Finch is a unified API for HRIS and payroll systems, leading the market with a network of 220+ data providers. Their platform has connected 50,000+ employers and synced nearly 14M unique employer records, including $99B in payroll earnings.


Early on, Finch realized that at scale, the costs for compute, network, and storage of CI could balloon up to numbers that’d make your eyes water. So, like many brave souls before them, they turned to Kubernetes Action Runners Controller (ARC) to build and maintain a self-hosted solution.

The Problem

Moving away from self-hosted runners.

As any developer who has tried can tell you, operating self-hosted runners on Kubernetes is a beast. It’s a constant battle to make it work with ARC and fine-tune auto-scaling just right so that your system can handle your team’s spiky CI workloads.

Ultimately, it became a huge pain to maintain and wasn’t stable enough for Finch. It frequently failed to pick up jobs, and at times, the pipeline would sit idle for 30 minutes waiting for a runner. More importantly, it wasn’t something that could be treated as an afterthought. CI shouldn’t be something you constantly have to think about, but with this setup, it was.

Sure, the self-hosted solution, on paper, was “cheaper” than GitHub-hosted runners, but it was proving to be anything but cost-effective. It wasn’t reliable, and the maintenance time was steadily spiraling out of control. Fernando Battistella, DevOps Engineer at Finch, was getting paged multiple times during the night, and it started to become a hidden, but very real, operational cost.

Eventually, they realized it was time to explore new options. Going back to the price tag of GitHub-hosted runners? Not happening. But what if there was a solution that combined reliability with low costs? Now, that’s a thought worth exploring.

Sometimes, the real cost of a solution isn't just in the price tag of the compute—it’s in the time, effort, and sleepless nights that go into keeping it running.

— Fernando Battistella, DevOps Engineer, Finch.

The Solution

Selecting Blacksmith over self-hosted.

Fernando, being the responsible buyer he is, set out to explore multiple options. But it didn’t take long for him to grow dissatisfied with the third-party services he tried. Under the hood, their architecture looked eerily familiar—Kubernetes, ARC, and a whole lot of grit to tackle autoscaling issues. And, as it turned out, those solutions faced the same reliability hiccups he was already dealing with.

Then, Fernando stumbled upon Blacksmith’s architecture: bare metal, microVMs, and all the right ingredients. “It looked like the kind of solution I would have built from scratch, if I had time and a couple more lifetimes,” he shared with us. After giving it a spin it worked. And it worked well. The out-of-the-box observability into the CI pipelines—something GitHub still barely offers—and the fact that he didn’t have to migrate away from the GitHub Actions ecosystem was the final nudge he needed to choose Blacksmith for Finch’s CI/CD pipelines!

Blacksmith just works. CI runners are not something I have to think about anymore.

— Fernando Battistella, DevOps Engineer, Finch.

The Result

The ultimate sleep solution.

Today, Finch runs all their CI jobs on Blacksmith, saving the hidden, operational cost of their most expensive engineers' time. Fernando can now get his time back to do far more interesting things than worrying over a failing CI. And yes, he sleeps better than ever! The whole team does.

I now have time to do other tasks, besides managing CI runners, that will help us grow even more.

— Fernando Battistella, DevOps Engineer, Finch.

If you're currently pulling your hair out dealing with self-hosted runners, let us help you too. We promise there is a better, end-to-end cheaper way.