Were you unable to attend Transform 2022? Check out all of the summit sessions in our on-demand library now! Watch here.

Explainable AI is vital for trust, compliance, and building better — and hopefully less biased — AI models. Both customers and regulators want to know more about what’s inside the black box. But most monitoring tools leave blind spots, alerting data scientists to issues but not their causes.

Today, AI startup TruEra launched a solution that allows enterprises to not only have a deeper understanding of their machine learning models, but also to identify the root cause of issues and contextualize them almost instantaneously. The tool, called TruEra Monitoring and based on the company’s AI Quality Platform, looks for AI quality metrics including bias, model stability over time, and which features are driving the score.

“If a model is experiencing bias, or is experiencing drift, we can pinpoint that to the exact features that are driving that disparity and then put that in proportion to how much they’re driving that disparity,” Will Uppington, cofounder and CEO of TruEra, told VentureBeat. “That’s pretty much instantaneous, whereas most solutions just don’t do that root cause analysis.”

Filling the monitoring gap

Current solutions can monitor infrastructure and KPIs, but by offering a way to monitor the inputs and outputs of machine learning models themselves, TruEra aims to fill what it calls the “monitoring gap.” In some benchmarks, the company said, the tool is more than 10 times faster than other solutions at a comparable level of accuracy.


MetaBeat 2022

MetaBeat will bring together thought leaders to give guidance on how metaverse technology will transform the way all industries communicate and do business on October 4 in San Francisco, CA.

Register Here

The other differentiator is that TruEra Monitoring is best integrated during development, offering diagnostic insights that can shape a better model from the early stages.

“The approach a lot of people have taken is they’ve modeled it after SAS infrastructure monitoring, which is kind of a standalone, very performance-oriented activity, whereas that’s completely the wrong approach for machine learning,” Uppington said. “We believe the right approach is that monitoring is just a continuation of the development cycle. Today’s operational data is tomorrow’s training data.”

Serving the customers

When using the tool, customers can interact with a variety of dashboards, including an overwatch page that summarizes each model being monitored. There are also links to debugging tools, so teams can act quickly when they find an anomaly. And for users who prefer to integrate the data into their existing platforms, there’s also an API to export the data.

HarperCollins, one of the first companies to deploy the tool, says it has been “very happy with the results” so far.

The book publisher has been using TruEra Monitoring since February, when it was still in beta. Matthew Bennett, North America CIO of HarperCollins Publishers, says monitoring of machine learning is one of the major missing pieces of IT management, and he’s excited the solution goes beyond current ad hoc tools.

“TruEra helped us improve the monitoring of some of our system’s outputs, shifting away from a manual, non-scalable approach,” Bennett told VentureBeat.

VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.