Aderant’s president and CEO Chris Cartrett and CTO Andy Hoyt give Legal IT Insider their insights into new AI associate Maddi, including what this significant new capability means for Aderant’s customers.
Aderant formally unveiled its new AI-powered virtual associate Maddi to its European user base on 21 June, as part of its Momentum Europe 2023 conference, hosted in London.
First unveiled at Momentum in Denver, Colorado, in May, Maddi is being applied across Aderant’s solutions and is designed to automate routine, repeatable tasks. It has been pre-trained on extensive data from several Aderant applications, as a result of which we’re told that it will require minimal training by law firms.
The first solution to leverage Maddi is new outside counsel guidance compliance tool Onyx, which extracts terms from OCG documents and categorizes them with minimal human intervention. It then helps to create and validate rules at each step of the time and billing journey. Maddi will also shortly be extended to timekeeping solution iTimekeep and Expert Accounts Receivable. Aderant is also looking into how to apply Maddi within its North American docketing solution Milana.
Maddi is built using open source technologies and speaking to Legal IT Insider, CTO Andy Hoyt, who joined Aderant in January from NCR Corporation and before that spent six years at IBM, said: “We are using lots of different open source technologies for different use cases. And so there’s Q&A; there’s vector searches, and there is natural language processing. They are all open source libraries that are out there. Even around the vision side where you take OCRs or image classifications, there’s LayoutLM open source libraries and communities.
“Then it comes down to the data that you get to train it on. The data set that’s out there and feeding into all of this is from the likes of BillBlast, iTimeKeep and Expert.”
The fact that the AI is pre-trained is central to Aderant’s messaging, and CEO Chris Cartrett said: “Versus this just being an open source AI or any kind of AI for that matter, this is about the training, and when users hear the name Maddi, they can know that when they roll it out, in the likes of iTimeKeep it has already been trained to know which narratives are acceptable to which clients.
“That to us is what our special sauce is. It is taking all this data and information, and we are in a very unique position to be able to train the AI. You’re not just starting with something new, you’re starting with something that is truly prepared for the types of problems that a law firm would want to apply it to.”
Within iTimeKeep, narratives entered incorrectly will mean that a bill is rejected by the client, and Cartrett says: “Based on previous bills that have gone to clients, Maddi will suggest narratives and better ways to say what you’ve said. It’s often how you are phrasing things that trigger an appeal or rejection by a client. The AI goes back and looks through other narratives that have been accepted and will give you those types of narratives as suggestions to replace whatever you may already have there.”
Within Onyx, which Legal IT Insider had a preview of ahead of general release, Maddi will take the data from Bellefield OCG Live and automate OCG compliance across time, billing and eBilling. Onyx, which is hosted in Microsoft Azure, integrates with all the major financial systems.
Cartrett says: “Onyx is unique in that it has been trained by thousands of existing OCGs and how those turn into rules and guidelines; how they impact time recording; and more importantly, what happens when they are going through the billing process. There’s nothing out there that has been trained on that type of data.”
One of the issues in training AI can be that the data it is trained with is not correct, but Cartrett says: “We handle this within billing and then we also have the leading eBilling tool: the way that BillBlast works is that it pulls that information back from each of your clients and we use it to educate the AI.”
Hoyt adds: “There’s millions and millions of narratives that have gone through the billing submission and that we know were either good or bad. Like with all things, you constantly learn, so maybe this worked for one client but didn’t work for another and got rejected. That then goes back into the library and retrains it. We have millions of records of data that we are able to mine and we feel really good about the accuracy because of the rich data set we have.”
Another concern that law firms typically have around training AI is whether their client data is being shared. Hoyt says: “When you look at public AI that is certainly of concern and that is one of the reasons for our approach, which is doing it locally and privately. All of the data is contained within the Aderant domain just like all other aspects of our applications are. So when we’re training the model, we’re training it locally, and that data is not getting shared. It’s open-source but we’re training it locally and it stays within Aderant.”
He adds: “If I build an application in the cloud and leverage technology such as Kubernetes, which is an open source technology around how you manage and run containers, when you do that, those containers maybe can run on your premises or elsewhere, but it’s proprietary, it’s segregated out, and that’s really the approach that we’re taking here.”
Feedback from the user community on Maddi has been extremely positive and Cartrett says: “We have people lined up and are scheduling people for implementations. We made the announcement in Denver and on the Onyx side had people literally sign up that day, which in the legal world doesn’t happen.
“People have known for a long time that AI would have some positive benefits, but the early products to market required so much training that it was exhausting. Now with Maddi, you’ve got something pre-trained, and that’s the difference.”