Best practices when preparing to adopt a contract review and analysis solution
Law firms rightly think long and hard about whether to implement new software to increase efficiency and deliver optimal client service cost effectively. When it comes to artificial intelligence (AI)-based contract review and analysis (CR&A) solutions, the thinking is no different. Once leadership and the IT team have made the decision to go ahead, it pays to take the time to prepare for what is involved internally to fully maximize the gains.
Here, we look at five key issues to consider for a smooth rollout and what best practice looks like to optimize performance.
Selection and integration
It sounds obvious but choosing the right CR&A tool for your firm’s — and clients’ — specific needs at the outset is critical. The analogy Product Manager Steve Fullerton at Thomson Reuters uses is, “You wouldn’t just pick any car to go off roading, you need a 4×4 with the right tires and sufficient ground clearance. If you try to do it in a family car, you won’t get very far. You may think one product does the same thing as another, but they don’t.”
This means thinking carefully about the various projects you will use the CR&A solution for and what it needs to deliver for clients. What will the different use cases be and what are the expected inputs and outputs for each? For instance, what kinds of contracts will need to be reviewed — are they single or bulk documents and what information needs to be extracted?
Having selected a solution with the right capabilities, the first step in deployment is to work out how best to prepare existing internal systems that will feed into the new tool for seamless integration. Consider what data needs to be made available and how content will be transferred from those internal systems.
Configuration and use case development
The next step is to configure the tool — and therefore the work product — appropriately. This is where doing the right groundwork in advance should really pay dividends in ensuring you get the right results for users and content for your client reports.
In practical terms, this means considering issues like how the information is automatically classified and organized into folders once documents are input into the system. After that, you will need to decide how the automation software should triage that information and what triggers and workflows need to be in place to deliver the right data, instructions, or tasks to the right people at the right time.
Fullerton advises that it can help to start at the end, by defining what you want to achieve and working backward to establish how to get there. Otherwise, “You might set things up a certain way, expecting to find this information or that data, and then you get to the end of the process and realise you’ve not captured everything required.”
CR&A tools will need to be configured differently for different use cases and although setup for some use cases may be fairly straightforward, firms should not assume they can take a one-size-fits-all approach.
“Different projects will have different contract types that need organizing and analyzing, plus the volume of documents will vary,” Fullerton explains. “So, an audit of hundreds of post-transaction documents for multiple deals will be a very different use case than handling documents for a single transaction, such as a portfolio sale or purchase. And then when you’re doing things like M&A due diligence, you’re likely to have multiple people working on a range of contract types, so there are even more variables.”
Templating and more
That said, Fullerton advises creating templates where possible, so that once a configuration has been created, it can be re-used many times over — albeit with adaptation where necessary — either on a client-by-client basis or a project-by-project basis, for consistency and ease of use. This could include everything from the setup steps used, to the definition of what information you are looking to capture, and in which format you report on that data.
Fullerton adds, “For certain projects that require something that’s over and above what’s ‘out of the box,’ there may also be a case for using your firm’s own standard legal wording to amend or extend the tool’s machine learning models or using in-house example documents as ‘training’ materials so the AI can better understand what data it needs to find.”
Managing source data issues
Finally, beware of poor-quality source data, particularly in historic files. Optical character recognition (OCR) systems may struggle to read handwritten data in older, printed documents such as signatures, party names, dates, or notes, or to understand non-contiguous text layouts such as columns or unusual sentence or paragraph structures. “You can get some unexpected results,” says Fullerton. “To mitigate this, try to digitize data in the optimal way. High resolution scans can help, but additional human oversight of documents and how well they are represented in the digital file may be required to ensure that the right data has been extracted.”
“Again, it all depends on the nature of the task. If you’re just looking for certain clauses, it’s easy to find those even in lower-quality documents, but if you’re looking for more discrete data like key dates or legal obligations, then be aware you may not get 100% accuracy.”
A massive leg up
In many scenarios, smaller firms especially will be able to use these tools off the shelf and simply “click and go.” However, for some types of use cases with more complex requirements, particularly in larger firms, more preparatory work will be required upfront to set the foundations for successful deployment. Therefore, it is important to understand and set expectations around what is possible and what may be required for success.
Once you have selected the right tool and developed a streamlined, repetitive methodology for contract review and analysis, you should see significant efficiency gains, with lawyers having to do far less detailed laborious checking. As Fullerton puts it, “It’s a massive leg up.”
Five top tips for success
- Select your CR&A solution carefully — they do not all have comparable capabilities
- Define what projects you will use it for and the expected inputs and outputs
- Prepare internal systems to ensure smooth integration and data intake
- Configure the solution appropriately for different use cases but use templates where possible
- Consider whether any additional post-AI extraction verification is necessary to reaffirm what the tool has found