Though much has been said about the cost savings that artificial intelligence and machine learning can bring to legal, there has been little discussion of the resources needed for a legal department to implement the technology.
Speaking at the Legalweek: The Experience 2017 Conference, Connie Brenton, senior director of legal operations at NetApp and chairman of the board of the Corporate Legal Operations Consortium (CLOC), cautioned against overlooking these challenges . She noted that while many are attracted to the novelty of AI, such technology takes "significant resources to get up and running."
Lessons From Cisco
While the resources needed for AI implementation will depend on a variety of factors specific to each company and their respective law departments, most looking to deploy AI will have to take into account similar cost considerations. Jennifer McCarron, technology program manager at Cisco, offers an example of what this may look like.
Cisco is "designing a pilot" for AI applications in the virtual assistant and contracting space, she said, adding that the company has previously had a relationship with legal services provider Riverview Law, but also that such efforts are only preliminary.
In considering AI, McCarron noted that one must always take into account the sometimes significant financial costs behind licensing and purchasing the solutions themselves. With Riverview Law's KIM virtual assistant, which can automate and create in-house workflows and processes, she estimated the "starting point for pricing" for 10 users at around $30,000.
While exact costs may change with discounts or bespoke implementations, Riverview Law's website noted that the "Foundation" cost of KIM was $12,250 for 10 users, although additional upgrades classified as "Professional" costs added to that base price.
For example, to have the solution tailored to a workflow or business processes, or fitted with "templates for auto-generated documents," costs $22,000 per enterprise. In addition, integrating KIM with existing systems costs $30,000 per enterprise, while the ability to tailor or add dashboards in the solution would run an organization an additional $37,000 dollars.
While the "Professional" costs are one-time expenses, the "Foundation" costs are per user, meaning that for companies like Cisco, "who [potentially] has hundreds of users," the price will dramatically increase, McCarron said.
While it's hard to pinpoint exact costs, McCarron estimated "to get something basic like a pilot of a virtual solution that solved some basic routing workflow system, for a 500-person user base like at Cisco, I would guess that would be around $250,000."
She added the company would also need to factor in additional costs for personnel, as implementing such technology would require the hiring of "two-and-a-half full-time employees internally," as well as "additional help from the vendor side. They would assign a solution architect and a project manager."
Further, such expenses would only cover deploying basic virtual assistant processes using KIM. For more high-level tasks, such as organizing and automating contracting and document management, deployment costs would increase. This is because the technology would have to be deployed over unstructured data—data that is yet to be tagged, classified, or understood in a digital format.
For large-scale projects with unstructured data, "it's going to require a little bit more complexity," McCarron said. "You're [talking] half a million dollars and up for these kinds of systems."
More Than Just Costs
But financial expense is only one of the costs incurred when turning to AI. Setting up such a novel and complex system also requires ongoing time commitments from many in-house and vendor teams.
"I would say end-to-end, you're talking five to six months for the smaller of the pilots of virtual assistant" to get from pilot to actual implementation, McCarron said. She noted this timeframe includes development work, sourcing infrastructure or cloud hosting, and clearing the project through various in-house security and compliance requirements.
And given that AI technology has to actually "learn" information, deployment will also mean devoting time and resources to helping the machine accrue the knowledge it will need to perform its work. Doing so can mean having professionals essentially teaching the system how to handle key pieces of legal knowledge like regulations, contracts, or other purposes for which the technology will be used.
Granted, the time and resources needed to effectively train such technology depends on the complexity of the tasks for which it will be employed. McCarron noted that in recent years, she has seen tremendous improvements in the abilities of machine learning.
"What I have seen from all the vendors is that it's very fast, and it doesn't seem longer than a few weeks to get a functional data set running. Once you have all the tech in place pouring over the contract, it seems to be very quick," she said.
Because of the heavy lifting and dedicated resources an AI implementation can take up, most early adopters are likely to be large corporations for whom AI can provide the most benefit for its cost. In addition to Cisco, McCarron noted that there are several other "larger behemoth" companies road mapping and implementing AI projects, noting Google's work to bring AI contract solutions from Seal Software into their legal operations as an example.
But medium-sized companies like PayPal and eBay "are definitely not doing it" yet, she added, an indication that the resources needed for AI may still be too cost-prohibitive for some.
So while the technology is still young and the market still evolving, for now, excitement over AI's ability to greatly modernize the legal industry is likely to be tempered by the reality of getting it up and running.