Serious Fraud Office CTO Ben Denisonhas turned to an AI robot to help review the abundance of legal documents it processes in pursuit of serious financial criminals.
The SFO analyses more than 100 million documents annually in the course of its investigations into major cases of fraud and corruption, which last year included bribery claims against Rolls-Royce that resulted in the largest ever fine imposed in the UK for criminal conduct.
Identifying the relevant information can cost the specialist prosecuting authority millions of pounds and thousands of human hours. The introduction of the LPP (Legal Professional Privilege) Robot can cut the costs to one-fifth of the human review.
“What we find is that when we’re investigating corporates in particular, we seize huge amounts of information in the course of our investigations,” Denison told CIO UK.
“The robot helps us to identify material that is covered by legal professional privilege, which we can’t use. It allows us to use technology to help identify this material and that makes the process much faster, and much cheaper. On the Rolls-Royce case where we have used it, it saved us 80% of the cost of reviewing it manually.”
Expanding volumes of data
Investigations undertaken by the specialist prosecuting authority can last for years and span multiple continents. As a comparison, the Panama Papers data leak of 11.5 million files was the largest leak in history and needed 400 journalists in 80 countries to spend more than a year working through that material before anything was published. That is an average-sized batch by the SFO’s standards.
The SFO’s investigation into Rolls-Royce trebled that to 30 million documents. It required 70 investigators to review them over four years before a £671 million settlement was agreed.
The volume of data is growing exponentially as cases become more complex. Denison believes that one in the pipeline is likely to reach around 100 million documents, stretching the SFO’s budget and workforce capacity.
“We’re using it on other cases now as well, and in time as our cases get bigger and bigger this sort of thing is going to be absolutely essential for what we do because it’s just impossible to review all this material manually when there’s so much of it,” he says.
“The average processing rate of those barristers is 300 documents a day, whereas the robot at its peak was getting through 600,000 a day, so it’s a significant difference in terms of speed of processing.”
The system is currently being used only to identify the communications between a client and lawyer, which as privileged information that must be excluded from an investigation.
Detecting this material is a far less time-consuming task for robots than it is for barristers. The LPP’s error rate of around 0.02% is also lower, and its accuracy and efficiency improve as it learns and is tweaked over time.
“It’s not necessarily that the humans get it wrong, but if you’ve got 10 or 20 different lawyers making judgements about something, it’s less consistent than if you have a robot doing it because you have one set of rules, one concept applied over and over again,” says Denison.
“It still goes to the lawyers for review, so there still is an element of human review as part of it, and we don’t see that ever going away. The robot just means that we make better use of the lawyers time as they can take a more targeted approach.
“It will lead to even greater efficiency savings, because if you can put the relevant material in front of the investigators at a much faster speed then it helps us to speed up charging decisions, and when we’re going to proceed to trial, things like that. It speeds up all of those processes.”
Convincing the doubters
The platform was produced by RAVN Systems, a London-based company whose software won a techies award for best enterprise startup in 2016 before it was adapted to meet the SFO’s need to identify privileged material.
Typically the information is reviewed by a selection of lawyers, accountants and investigators. Any sceptics among them were converted to the AI cause following an explanation and demonstration of the system, backed with metrics on its accuracy and an understanding that it would remove an often mundane and repetitive task.
“I don’t think it’s ever going to completely replace humans in a task like this,” says Denison. “What it does is automate as much of the process as we and the defence teams are confident with, and allows us to just make better use of resources.
“For us, that’s really important because we do have limited resources, we have fixed budgets and headcounts as others in the public sector do. In time I think it will be a standard part of our toolkit and we will use it in every case.”
The AI robot is one of a number of digital solutions to legal problems that Denison has helped introduce. The SFO is also replacing the evidence review system that is the primary tool used in SFO casework with an e-discovery platform called OpenText Axcelerate with built-in machine learning capabilities.
The SFO’s core budget of £36 million leaves it unable to compete with City law firms on salaries, but the authority can offer staff unique cases and a superior work-life balance that automation gives them more time to enjoy.
“A draw for them is the quality of the work and the quality of the experience they get from working here, so you will find a lot of lawyers and others will come here from the private sector, and they may go back to the private sector afterwards, but they’re able to get experience here that they wouldn’t find anywhere else,” says Dennison.
“For example with digital forensic, there is a huge market for that in consulting firms, in forensic accountancy, and of course in law enforcement too.
Integrating systems and improving collaboration
Denison joined the SFO in August 2014, after more than four years as CIO of The Bar Council, the UK’s professional association for barristers.
He recently took over a number of other teams in the SFO, including those that manage and process the evidence. An integrated IT setup helps align and coordinate the variety of work undertaken across the authority.
“What we’re able to do by bringing the teams together is to look at any scope for further integration of processes and systems. The hope is that that will allow us to operate more efficiently and in particular through the use of technology where we can streamline some of those processes.
“One of the things about the public sector is I think we’re better at collaborating because we’re not competing with each other and so you’re able to learn from colleagues and peers and put that to use to help benefit all of us.”