Just ten months after their $5 million seed funding, Aporia, the end-to-end observability platform for machine learning, announced today a $25 million Series A funding round led by Tiger Global. The round included participation from Samsung Next, and existing investors TLV Partners and Vertex Ventures.
Whether we like it or not, AI is part of our daily lives; a majority of organizations have adopted AI to support a variety of functions including product, customer support, marketing and sales. As a new necessity, AI has, for the most part, made our lives better, but like software, it also has issues and bugs. The difference is that, unlike classic software, AI models fail "quietly", so failure detection occurs at a very late stage, which can lead to significant damage already having occurred once detected. This lack of visibility of whether ML models are accurate, equitable, and fair can have devastating outcomes for businesses and people. For example, Zillow, the online real-estate marketplace company, reported that they failed to accurately leverage AI to forecast house prices, which led to hundreds of millions in losses and thousands of employee layoffs. Another incident where AI had unexpected behaviour was when Amazon used an AI recruitment tool which led to the discrimination of women candidates.
AI is somewhat of a black box, so it makes issues more difficult to spot, resulting in a potentially larger impact. These issues are now a daily reality that companies are grappling with and that have the potential to affect us all. Aporia, the Israeli customizable machine learning observability platform, is tackling that challenge, by ensuring an organization's use of AI is ethical and responsible.
In a conversation with Geektime, Aporia’s CEO and co-founder, Liran Hason, explained that to shed light on the AI black box and ensure the safety of a company’s highly specialized AI models, data scientists require a tailor-made observability solution. But such a solution is expensive and time-consuming to build from scratch. Aporia’s monitoring system provides organizations with full transparency of AI model decisions and knows how to identify and alert to significant events in model functioning such as performance degradation, concept drift, suspected discrimination and more. Essentially, within minutes, data scientists can easily create customizable monitors to detect a wide range of issues including biased predictions, unexpected changes in the format of the input data, or degradation in a model’s performance over time. The platform helps interpret how models are making predictions, by showing how much each input parameter contributed to the decision and how the prediction shifts with changes to the input data. It thus provides full visibility of how models are performing in the real world, with actionable insights that help data scientists investigate and get to the root cause of any issue. The best part is that it can be seamlessly integrated with existing infrastructure so ML practitioners can easily get started. Hason continued by saying that what sets Aporia apart from the other similar platforms dealing with this issue, is that Aporia’s system is flexible– it allows monitoring of over 50 different types of behaviours and allows data delicacies to implement the monitoring logic themselves.
Put it to the test
Hason gave us a real example of how Aporia was used by a financial company. As we all know, financial entities can use AI models to assess the risk of a candidate applying for a loan, which they then apply automatically to approve or refuse an application. For example, company users can apply for a loan automatically through the company website or the app. In one case, they had a company that had a developer who worked on the Web app referred to the "Income" field as annual income, while in the same token, a data scientist developed a model that referred to the "Income" field as monthly income. As a result, every loan application that came to the model was experienced with a salary 12 times higher than what was actually the case, which obviously affected the system's decision to approve of a loan. Luckily, Aporia detected the flaw and alerted the team quickly. Without their monitoring, this malfunction could have lasted for many weeks and created significant damage to the company.
Fortune 500 companies and data science teams in every industry across the globe have been adopting Aporia’s solutions, so much so that they have experienced a 600% growth in customers over the past six months. Some of their more notable clients include leading companies like New Relic, Lemonade and Armis. The funding from this round will enable significant growth in customers and users. Aporia also plans to increase the range of use cases their solution addresses, triple its team’s headcount by the end of the year, and expand its presence in the U.S. market.
“Aporia has demonstrated unbelievable growth since its launch and has amazing momentum. They are quickly becoming a leader in the space of ML observability,” said John Curtius, Partner at Tiger Global. “Executives at global enterprises understand the benefits of artificial intelligence and how it's impacting virtually every industry, but the risks keep them awake at night. Aporia is positioned to be the solution every organization turns to for ensuring their responsible use of AI.”
Aporia was founded in 2019 by Liran Hason (CEO) and Alon Gubkin (CTO). To date, they have raised $30 million.