KEY TAKEAWAYS:
- Know your AI tool inside out—including how it was trained and its limitations, to ensure it aligns with your needs.
- Simplify and structure your contracts to maximize AI performance, making it easier for the tool to accurately extract and analyze key data points.
- Combine AI with human oversight to catch errors and nuances that AI might miss so that you ensure an effective process.
One of my passions outside of work is cooking. I love finding and following recipes to create the best dishes. Imagine you just purchased a new AI contract review tool, but instead of learning how it works and following best practices, you continue using your old methods. It’s like trying to use a new mac and cheese recipe but ignoring the instructions because you’re used to the boxed version. The result? Your dish, or in this case, your contract management, will fall short of expectations.
Instead of repeating a cycle of disappointment, learn how to optimize your processes to maximize on the benefits of the new tool. When it comes to AI contract review tools, the structure and layout of the contract templates that you upload into the tool for analysis and extraction are critical. This article lays out five tips for drafting contract templates to improve how easily and accurately they are ingested and used by AI review tools.
Understand How OCR Scans Your Contracts to Feed the AI
Before jumping in, let’s step back to the bird’s eye view of AI in CLM. To grasp the strengths and weaknesses of an AI-driven CLM tool, understand how AI “reads” contracts. Optical Character Recognition (OCR) is often the first step in AI-driven contract analysis, especially for scanned documents. Understanding how OCR works—and its limitations—is crucial.
OCR converts scanned images into machine-readable text, but it’s not perfect. Complex layouts, low-quality scans, or multilingual documents can trip up even the most advanced OCR systems, leading to errors in the data fed to the AI. If the OCR misinterprets text or fails to correctly capture the document’s structure, the AI might draw incorrect conclusions. The following tips will ensure you get the most out of your tool by conforming to the best practices of OCR and AI in contract review.
1. Don’t Use Tables
Tables in a contract are easy for humans to read but can confuse OCR. With AI, a table’s structured data might be flattened into a list of text strings, mixing headers with data points and disrupting the AI’s ability to extract accurate information.
Want proof? Look at these two examples of the same information captured by a paragraph, then by a table:
Upon first glance, the table is much easier to decipher than the clause written out in paragraph form. Due to this, many companies structure the details of their agreements in table format. It’s simple to review the contract and identify the key details. However, when subjected to an OCR scan that converts the PDF image into a text string processed by AI, the table produces something like this:
If that sounds confusing, think about how tricky it would be for a non-human trying to extract the key data points and display them correctly. Spoiler: it’s not a smooth ride. If I’m in luck, the AI might successfully identify the Effective Date and Term, but chances are it will miss out on the Renewal Type and Notice Date. These are crucial details for getting good output from your AI tool. So be to sure to consider how intricate formatting affects your tool’s efficiency. While tables might be easier for us humans to read, plain text is easier for AI to read.
2. Don’t Use Columns
Other special formats can also cause problems. The notorious two-column page layout might reduce your page count, but some OCR tools may struggle to convert two-column page layouts into simple text. Many OCR tools are designed to process text in a linear, left-to-right manner, which can lead to issues when handling multi-column layouts. The tool might read across both columns like a single block of text, resulting in messy or nonsensical output.
3. Don’t Use Special Fonts
Special fonts, such as those that are highly stylized, have unusual spacing or include embellishments, can cause the OCR tool to misinterpret letters or numbers. This can result in incorrect characters being identified. Or the tool might fail to recognize the text altogether. Helvetica is the world’s most popular font for a reason.
Learn More: The Recipe for Better Contract Design
4. Keep Headers and Footers Simple
Headers and footers pose challenges for OCR tools, as they can disrupt text flow between pages and misinterpret contract language. This issue is particularly common with eSignature tools like DocuSign, which adds a footer to your signed document, causing examples like this:
“NEWCO shall indemnify OLDCO and its Affiliates and their officers, directors, employees, and agents from, and defend and hold such parties harmless from and against, any Losses suffered, incurred or sustained by such parties or to which such parties become subject, resulting from, arising out of or relating to the following: 1. the inaccuracy, untruthfulness, breach or alleged breach of any representation, warranty or covenant made by NEWCO under this Agreement; 2. personal injury (including death) or property loss or damage resulting from NEWCO’ or NEWCO agents’ acts or DocuSign Envelope ID: C035BOF2-C65C-4EF0-8E25-3ED9B6834088 omissions; 3. the negligence or willful misconduct of NEWCO; and 4. the failure of NEWCO to comply with any applicable law.”
Including the DocuSign Envelope ID will interrupt the AI’s capture of the (extremely complex) indemnification clause, which reduces the user’s ability to quickly identify each item for which NewCo will indemnify OldCo. The main point is to keep the header and footer simple.
A quick tip: You can turn off DocuSign’s tag on the document in their settings.
5. Don’t Use Overly Complex Legal Language
Introducing AI into your workflow can maximize efficiency, but overly complex legal language can have the opposite effect if your contract language looks like a $ 1,000-per-hour outside counsel wrote it. AI might struggle with intricate clauses, leading to errors in data extraction.
Complex Drafting
Let’s look at an example of a Limitation of Liability clause, which is found in most contracts and is the most negotiated section of any agreement.
“Notwithstanding anything to the contrary contained herein, except in respect of liability arising from gross negligence, willful misconduct, or breach of confidentiality obligations, under no circumstances shall either party be liable to the other for any indirect, incidental, consequential, special, or punitive damages, including, without limitation, loss of profits, revenue, data, or use, even if such party has been advised of the possibility of such damages, whether in an action in contract, tort (including negligence), or any other legal theory, arising out of or in any way connected with this Agreement, including but not limited to the performance or breach hereof, and regardless of whether such damages were foreseeable.”
Learn More: 10 Ways to Integrate Plain Language into Contracts & Advice
How AI Reads It
This clause is packed with legal jargon, multiple exceptions, and complex conditions. Here’s how such a clause might confuse AI:
- Multiple Conditions: The clause includes several layers of exceptions that the AI needs to parse accurately. Misinterpreting or overlooking these exceptions could lead to incorrect conclusions about the liability of the parties involved.
- Legal Terminology: Terms like “indirect, incidental, consequential, special, or punitive damages” are specific legal terms with particular meanings that may vary depending on the jurisdiction. If the AI isn’t trained on this specific legal terminology, it might fail to fully understand the implications of these terms.
- Nested Structure: The clause’s nested structure—combining multiple concepts within a single, extended sentence—requires a deep understanding of legal syntax. AI might struggle to correctly associate the various elements, leading to errors in data extraction.
- Contextual Understanding: Phrases like “including, without limitation” introduce additional concepts that further complicate the AI’s task. The AI needs to understand that these phrases are meant to broaden the scope of the clause, not limit it.
If AI fails to accurately parse and interpret this clause, it could result in an incorrect risk assessment or other misunderstanding. This could lead to flawed risk management or slow down the speed of your negotiations. Using modern legal language in your contracts is essential to the modern contracting process, both for AI tools and for the counterparty.
Ensure Human Oversight Because AI Isn’t Perfect
AI tools are revolutionary, but they’re not perfect. You’ll still need human oversight to ensure the accuracy of the data extracted by AI. No AI tool is 100% accurate, and relying solely on AI without human review can lead to missed details or misinterpretations.
