Industry:

Financial Technology

Website:

www.upstart.com

Use Cases:

Cost allocation, multi-cloud visibility, cost optimization, shift left, FinOps partnership, financial reporting

Project Leads:

Chris Gray, principal engineer at Upstart

Overview

About

Upstart is the leading AI lending marketplace, connecting millions of consumers to more than 100 banks and credit unions that leverage Upstart’s AI models and cloud applications to deliver superior credit products.

Challenge

Upstart wanted to reduce its cloud spending by at least $10M, which required a single pane of glass for all cloud costs, engineering cost accountability, and a FinOps partnership.

Solution

Implemented CloudZero to get complete cost visibility, drive engineering cost accountability, and automate financial reporting.

Outcomes

  1. CloudZero helped reduce annual cloud costs by $20M
  2. Shifted cost accountability left
  3. Saved 5 hours a month on financial reporting
  4. Calculated core unit cost metrics

The Challenge

Upstart is the leading AI lending marketplace, and it’s dedicated to improving access to credit for all Americans. Whereas traditional lending relies almost wholly on a single metric — the FICO score — Upstart uses AI and alternative data, like education and employment status, to more accurately assess borrower risk and expand access to credit.

Like many companies, Upstart faced challenges last year as both interest rates and inflation weighed on consumers and lending slowed. 

“For us, that meant one thing: It was time to get lean,” said Chris Gray, principal engineer at Upstart.

Like many SaaS companies, for Upstart getting lean started in the cloud. Prior to engaging CloudZero, Upstart had a sizable cloud footprint, spread out across AWS (the biggest slice of the pie) and a number of other service providers. Because Gray had been running Upstart’s cloud infrastructure team, Upstart anointed him the cloud cost reduction czar.

“First, we just wanted to put a dollar and cent goal on it,” Gray said. “So, we targeted $10M.”

Including the savings objective, Gray then broke his cost-optimization goals into three main prongs:

  1. Single pane of glass. Upstart’s vendor diversity meant they had a lot of cloud cost data in a lot of different formats and platforms. Chief among Upstart’s goals was unifying all cost data in a single pane of glass, where it would be easier to explore, analyze, and optimize.
  2. Cost per product. Upstart had five major products spanning several loan categories, including personal loans, auto loans, mortgage loans, and more. Given the different nature of these loans — size, recipient profile, frequency — the most important unit cost metric for Upstart was cost per product.
  3. Shift left. Gray recognized that the best way to manage costs was to get engineers in the habit of making cost a nonfunctional requirement from day one. So, he needed a platform that would fuel this cultural change within Upstart.

The Solution

Gray had experimented with other cost optimization vendors and couldn’t find the right one to fit Upstart’s exact needs. So, he turned to CloudZero.

“The first thing we talked about was getting visibility into our AWS accounts,” Gray said. One of Upstart’s most important AWS visibility areas was RedHat OpenShift on AWS (ROSA), for which CloudZero quickly developed support. “The fact that CloudZero so effectively prioritized ROSA support was really impressive,” Gray said.

CloudZero ingested and allocated all of Upstart’s cloud costs, giving them the single pane of glass they craved at the outset of the engagement. From there, it was just a matter of engineering-led optimization — getting the right data to the right people, racking up savings, and driving efficiency.

The Results

Reduced cloud costs by $20M

Upstart had set an ambitious savings goal of $10M when the company set out to cut cloud costs in early 2023—and CloudZero helped the company to double that figure by the end of the year.

The savings came from a variety of places, one of which was a whale of a data-analytics project.

“The billing data from the service we used was not the easiest to parse,” Gray said. “Once we got it into CloudZero, we looked through all of the custom usage metrics we were getting and pinpointed some runaway spend activities that needed to be resolved.” Executing on its cost-reduction plans led to $700,000 of annualized savings, paying for CloudZero many times over.

Other savings came directly from Upstart’s FinOps Account Manager (FAM), Zongde Li. Gray recounted a story about Upstart’s AWS Config bill, which led to $168,000 of annualized savings with a minimum effort. “Without CloudZero, we would’ve overlooked it,” Gray said. “But having the insight that our Config costs were higher than other CloudZero customers of our size prompted an educated investigation.”

Lastly — and perhaps most importantly — savings came directly from Gray’s engineers.

Shifted cost accountability left while maintaining speed of innovation

“Giving all our teams access to quick views of their spend has enabled them to understand where their costs are — and, when necessary, how to optimize,” Gray said.

Gray described Upstart’s pre-CloudZero engineering culture as being primarily focused on creating new features. “Value was in feature delivery,” Gray said, “and cost was a lesser part of engineering decisions.”

Cost optimization was a reactive process. Now, it’s a proactive process.

“I now have teams coming to me saying, ‘Hey, we saw costs going up, we accomplished xyz tasks, and we think we saved the company money. Can you help us show that?’” Gray said. He said he gets these kinds of stories frequently, meaning his engineers are identifying and solving cost problems autonomously.

The best part, he said, is that it hasn’t hurt Upstart’s speed of innovation at all. On the contrary, it may have expedited it. “You have to consider costs eventually,” he said, “and when you do it late, you tend to have to go back and spend a bunch of time rearchitecting. Considering cost upfront makes your innovations much more durable.”

Saved 5 hours/month on financial reporting

Before CloudZero, Upstart’s finance team had asked Gray to compile reports on per-organization costs, including machine learning, IT, data engineering, and more. It was a “cumbersome process,” Gray said.

With CloudZero, he got Upstart’s cloud costs allocated by organization, and then created a custom dashboard in CloudZero Analytics that automated the creation and dissemination of cost-per-organization reports. Gray estimated that this report alone saved him five hours a month in labor.

Calculated core unit cost metrics: cost per product, cost per organization

In addition to cost per organization, CloudZero also successfully calculated Upstart’s cost per product, enabling them to get cost data in the hands of the most relevant engineers. In the future, Gray said he plans to use CloudZero to measure Upstart’s cost per origination and cost per verification — unit costs that sit at the core of Upstart’s business value.

Gray is very pleased with CloudZero’s impact. “It’s a visibility platform with great insights into opportunities for cost savings,” he said. “It makes it simpler to digest how you’re spending in the cloud — and how to optimize.”