There's no shortage of guidance on how to coach a patient who's on a GLP-1. Protein targets, strength training, what to do when the nausea hits. It's useful, and it's all about one coach and one patient.
What you can't find anywhere is the layer above that. How many patients one coach can carry. How often to make contact, and when. What the software should be handling so a coach isn't doing it by hand at 6pm. We went looking for caseload ratios, credential standards and cadence benchmarks, and outside vendor marketing there's nothing published. So what follows is Avidon's operating model rather than a summary of the literature. The evidence underneath it is real and cited. The staffing conclusions drawn from it are ours.
What GLP-1 Coaching Has to Cover
GLP-1 coaching is the behavioral work wrapped around a metabolic prescription: contact between clinical visits, habit change on nutrition, activity and sleep, and outcome tracking that keeps running whether or not the patient is currently on the drug.
That last clause is the one that sets your staffing. A program scoped to the months a patient's actually taking something is scoped to a fraction of the relationship, and it turns out to be the wrong fraction.
The Panel You're Staffing For Isn't the One You Planned
At ENDO 2026 in June, Sainikhil Sontha of the Boston University School of Public Health presented an analysis of more than 60,000 Americans with type 2 diabetes. About 40% had stopped their GLP-1 within the first year and nearly 60% within two. That's a conference presentation rather than a peer-reviewed paper, so hold the decimals loosely. The half that matters for staffing is what came next:
Real-world data points the same way. A Cleveland Clinic cohort published in Diabetes, Obesity and Metabolism in March 2026 followed nearly 8,000 adults who started semaglutide or tirzepatide and then stopped. Of those:
So a real panel at any given moment is a mix. Some patients titrating up, some steady, some paused, some mid-switch, some coming back. A single flat ratio across that mix is what makes a program feel underwater in some months and idle in others, and it's why headcount conversations tend to go nowhere. The timing detail behind all of this sits in our piece on GLP-1 discontinuation coaching.
Intensity Is the One Thing You Can't Cut
Here's the uncomfortable part for anyone planning to bolt on a light coaching layer and call it a program.
What has been tested is whether structured support changes what happens after treatment stops. Jensen and colleagues, writing in eClinicalMedicine in 2024, randomized patients who'd lost a mean 13.1 kg on a low-calorie diet to supervised exercise, liraglutide, both, or placebo for a year, then followed 109 of them through a second year with no treatment at all. Regain during that untreated year was 6.0 kg larger for the group coming off liraglutide alone than for the group coming off supervised exercise. Across the full two years, the group that had both finished 5.1 kg lighter than the drug-only group, with body fat 2.3 percentage points lower.
Two caveats worth saying out loud: it's liraglutide rather than semaglutide or tirzepatide, on a small post-treatment group, and supervised exercise is one form of structured support rather than a stand-in for coaching generally. But the direction is a staffing finding as much as a clinical one. What separated the outcomes was structured, supervised support. Not the presence of support. A quarterly check-in call isn't a smaller dose of that. It's a different thing, and it doesn't do the same job.
Our Staffing Model for GLP-1 Coaching
Nobody's published one, so here's what we'd build. The reasoning is visible on purpose, so you can argue with a specific piece of it instead of the whole thing.
Size Caseload by Phase, Not by Headcount
A patient in the first eight weeks after stopping needs several times the contact of a patient who's been steady on maintenance for a year. Give every coach a flat 150 and you've built a model that's wrong in both directions at once, overloaded whenever a cluster of patients pauses and slack whenever they don't.
Weight the panel instead. Count a recently stopped or switching patient as several maintenance patients, rebalance monthly, and let headcount follow the weighted total rather than the raw one. That's ordinary caseload management with one extra variable, and the variable is medication status.
Put the Cadence Where the Risk Is
Regain follows a decay curve, not a straight line. A systematic review and nonlinear meta-regression published in eClinicalMedicine in March 2026 pooled six randomized trials and 3,236 participants and put the half-life at twenty-three weeks, with roughly 40% of the on-treatment loss still in place a year out.
So the heaviest contact belongs immediately after a stop or a switch, tapering after. Most programs run precisely the opposite, loading attention at intake and thinning it out by the time the outcome is actually being decided.
Automate the Contact That Doesn't Need a Coach
A coach's hour should go to the conversation. It should not go to noticing that a refill didn't happen, sending the reminder, or logging that the reminder went. If a discontinuation only reaches your team as a gap in a report three weeks later, the window that mattered is already gone, and no amount of caseload math fixes it.
You need a definition of "stopped" that a system can act on, and the ENDO researchers used a workable one: a gap of more than sixty days in filling the prescription. That's a trigger you can wire up. It's also the closest thing to a published operational standard that exists here, which tells you how young this field is.
The test is simple. If a patient stopped their prescription in the first week of the month, does anything in your system fire that month? If it doesn't, that's the first thing to fix, ahead of any ratio. Building the panel around that trigger is what our work with GLP-1 and metabolic clinics is organized around.
How to Tell If the Model Is Working
Average weight loss across the panel won't tell you. It blends patients in four different states and moves for reasons that have nothing to do with your team.
Watch detection lag instead: the days between a patient stopping and somebody on your team knowing. If that's measured in weeks, nothing downstream of it can work. Then watch what share of paused patients are still in contact ninety days later, because that's the number the post-treatment evidence says is load-bearing.
None of this needs a bigger team. It needs the panel counted differently, the contact moved to where the risk actually is, and a system that doesn't wait three weeks to notice a stop.
