Why GLP-1 providers lose patients early, and where AI can help

Most GLP-1 providers are thinking hard about acquisition. I think the bigger long-term question is retention.

As the market becomes more competitive, price will keep getting squeezed. Patients will have more choice, medication options will expand, and switching will become easier. In that environment, the providers that win will not simply be the ones that acquire the most patients. They will be the ones that keep patients engaged long enough to benefit.

That was one of the main themes I discussed with Dr Ashwin Sharma on the GLP-1 Digest podcast. The question we kept coming back to was simple: where does AI create real value for GLP-1 providers?

My view is that the value is not in adding another chatbot. It is in building a behavioural intelligence layer around treatment.

GLP-1 medications work, but the medication is only one part of the intervention. In the trials, these drugs are not studied in isolation. They are delivered alongside behavioural and lifestyle support. In the real world, and especially in direct-to-consumer care, that support is often thinner, more reactive, and harder to personalise at scale. That creates a retention problem.

Patients do not usually disengage at random. There are predictable points in the journey where the risk of churn increases.

The first is onboarding. The early weeks are fragile. Patients are adjusting to treatment, worrying about side effects, and trying to work out what is normal. Many have already seen alarming stories online, and some are anxious before they have experienced a side effect at all. If support is slow, unclear, or unavailable, confidence drops quickly. A one or two day delay in a reply may sound minor operationally, but in those first few weeks it can weaken the relationship between a patient and their provider.

The second is the plateau. Weight loss slows, expectations stop matching reality, and motivation drops. Some patients assume the medication has stopped working, when what they need is better expectation setting, reassurance, and a plan.

The third is maintenance. This is the least developed part of most GLP-1 pathways and, in my view, one of the biggest commercial opportunities. Many providers are still focused on acquisition, onboarding, and dose escalation. As more patients reach twelve months and beyond, the real question becomes how you keep them supported, engaged, and progressing over the long term.

This is where AI can help, if it is designed properly. Not by replacing clinicians, but by filling the gaps that traditional services struggle to cover. Patients do not only need support during office hours. Worry appears at night, cravings happen at weekends, and questions come up in the moment. Human teams are usually asynchronous, and that is unavoidable. AI can provide immediate, consistent support at the moment the patient actually needs it.

That matters commercially as well as clinically. A patient who feels unsupported is more likely to churn. A patient who feels understood, reassured, and guided is more likely to stay engaged long enough to see the benefit.

Something many providers underestimate is that patients will often ask AI things they would not ask a clinician, dietitian, or health coach. I have seen this directly in user research. People worry about being judged, or feel they should already know the answer. With AI, that barrier often disappears.

That produces a different kind of data. Not just app usage, but behavioural data. The questions patients ask, the timing of those questions, the concerns they repeat, the moments they go quiet, the support they accept or ignore, and the patterns that appear before a cancellation. These are all signals, and for GLP-1 providers they may become one of the most valuable assets they have.

The real commercial questions are behavioural. Which patients are at risk of stopping. Who needs more reassurance during onboarding. Who is struggling with expectations, or approaching a plateau, or likely to switch provider. Who needs a human rather than AI. And which behavioural intervention works best, for which patient, at which point in the journey.

Most providers are sitting on the raw material to answer these questions, but the data is fragmented. Support tickets, clinical messages, app behaviour, weight data, cancellation reasons, qualitative feedback, and AI conversations tend to sit in separate places. The opportunity is to connect that data and turn it into usable intelligence. That is where AI becomes much more than support automation.

Conversational AI is useful. It provides round-the-clock support, answers routine questions, and takes pressure off human teams. The larger opportunity is agentic and predictive: systems that pull data together, identify disengagement risk, trigger workflows, personalise interventions, and learn which kinds of support are most associated with better retention and outcomes. That is a very different proposition from putting a chatbot into an app.

It is also much harder to do safely. One mistake I see is treating AI as a feature rather than a system. A provider adds an AI coach, puts guardrails around a frontier model, and assumes the behaviour change problem is solved. It is not. Patient-facing AI in obesity care needs careful design, testing, deployment, and monitoring. These systems interact with real patients, many of whom are worried, frustrated, or vulnerable, and asking questions that sit close to clinical territory.

When we first put patient-facing AI live at scale, my team reviewed every conversation. That experience changed how I think about AI safety. You cannot assume a large language model is clinically safe out of the box, and you cannot rely on a disclaimer to solve regulatory risk if the system is drifting into clinical advice.

Intended purpose matters. If an AI is advising on side effects, titration, tapering, or clinical decisions, it may be crossing into medical device territory. That line is not always simple, but providers need to understand it. Calling something wellness does not automatically make it wellness. This is why pre-deployment testing and live monitoring matter.

At Sacher AI, this is a major focus of our work. We help organisations design, evaluate, and monitor AI systems for healthcare and behaviour change, including through tools such as PromptSafe. The aim is not to slow innovation down. It is to make sure AI can be used safely, responsibly, and effectively in real patient pathways.

The providers that get this right will not simply automate support. They will build better retention systems. They will know which patients need help before they cancel, understand which moments in the pathway create risk, and use AI for immediate support, human teams for higher-value care, and behavioural data to keep improving the system over time.

The market will not be won by price alone. Medication costs will change, new drugs will arrive, and generics will come. What patients will increasingly pay for is the quality of the support wrapped around treatment. That support may be AI, human, or, more likely, a hybrid of both, but it needs to be personalised, timely, and built around real behavioural need.

My view is simple. The next competitive advantage in GLP-1 care will not be choosing the right medication. Most providers already have access to the same medications. It will be understanding human behaviour well enough to help patients stay on treatment long enough to benefit. That advantage will go to the providers that build the best behavioural intelligence layer around treatment.

I discussed these ideas in more detail with Dr Ashwin Sharma on the GLP-1 Digest podcast, including AI for patient support, churn prediction, behavioural data, regulation, safety monitoring, and why maintenance may become the most important part of the GLP-1 journey.