Though e-discovery market in the U.K. isn't as mature as it is in the U.S., the country is making strides bringing e-discovery into the mainstream. The U.K.'s Serious Fraud Office (SFO), for instance, recently announced that it has brought in OpenText's AI-powered Axcelerate document review system to help with its investigative efforts.

To be sure, the agency is far from the only government entity leveraging e-discovery technology in its operations. But the SFO is not your average e-discovery buyer. “We are not a law firm, we are not an in-house team, we have requirements that are specific to us because we are an investigator and prosecutor,” said Ben Denison, chief technology officer at the SFO.

Even among U.K. regulators, the SFO is unique. Whereas most regulators will handle a high volume of cases, the fraud agency focuses only on a select number. “We take on seven to 10 new cases each year, and each case can take five to seven years,” Denison said.

What's more, the cases can include enormous amounts of documents. A fraud case against Rolls-Royce PLC that was settled in 2017, for instance, had around 30 million documents, one of the largest cases the agency ever handled, Denison said.

Given the SFO's requirements, its demand for artificial intelligence (AI)-powered e-discovery technology is unique. The agency needs an e-discovery tool that can be responsive to documents that include privileged content. Denison noted that in any case the SFO takes on, “companies will turn over emails, servers and laptops, which will include privileged material that we can't use as part of our investigation.” Streamlining the process to identify privileged material, therefore, is a key criteria for the agency.

Another requirement is that any tool be able to adapt to the SFO's specific evidence handling workflow and integrate with all the systems already in use.  After seizing materials from a company, “that material is all then put into evidence back and tracked into our evidence systems… and that approach is different form a regulatory and a law firm,” Denison said.

Any e-discovery system the SFO uses has to be capable of tracking a large amount of documents, so the agency looks for platforms with advanced analytics tools, “which really help you cut through large amounts of data,” including email threading and graphical representations around communications between different parties, Denison said.

The SFO has used other AI-powered e-discovery technology besides OpenText's Axcelerate. The agency for example, had an ongoing relationship with RAVN Systems when it was handling the Rolls-Royce case. RAVN was able to adapt their AI-powered Applied Cognitive Engine (ACE) to help SFO review the Rolls-Royce investigation documents.

The platform was a great help for the SFO because it could “go through and say with certainty what documents were not privileged,” Denison said. He added that, with the technology, the agency was able to review documents faster and for about 80 percent cheaper when compared to manual human review.

The SFO was also able to quickly train RAVN to identify content that was not privileged “because we were doing the manual review for some time, and we had a fairly large seed set we could work with,” Denison explained. But training AI e-discovery platforms to find privileged content in other data sets, he added, required them to create entirely new seed sets.

“What we have seen with the privilege example is that it varies form case to case, so you would effectively have to come up with a seed set for each case and retrain it,” Denison said.

Beyond RAVN, the SFO looked to bring more AI e-discovery capabilities in-house when it decided to update its old e-discovery platform, which it had used since 2009. It reached out to a range of vendors with its requirements and eventually whittled the choice down to three e-discovery platforms. Recommind was chosen by the agency—though the SFO soon realized they would be working with OpenText. The day the bid was accepted “was the day OpenText announced they were acquiring Recommind,” Denison recalled.

Denison noted that the new OpenText platform met all the agency's requirements and that the SFO is deploying it for a case that has overtaken the Rolls-Royce investigation in terms of sheer number of documents to be reviewed. The new case “has just passed 50 million and continues to grow.”

The SFO has also been impressed by how much help AI has been and believes it will be a game-changer for the agency in the months and years to come. While predictive coding in the U.K. has only thus far been used in civil cases, the agency believes it's only a matter of time before it makes its way across the country's judicial system.

“Predictive coding isn't used in criminal case in any jurisdictions … but somebody is going to be the first to do that, and it may be us,” Denison said.