The franchise's small-group model put up to twelve members lifting at once under a single coach's supervision. That ratio worked for running a class, but not for catching every mistake: a coach could only watch one bay at a time, and a form error on a squat or deadlift in a bay they were not looking at could run for the rest of the session.

Rep counts came from members themselves, called out or logged after the fact, and were unreliable by nature. Without a dependable count, neither members nor coaches had an objective record of how a lifter's numbers were actually moving over weeks or months, so progress was mostly a feeling rather than something anyone could point to.

The consequences showed up in the numbers the franchise tracked at network level. New-joiner injuries were trending upward, and so were early cancellations, both signals of a class experience that was not catching problems early enough to keep new members safe, confident and coming back.

We built a camera-based motion and pose analysis system for each lifting bay, rather than asking a coach to be in more places at once. A pose-estimation model tracks joint angles through every rep a member performs, giving the bay an objective read on form without any wearable or extra step from the lifter.

The system counts sets automatically, replacing self-reported numbers with a dependable record, and surfaces colour-coded form flags on the bay's display in real time. That turns the coach's attention into a directed resource: instead of scanning the whole room and hoping to catch the right moment, they can see which bay needs them right now.

Post-session summaries with rep quality and load progression sync to the member app, so lifters get an objective record of their own progress for the first time. Studio-level dashboards roll the same data up to the franchisor, giving the network a view across all its studios rather than one class at a time.