2024–25 · Groupon · via TrustSoft · US · Case study

A team asks for a cloud project. Hours later it exists.

Eighteen months embedded on one long-running project at Groupon, through TrustSoft: the team that runs the shared cloud foundation every development team builds on, across AWS and Google Cloud. My part was making a new project something a team could get without waiting.

The problem

Groupon's development teams needed new cloud projects constantly — a place to deploy, with the right permissions, networking and guardrails already in place. Each one was set up by hand: days of waiting on a small platform team, and no two projects came out quite the same.

At that scale a queue for cloud projects is a queue for everything downstream of them.

What existed already

A serious platform: an enterprise landing zone — the shared foundation every team's cloud sits on — across AWS and Google Cloud, run as Terraform and Terragrunt at enterprise scale, plus a set of internal developer tools the teams already relied on.

It was the second time I had walked into a many-teams, one-foundation problem. The first was TrustSoft, where I had built the foundation myself.

Whose part is whose

Theirs

Groupon's platform team

The landing zone across AWS and Google Cloud

Terraform and Terragrunt at enterprise scale, and the internal developer tools teams already used. I operated and maintained it with the team.

This part is mine

I built this

Self-service cloud projects, end to end

Fully automated Google Cloud project creation — Python, with Terraform templates built in

Standard patterns and templates, so every project came out the same

The internal developer tools, improved (Python, Terraform, Docker)

Demos of each piece to the teams who would use it

Who used it

Every development team

Any team that needed a new project

development teamsAWSGoogle Cloudone standard on both clouds

I was also the person those teams came to when a deployment would not go through — support, not just tooling.

What changed

Days → hours

Project provisioning lead time dropped from days of manual ticketing to automated creation in hours.

Two clouds

AWS and Google Cloud, one set of patterns, one standard for how a project is set up.

Every team

Self-service provisioning eliminated the platform team bottleneck for new GCP and AWS development projects.

What did not go well

Automating project creation removed one queue and exposed the next: I stayed the person teams came to when a deployment would not go through. Self-service moved the bottleneck; it did not delete it.
It was their platform, and I was there through TrustSoft. I could automate around the platform team's process, not redesign it — some of what I would have changed stayed as it was.

What it taught me

The next thing to automate is the answer to the question teams ask most, not the thing that is most interesting to build.
Demo it to the people who will use it, in their words, before you call it done. The demos changed what I built more than the specification did.

Where it went

I demoed and presented what we built directly to Groupon's teams, and worked with their engineers on the deployment practices that went with it.

What it left me with

The same problem for the second time, at American scale: many teams, one foundation, and the foundation only earns its keep if getting onto it is effortless.

The first time was at TrustSoft; the third is Glow.

← All work

Next · Glow →