Arize AI, a machine learning (ML) observability and model monitoring platform, announced it has raised $19 million in Series A financing. The round was led by Battery Ventures with participation from previous investors Foundation Capital, Trinity Ventures, The House Fund, and Swift Ventures. Dharmesh Thakker, general partner at Battery Ventures, will join the Arize AI board.
Machine learning is the backbone of modern technology, powering artificial intelligence (AI) systems that touch all aspects of life. But these systems are extremely complicated, and many ML practitioners don't have the right tools or telemetry to understand how or why their creations work.
Common problems plague all ML models, and leaving them undetected and unmitigated eventually disrupts business and creates negative experiences for customers and ultimately communities and society. Adding to the complexities of AI is a lack of transparency and observability – therefore creating poor accountability – which can lead to biased ML models that fail to protect individuals against discrimination, promote inclusion, diversity and equity, and safeguard equality.
The Arize AI founding executive and engineering teams, composed of industry veterans from industry-leading organizations including Uber, Google, Apple, Slack, Adobe, and PagerDuty, have charted a course for the industry's first full-stack ML observability and model monitoring platform. The platform is the only solution designed specifically for ML engineers, data scientists, and other practitioners responsible for deploying and maintaining ML models that drive business decisions and processes.
Arize AI's Aparna Dhinakaran is one of the few female co-founders in the AI industry. Prior to co-founding Arize AI, Dhinakaran was a Ph.D. candidate in computer vision at Cornell University and before that led the design and development of Uber's first model lifecycle management system. A season 32 contestant on The Amazing Race, she tested her limits and learned valuable life lessons, finishing amongst the race's top 5 teams.
"In the same way that tools had to be created in the software industry to track issues, manage version history, oversee builds, and provide monitoring, we're seeing a similar trajectory in the ML space," said Aparna Dhinakaran, co-founder and chief product officer at Arize AI. "Without the tools to reason about mistakes a model is making in the wild, teams are investing a massive amount of money in the data science laboratory but essentially flying blind in the real world."
Since announcing its seed financing in October 2020, Arize AI has gained traction amongst enterprises such as Adobe and Twilio that are looking to ensure production models perform as designed in the research and building phase. Other customers include organizations in financial services, fintech, healthcare, insurtech, ad tech, retail, and other industries that rely on AI for fraud detection, pricing, demand forecasting, and service delivery optimization.
"Despite the significant effort and resources put into building and shipping models, no model is going to be perfect at processing and understanding natural language under natural conditions," said Brendon Villalobos, machine learning technical lead at Twilio. "The Arize AI platform provides an intuitive UI that's easy to use and can monitor drift and performance of all models across our most advanced communication deployments. With Arize, our team of practitioners can now quickly and easily observe and continuously improve models, solving one of the core pain points to keeping our ML initiatives on track."
"In our business, ML models make decisions daily that determine if our customers make or lose money by deciding what ad spots to buy," said Alok Kothari, director of machine learning at Adobe. "The ability to quickly change what we've built, understand how it's different from previous models and know where it has problems is mission-critical, particularly in the context of our commitment to innovation and leadership in the increasingly privacy-focused advertising environment."
"While almost every business is massively investing in artificial intelligence for competitive advantage, very few are able to deliver models continuously with a high return on that investment," said Jason Lopatecki, CEO of Arize AI. "Arize AI is successfully eliminating barriers to a future where ML practitioners understand why a machine learning model behaves the way it does after it is deployed into the real world. Ultimately, as AI systems become increasingly complex, their capabilities and limitations will become more profound and will require a highly advanced level of useful, meaningful human oversight to ensure they are contributing to, not detracting from, societal well-being."
"As the world becomes increasingly AI-centric, there will be a few primary categories of ML infrastructure tools that truly matter for data organizations," said Battery's Thakker. "Billions have been invested in two categories: data preparation and ML model building; leading to a flood of models being deployed across every industry. However, the actual value of a model's impact on business and customers is often hazy at best. Similar to how solutions help teams manage their software infrastructure investments, organizations that are serious about ML need to employ a toolchain for ML infrastructure."