Company
MyCoach Pro is a French SaaS company providing an Athlete Monitoring System (AMS) for sports organizations, from amateur clubs to professional structures. The platform is built around three steps: collect and centralize performance, medical, video, technical, scouting, and administrative data in one place; analyze and decide using dashboards and reporting built for each role; and share that data seamlessly across coaches, medical staff, analysts, scouts, sporting directors and athletes. MyCoach Pro’s AMS is used by clubs across 5 regions and 34 countries, spanning 13+ sports including football, rugby, basketball, and handball.
The Challenge: Manual Scaling Was A Limitation
As a SaaS provider hosting a multi-tenant application on Kubernetes since 2018, MyCoach Pro’s infrastructure had one job: stay up. “It’s really important to us because it allows us to be resilient and provide the best platform to our customers,” says Clément Agarini, VP of Engineering.
But keeping that infrastructure rightsized was a persistent drain on the team’s time. Multiple times a year, the team had to manually resize node pools – moving workloads from one pool to another to clean and reorganize the cluster to keep up with traffic growth and increased platform usage. The test environment, meanwhile, sat chronically overprovisioned simply because no one had time to optimize it.
Pod sizing added another layer of guesswork. Every time MyCoach Pro deployed a new service, the team had to manually estimate CPU and RAM requirements – a slow, inexact process that routinely led to inefficient resource allocation.
The team had wanted automation for years. Early attempts to explore Kubernetes autoscaling on GCP felt too complex to configure correctly, especially given requirements around minimum resource guarantees for RAM and CPU, so MyCoach Pro defaulted to committing to fixed node pools instead. The manual approach persisted until a Cast AI demo changed the situation entirely.
The Solution: Gradual Automation Rollout, Immediate Results
MyCoach Pro chose Cast AI after seeing how straightforward the setup was during an initial demo.
“I think Cast is pretty easy to sell – the ROI is awesome from the start. It’s easy to install and easy to manage, so I didn’t take time to look at anything else.”
Clément Agarini, VP of Engineering at MyCoach Pro
From Testing to Production
Onboarding was intentionally gradual. The team began by sending data to Cast AI to model expected savings before touching anything in production. From there, they rolled Cast out to the test environment first – running it entirely on Preemptible VMs – and immediately saw over 70% cost reduction, bringing the environment to under $10 a day.
The team had long known Preemptible VMs represented a major savings opportunity, but managing the rescheduling complexity had always been a barrier. Cast AI handled that automatically, continuously rebalancing the cluster throughout the day to maximize Preemptible VM usage while maintaining stability.
Production required more care. An initial configuration led to Google aggressively reclaiming Preemptible VMs, causing brief stability issues. The Cast team resolved the problem within two to three hours, landing on a balanced mix of Preemptible and on-demand VMs that preserved both savings and reliability. MyCoach Pro was running in production within two weeks.
Support Along The Way
“The Cast team is very responsive – we get answers to our questions in minutes. Your team helped me understand why we had the surge in CPU and memory usage and how to configure it properly – I got someone on a call in the afternoon when we saw the issue in the morning.
We had a minor issue with scaling up at some point because one of our workloads needed to scale aggressively. But someone from support helped us make sure that this particular workload – which we know can scale up and down quickly – was properly configured, and now everything is in place.”
Clément Agarini, VP of Engineering at MyCoach Pro
Throughout onboarding, the Cast team also helped surface and fix infrastructure issues that had gone unnoticed for years. Because node pools rarely changed under the old setup, problems with workload configuration simply never surfaced. Onboarding brought them into the open – including the need to configure Pod Disruption Budgets to ensure at least 50% of critical pods remained available during node changes or upgrades.
“Thanks to the help we got when onboarding, we were able to fix things that should have been fixed a long time ago.”
Clément Agarini, VP of Engineering at MyCoach Pro
The Results: GKE Infrastructure on Autopilot
With Cast AI automatically managing cluster rebalancing, pod rightsizing, and Preemptible VM scheduling, the operational picture shifted dramatically.
Dramatic Improvement In Resource Utilization
Automated workload rightsizing eliminated the manual guesswork on pod sizing. Cast continuously scales CPU and memory requests to match the actual workload.
In one example, MyCoach Pro reduced requested CPUs across the cluster from 21.305 to 12.372 – a 42% reduction:
At the same time, Cast ensures that memory requests remain stable to eliminate the risk of OOM kills:
Significant Time Savings
Time savings may be the most tangible shift day to day.
“Multiple times a year, I needed to manually rescale the node pool and move things from one node pool to another in order to clean it and reorganize. With Cast, I’m saving time by delegating this task to a solution that does it better than me. On top of that, I run the same workloads for a fraction of the price.”
Clément Agarini, VP of Engineering at MyCoach Pro
A New Level of Operational Confidence
Operational confidence has grown alongside the automation. With the team shipping new services weekly, MyCoach Pro no longer has to manually assess whether the cluster has enough capacity for what’s coming next. Cast automatically rightsizes every new service as it’s deployed.
“I can be confident that we only use what we really need to run the platform.”
Clément Agarini, VP of Engineering at MyCoach Pro
60% Cost Savings
Cost savings came in at 10% of the total monthly cloud budget – a 60% reduction in compute spend that the team describes as “huge” for an engineering org that made a deliberate decision years ago to forgo a dedicated sysadmin and move fully to the cloud in the name of cost efficiency. Cast AI became the missing piece that made that bet pay off on the compute side.
By automating infrastructure management on both the cloud and Kubernetes side, Cast helped MyCoach Pro reduce its overall compute costs and enable the use of Preemptible Instances.
Example cost savings
Automation Became The New Normal
Today, the team’s Monday morning routine is to check the Cast AI dashboard – and that’s it. Their advice to other engineering teams considering the switch:
“Put infrastructure on autopilot. Just don’t look at it – use it. You’ll be more cost-efficient, and you won’t have to look at the node pools on a daily basis. You’ll also benefit from Preemptible VMs and be able to rebalance the cluster to keep costs as low as possible every day.”
Clément Agarini, VP of Engineering at MyCoach Pro
