
Pre-deployment testing for conversational AI agents. Run your agent across hundreds of realistic and adversarial simulated conversations before launch, see exactly where it fails, and get specific fixes.
Watch the 2-minute overview and start a free trial, no card required.
- Realistic personas you define, plus adversarial personas we generate.
- Evaluators that are plain-language checklists of observable behaviours, not abstract scores.
- Audit-ready, exportable reports for governance review.
What we do
Services across the full AI lifecycle
GLP-1 and behaviour change
The behavioural layer for GLP-1 and obesity care: personalised support that improves adherence, retention, and long-term outcomes at scale.
Find out more →
AI agent design and delivery
End-to-end design and delivery of patient-facing AI agents, grounded in behaviour change science, workflow fit, and clinical and behavioural safety.
Find out more →
AI safety and evaluation
PromptSafe pre-deployment testing and the FAST evaluation framework, giving audit-ready evidence before and after launch.
Find out more →
Research and evidence
Peer-reviewed research that generates credible evidence of effectiveness, safety, and impact.
Explore research →
How we work
A clear, low-risk way to start.
01
Start with a conversation
A short discovery call to understand where you are and what you are trying to change.
02
Scope the right first step
We recommend the simplest useful starting point: strategy, a build, evaluation, or research.
03
Work as an extension of your team
Hands-on delivery alongside your people, not slides handed over from the outside.
04
Prove it
Evidence, safety, and outcomes you can stand behind, designed in from the start.
From the blog
Latest thinking on behavioural AI in health.
Digital health
Where the real value is in AI for GLP-1, beyond the chatbot
When people picture AI in GLP-1 care they picture a chatbot. The chatbot is real and useful, but the value has moved to agents, real-time data, and the gap between engaging with an app and changing behaviour.
Read article →Behavioural science
The behavioural layer: why the drug or device is never enough
GLP-1 medicines and devices work. Sustained outcomes come from the behavioural layer around them, the support, content, and design choices that help patients change and hold that change.
Read article →Safety
Right answer, wrong action: behavioural safety in clinical AI
Patient-facing AI can be clinically correct and still make a patient more likely to do the wrong thing. Behavioural safety is the layer most teams are missing.
Read article →