Arize AI, a machine learning operations platform startup, today announced that it has raised $38 million in a Series B round led by TCV with participation from Battery Ventures and Foundation Capital. Bringing Arize’s total capital to $62 million, CEO Jason Lopatecki says the new money will be used to scale R&D and double the company’s headcount from 50 over the next year.
Machine learning operations, or MLOps, is about deploying and maintaining machine learning models in production. Like DevOps, MLOps aims to increase automation while improving the quality of production models—but not at the expense of business and regulatory requirements. Given the interest in machine learning and AI more broadly in the enterprise, it’s no surprise that MLOps is expected to become a large market with IDC placing the amount is about $700 million by 2025.
Arize was founded in 2019 by Lopatecki and Aparna Dhinakaran after Lopatecki sold a previous startup — TubeMogul — to Adobe for about $550 million. Lopatecki and Dinakaran first met at TubeMogul, in fact, where Dinakaran was a data scientist before joining Uber to work on machine learning infrastructure.
“After watching team after team – year after year – fail to understand what was wrong with the models delivered to production and struggle to understand what the models were doing once they were deployed, we came to the conclusion that something was fundamentally missing Lopatecki told TechCrunch in an email interview. “If the future is driven by AI, then there needs to be software that helps people understand AI, break down problems and fix them. AI without the ability to monitor machine learning is not sustainable.”
Arize is certainly not the first to tackle this kind of data science challenge. Another MLOps vendor, Tacton, recently raised $100 million to build its experimental machine learning model platform. Other players in the space include Galileo, Modular, Portal and Grid.aithe latter of which secured $40 million in June to launch a gallery of components that add AI capabilities to apps.
But Lopatecki argues that Arize is unique in several ways. The first is the focus on observability: Arize’s embedding product is designed to peer into deep learning models and understand their structure. “Bias Tracking” complements it, a tool that monitors for biases in models (e.g. facial recognition models that recognize black people less often than subjects with lighter skin) — and tries to trace back to the data causing the deviation.
More recently, Arize debuted built-in drift monitoring, which attempts to detect when models become less accurate as a result of outdated training data. For example, drift monitoring can alert an Arize client if a language model answers “Donald Trump” in response to the question “Who is the current president of the United States?”
“Arize stands out… [because] we are laser-focused on doing one hard thing well: machine learning visibility,” Lopatecki said. “Ultimately, we believe the machine learning infrastructure will look like a software infrastructure with a number of market-leading, best-in-class solutions used by machine learning engineers to build great machine learning.”
Arize’s second distinguishing feature, Lopatecki says, is its expertise in the field. Both he and Dhinakaran come from academia and draw from practitioner roots, he notes—having built machine learning infrastructure and managed model problems in manufacturing.
“Even for teams that are experts and thought leaders, it becomes impossible to keep up with every new model architecture and every new breakthrough,” Lopatecki said. “Just as quickly as teams finish building their latest model, they typically move on to the next model that the business needs. This leaves little time for deep introspection into the billions of decisions these patterns make daily and the impact these patterns have on both business and people… That’s why Arize spent over a year building a pattern monitoring product for in-depth training and designed workflows to troubleshoot where they go wrong.
Some might argue (correctly) that Arize’s competitors also have experts in their ranks as well as surveillance and monitoring solutions in their product suites. But judging by Arize’s impressive list of clients, the startup is making a pretty damn convincing pitch. Uber, Spotify, eBay, Etsy, Instacart, P&G, TransUnion, Nextdoor, Stitch Fix and Chick-fil-A are among Arize’s paying customers, and the company’s free tier — which launched earlier this year — has more than 1,000 users.
Mom, however, is the word for annual recurring revenue. Lopatecki was adamant that the Series B capital would give the company “wide runway,” macro environment be damned.
“In healthcare, there are teams using Arize to ensure cancer detection models using imaging are consistent in production across a wide range of cancers. Additionally, there are teams using Arize to ensure that the models used in standard of care decisions and insurance experiences are consistent across racial groups,” Lopatecki added. “As models become more complex, we’re seeing even the largest and most sophisticated machine learning teams realize that they’d rather invest their time and energy into building better models than building a tool to machine learning monitoring… Arize helps practitioners improve ROI from models and quantify results for business leaders [and provides] the market-leading AI investment risk monitoring software.”