Artificial intelligence (AI) and machine learning (ML) technologies are having a significant impact on today’s enterprise, particularly in the professional services space where they are driving greater efficiency and productivity. But before companies jump into acquiring the latest products, it is vital to understand how an AI strategy can assist with the overall objectives and solve the key challenges in the business. Only then will AI deliver real value to the business.
Legal and financial services firms are leading the way in this regard. Instead of following the trend of deploying generalized AI tools that can be used “horizontally” across many industries and workflows, an approach exemplified by tech giants like Microsoft, Google and Amazon, legal and financial organizations have chosen a more targeted approach. In my experience as a developer of automation systems, businesses that adopt a “vertical AI” strategy with a narrow focus on making incremental improvements to every day, time-intensive tasks, ultimately serve both employees and clients better.
What is vertical AI?
Systems based on vertical AI are typically full-stack products — meaning the solution is integrated “vertically” from the end-user interface all the way down to the level of data analysis and ML. The design of such systems for professional services firms requires deep subject expertise and industry domain knowledge. Vertical AI systems are usually designed by a collaborative group of industry experts, data scientists and software engineers who are working together to solve specific business problems.
In vertical AI systems, it is important to build a suitable infrastructure to implement the AI workflow. The chosen infrastructure must align closely with industry-specific processes and workflows. Also, for professional service organizations like law firms, banks and insurance companies, the security of internal data is a critical factor in selecting infrastructure. Typically, such firms have large volumes of business and legacy data collected inside their security perimeter. Such data is decentralized across various services such as CRMs, document management systems, billing and timekeeping systems, email and so on, and it is stored in a variety of formats, both as structured and unstructured data.
A primary task for a vertical AI design team is to integrate these on-premise legacy systems into a single data source without compromising the security policies of the firm. The system must provide enterprise-level authentication and support various APIs, including low-level protocols like TCP/IP. The data needs to be collected, preprocessed and used for building, training and optimizing the ML models. Building an optimal data pipeline for a firm’s AI workflow is an important part of the vertical AI solution design process. Each solution will be different, depending on the firm and the business problems it seeks to address.
Some vertical AI systems preprocess and structure the data on edge and send it to “data lakes,” which are storage repositories holding large amounts of raw data in its native format until it is needed. At the same time, important data for training the ML models should be copied from legacy systems and sent to data lakes as well. Before the training begins, the data needs to be converted to a unified format. Vertical AI solutions, then, function as a kind of a “smart” layer between abandoned data collected by legacy systems and the end-user.
Another design consideration in vertical AI solutions is the choice between cloud and on-premises infrastructure for AI. A recent Moor Insights & Strategy study notes that enterprises focused on long-term solutions have a strong tendency to use on-premise AI. One reason for this is security. Particularly for professional services firms, sending highly confidential data outside their firewalls and facilities is too risky. Highly regulated industries like legal and financial services usually prefer secure, on-premises IT. Another reason for choosing on-premises architecture is the phenomenon known as “data gravity.” Large datasets that are generated on-premise are hard to move in the cloud. The larger the dataset, the more likely it is impractical to move it to the cloud, and the more likely “gravity” will keep it on site. For that reason, it makes sense for an AI-based solution to operate near the data that it will be using.
Promising Use Cases
Today, there are already a number of vertical AI deployments in both the business of law and the practice of law. Vertical AI is widely used for email and document management, legal analytics, due diligence and contract analysis. A good example is Kira Systems, which is machine learning software that helps attorneys to perform more accurate and much faster contract review by identifying, extracting and analyzing relevant content. BlackBoiler, another AI-assisted contract review platform, has saved users substantial amounts of time by automating the part of the review workflow that is repetitive and redundant. Ultimately converting time into measurable cost savings.
Another example of vertical AI is COIN, a software solution developed in-house by JPMorgan that automates contract review for certain classes of contracts. The software uses image recognition to identify patterns in the agreements and can extract 150 attributes from 12,000 commercial credit agreements and contracts in only a few seconds. According to the JPMorgan, that is equivalent to as much as 360,000 hours of work by lawyers.
While vertical AI presents a number of unique challenges requiring a strategic approach to solution design and intensive collaboration by a team of experts from diverse disciplines, the payoff can be spectacular for firms seeking to automatic basic, labor-intensive and often tedious tasks in the services environment. To date, horizontal AI has attracted most of the attention in the business press, but vertical AI is poised to have a much larger impact on efficiency and the bottom line in the long run.
This article originally appeared in Forbes Technology Council.