File Stack and Magnifying Glass
Important detail: automation brings efficiency, but nuance is a challenge © Getty Images/iStockphoto

When a machine is analysing a legal document, every word counts. For example, the insertion of “not” — just a three-letter word — changes the entire meaning of a contract’s clause. But while teaching machines to assess legal data is not easy, the benefits of doing so — from increased efficiency to better risk management — have spurred efforts to train computers to become more sophisticated in how they interpret the written word.

“The real challenge is teaching machine learning models to understand the wording,” says Lucy Shurwood, a partner at Pinsent Masons. The law firm is developing its own artificial intelligence platform, called TermFrame, to extract, review and analyse contract risks. “Something might be worded differently but mean the same thing and sometimes one word change will fundamentally alter the meaning of the clause,” she adds.

Natural language processing — systems that interpret speech and text — is already part of this effort. However, the next step is to enable machines to make qualitative analyses of legal documents. “The golden ticket is being able to classify them,” says Ms Shurwood.

In the case of contracts, this might mean being able to identify unnecessarily obstructive clauses such as the need for a specific party’s consent before contractual obligations can be transferred. “If we can capture and understand what our clients are doing, we can advise them better because we can spot trends in the way they’re contracting,” she says.

To teach computers to make these qualitative assessments, Pinsent Masons is creating machine learning models by inputting different examples, which can take up a lot of time. “When we were doing some analysis of our contracts, we had 7m examples of clauses — and that’s just for one provision,” says Ms Shurwood.

However, as machines read and analyse more data, the process will speed up. “As new models come into existence that can understand the context and the syntax, that task is becoming easier,” she says.

Not all obstacles to handling legal data are digital. “The context is always different and the law is always changing,” says Vasant Dhar, professor of information systems at New York University’s Stern School of Business. “That makes it a messy arena for AI to operate in.”

Separating legal data from business and operational data is not always easy. Contracts, for example, contain operational data such as the names and addresses of relevant parties, as well as legal data such as change of control provisions — agreements that give a party certain rights, such as payment or termination.

Also, legal data is often scattered across an organisation. In the case of allegations of wrongdoing, those relating to harassment and discrimination might exist in the HR department, those relating to fraud and bribery in compliance, and those relating to embezzlement with the chief security officer.

“Each has different systems and nearly all [organisations] have these entirely siloed,” says Eugene Soltes, a Harvard Business School professor, who specialises in studying corporate wrongdoing.

Another hurdle is that legal data tend to be particularly confidential by nature. This means that even within an organisation the information is often unavailable to analytics experts such as the statisticians and programmers.

“These are the skeletons in the closet that, short of the legal division, outside counsel and the chief executive, people don’t know or talk about,” says Prof Soltes. “So it’s not typically exploited.”

Those that overcome the technological and organisational hurdles, however, can reap benefits that range from greater efficiency in processes and lower legal spending to better management of risk and compliance.

Analysis of past contracts might reveal that legal teams are spending too much time negotiating low-risk items, while those where risks are higher are receiving insufficient attention.

In litigation, more accurate predictions of the likelihood of winning a case could help companies decide whether and how to settle.

By examining claims data, technology can enable companies to run fraud detection investigations much faster, says Greg Schneider, co-founder and executive director of Quantium, a data analytics company.

“It’s going through a thousand cases to find ones that just don’t look right,” he says. “You can do it at [an acceptable] level of quality and time and, because of computing, it’s infinitely scalable.”

Use of data enhances management of internal risks too, says Ms Shurwood, who cites employment claims. “We can identify geographic trends or trends in types of claims that we can flag up to the client,” she says. This might reveal that discrimination claims are frequently filed against the client in one business unit, enabling it to take action.

For multinationals, regulatory compliance technology — known as “regtech” — helps ensure they are meeting legal obligations in dozens of jurisdictions.

“A big regtech idea is to computationally process these obligations and have a flag any time those obligations change,” says Daniel Martin Katz, professor of law at Illinois Tech and co-founder of LexPredict, a consulting and technology firm.

Even so, he adds, given the cognitive biases present both in humans and algorithms, technology will not provide a complete solution. “You want to correct the biases of one with the other,” he says. “If you can intelligently thread them together, that’s ideal.”

A graphic with no description

Anheuser-Busch InBev

Help digital transformation
After rapid expansion, including the $100bn acquisition of SABMiller in 2016, global brewer AB InBev increased its operations in high-risk countries, such as Ukraine. Its legal and compliance teams conducted a due diligence exercise to improve risk analysis and regulatory observance.

Working with the head of technology, data and operations, the compliance team began a data project covering 12 work streams to improve compliance in areas including the prevention of fraud, corruption, antitrust risks and money laundering. This has helped to reduce the cost of investigations. Previously, an investigation could cost up to $2m. Now, it costs a ninth of the sum for twice the scope.

Ansarada and Quantium
Inform deal and litigation strategies
Due diligence surrounding mergers and acquisitions can be costly and slow and there is no guarantee a potential buyer will make an offer. Data science group Quantium worked with Ansarada, which provides online data rooms to store and review documents during the due diligence phase of a deal, to develop a measure of how likely a suitor is to bid.

The artificial intelligence technology, launched in 2017, extracts information on the behaviour of potential bidders and an algorithm gives the probability that each will make an offer.

By tracking 57 attributes of bidder behaviour for seven days, the data model was 97 per cent accurate on a test sample of 3,500 deals. The automated process is less expensive than an investment banker’s assessment.

Improve business operations
Miner BHP used the Thomson Reuters Legal Tracker to track and code data about its spending on legal services. It identified areas with the greatest number of legal problems and highest cost. The data analysis led to BHP’s decision to hire specialists, including labour law experts, on the in-house legal team to manage these areas internally.

By taking this approach to the staffing on its legal team, the company has helped reduce its external legal spending by 56 per cent and internal legal spending by 40 per cent since 2014.

BT and iManage
Improve business operations
BT operates across 170 countries, working with legal documents in nearly 100 languages. The UK telecom group started using iManage’s RAVN artificial intelligence software in 2016 to upload 1m contracts and analyse data such as their key terms. About a year into the partnership, the company now expects the process to save tens of millions of pounds each year by ensuring BT charges and is being charged correctly. The process has allowed BT to handle more work internally, adding to its savings.

Cisco and DataNovo
Evaluate intellectual property
Technology conglomerate Cisco has acquired 13 businesses since January 2017 as part of its strategy to expand its software and subscription services. Understanding the intellectual property assets at stake is complex and time-consuming.

Cisco partnered with DataNovo, a legal analytics start-up, which allowed it within days to evaluate a target company’s patent portfolio, a process that can take patent lawyers months. The tool searches through an art archive to assess the validity of patents and find infringement. This helps in the assessment of the value of takeover targets and allows Cisco to move faster than rival bidders.

D2 Legal Technology
Deliver new business insights
The legal analytics supplier is helping global financial institutions meet new regulatory requirements and improve their management of legal documents, data and risk.

It worked with Barclays to build a new governance structure across the banking group that covers the legal information held in more than 50,000 master trading agreements and hundreds of millions of trades across many databases. D2 Legal Technology helped identify where the group’s legal data was held, connect it and improve the types of data points Barclays captures.

Huawei and Lex Machina
Inform deal and litigation strategies
Lawyers at Chinese telecoms company Huawei use Lex Machina’s analytics to choose US intellectual property litigators based on their relevant experience rather than on reputation or sales pitch.

The tool also guides litigation strategy by analysing a judge’s history of granting certain types of motions and showing whether the opposing counsel has a pattern of settling. Huawei estimates the tool has helped it save between $750,000 and $1.5m in litigation expenses in the past year.

Hub International and Exigent
Improve business operations
Offices are the second biggest cost for insurance broker Hub International, after its employees. Exigent used its proprietary software and a team of data analysts to examine the company’s 650 leases, combining it with data information from human resources and finance. The tools provided the company with greater insight into costs and helped it make more informed decisions on leases. Hub estimates it will reduce overall property spending by 10 per cent a year in the next five years.

Savills and Leverton
Deliver new business insights
Extracting financial information from property leases is time-consuming and expensive. Savills uses Leverton’s machine learning to automatically extract this information from client leases. The data are then analysed and visualised using Savills’ own K3 tool to help clients make better decisions about their properties. The automated process is 30 per cent faster than doing it manually and allows Savills clients to link the analysis to specific clauses in their original lease documents.

Westpac Group, Lawcadia and Red Marker
Help digital transformation
Australian bank Westpac is working with technology company Lawcadia to capture data on its outside legal spending. It also uses Red Marker’s Artemis software to evaluate marketing material and ensure compliance. The generated data are driving more consistent management of legal policies across the group.

Case studies research: RSG Consulting

Get alerts on Professional services when a new story is published

Copyright The Financial Times Limited 2022. All rights reserved.
Reuse this content (opens in new window) CommentsJump to comments section

Follow the topics in this article