Perfectionism and all-or-nothing thinking
Helps people treat a slip as a normal part of change and get back on track, so one bad day does not end the whole effort.
What we do
Three ways to work with us, all in GLP-1 and obesity care.
Sacher AI is an independent behavioural AI and evidence partner for GLP-1 and obesity care. Teams come to us to decide what to build, to design and deliver it, or to test and prove that it works.
Offer 01
Decide
Find out whether an idea is worth building before you commit the budget. A fixed fee Evidence Phase of 8 to 12 weeks tests the idea using the Before You Build Method, and ends with a clear recommendation to go ahead, rework or stop.
In GLP-1 and obesity care
Whether an AI support service would actually keep more people on treatment through the first months, who would deliver it, and who would pay for it, answered before the budget is committed.
What it covers
- The five uncertainties: need, workflow, action logic, commercial viability, and regulation
- Assumptions tested with real users and the people who would have to act on the output
- A decision-ready pack: recommendation, evidence, and the assumptions still open
How we work
Conversations across the care pathway
We speak to organisations across the care pathway, such as telehealth services, pharmacies and digital health companies, to test whether a concept fits how care is delivered in practice.
Midpoint review
Halfway through, we review what we have learned with you. Together we decide whether to carry on as planned, change direction or stop early.
Evidence trail
Every finding in our recommendation links back to its source: who said it, where and when. Your team can check how we reached each conclusion.
Offer 02
Design and deliver
Turn an idea into a tested prototype or MVP, working alongside your own team. We cover the technical planning, product, design, behavioural logic and conversation design. Your team keeps the production build.
In GLP-1 and obesity care
An AI coach that helps people on GLP-1 treatment keep up the eating, activity and routines that protect long-term results. Anything clinical, including side effects and dosing, goes straight to their clinical team.
What it covers
- Technical planning, product and design
- Behavioural architecture, conversation design, and prompt strategy
- Delivery alongside your product and engineering teams
- Testing and iteration against clinical and behavioural requirements
- A technical plan your engineering team can take into production
How we work
Synthetic-data demonstrator
We build an interactive demonstrator on synthetic data. Your team and stakeholders can try something that works, without using any patient data.
We don't sell a patient platform, and the production build stays with your team.
Support designed around how people think, feel, and act
Two people on the same medication can need completely different support. Behavioural phenotyping identifies the drivers behind each person's eating, activity, and adherence, and turns them into tailored support. Three examples:
Emotional eating
Helps people spot what sets off their eating, and tell physical hunger apart from eating to cope with stress or low mood.
Social and environmental pressure
Helps people plan what to say and do in the social situations that most often throw them off course.
RAPID: a disciplined path from idea to a tested prototype
Research-Assisted Prototyping and Iterative Development. Each idea moves through structured, auditable stages, with behavioural science, clinical thinking, and governance built in from the start.
- 01
Discovery and research
We select the right subject-matter experts from our team and network, review the science and the market, and turn an ill-defined business problem into a well-defined AI problem, with a model design and success metrics.
- 02
AI model design and prototyping
Our engineers build the solution: choosing a foundation or custom model, engineering prompts, fine-tuning where useful, and implementing guardrails to meet your AI safety standards.
- 03
AI metrics evaluation
We evaluate the model with our own tools, including behavioural-science persona simulators, iterating against domain-specific metrics and safety assessments.
- 04
UI design and prototyping
A software team builds an MVP web or mobile app with a simple interface, so real users can engage with the model, with a technical plan your engineering team can take into production.
- 05
MVP user testing
We put the prototype in front of small user cohorts, gather and analyse qualitative feedback, and feed it into the next iteration, with ongoing support available.
Offer 03
Test and prove
Show that your AI is safe and that it works. We test it with PromptSafe, evaluate it independently using the FAST framework, and can take the research through to peer-reviewed publication.
In GLP-1 and obesity care
Testing how an AI assistant responds when someone on GLP-1 treatment says they are skipping meals, feeling unwell, or thinking of stopping, before any patient sees it.
Right answer, wrong action
A patient tells your AI they have been skipping meals and the weight is coming off faster. The AI replies: that is great, well done for staying focused. Nothing in that answer is factually wrong. But for someone with a history of disordered eating, it has just encouraged the thing your programme exists to prevent. Accuracy testing will not catch it, because there is nothing inaccurate to catch.
Behavioural
Usually missing
What is the conversation doing to the person?
Clinical
Is the information accurate and in scope?
Technical
Does the system work and stay reliable?
What it covers
- PromptSafe pre-deployment testing across realistic and adversarial conversations
- The FAST framework: fidelity, accuracy, safety, and tone
- Independent evaluation before or after launch, with audit-ready evidence
- Research from study design through to peer-reviewed publication
How we work
Support, not clinical decisions
Our behavioural AI supports people between appointments. Clinical decisions, such as whether to stop, pause or switch treatment, stay with the person's own clinician. We design that boundary in from the start.
Concept research with people on treatment
We test concepts with people on treatment in short, structured interviews, through an independent recruitment partner, with the interview guide and screens agreed with you before fieldwork starts.
Clinical and behavioural safety, from the start
No AI is risk-free. What we can do is design for that risk, evidence it, and keep qualified people in the loop.
Human in the loop
AI assists with drafting, analysis, and flagging. Qualified people review and approve, and remain responsible for clinical and operational decisions.
Retrieval-based by default
Systems draw on approved content rather than generating free-form clinical advice, with clear escalation to a clinician when needed.
Auditable by design
Interactions are logged so teams can review inputs, outputs, and escalation, and evidence their oversight.
Aligned with recognised standards
We design to align with MHRA and CQC expectations, MHRA's Good Machine Learning Practice, and ISO/IEC 42001, and work closely with your regulatory affairs and quality assurance teams where a system may be a medical device.
Next step
Let's talk about what you are about to build.
Book a discovery call and we will work out whether the right first step is Decide, Design and deliver, or Test and prove.