Much has been said and written about the impact that generative AI will have on more traditional automation, such as document automation.
While some have forecast the end of document automation, others have suggested it will have a more positive impact. For example:
- Allowing people to draft or improve their contract templates more easily than has previously been the case. The absence of good templates has often been a key reason why document automation projects haven’t had uptake or been successful.
- Using AI to automate the automation of templates. AI could review a document or template, identify the areas where document automation is likely to need to insert variables and have a first go at creating those variables. This could greatly speed up the time people like us need to create templates.
- Similar to the point above, making it easier to integrate document automation into broader business flows via tools such as PowerAutomate or Workato.
- Taking a document that has been assembled by document automation and making it easier to make further changes for more unusual situations (e.g. word add-ins that enable the user to look up approved clause banks and to make changes that it wouldn’t make sense to try and automate with traditional document automation).
- Related to the above, enabling the negotiation of documents generated via document automation without the need to go back into the document automation tool and re-run an entirely new set of documents.
These are all good and exciting things I’m looking forward to.
What I wanted to look at today, though, is another specific area, which is how generative AI can be incorporated within the user experience of a traditional document automation workflow.
As many of you will know, a typical contract has several section types:
- There are a lot of sections that are actually just basic data. Party names, addresses, roles etc. These are mundane and don’t add much value, but they must be entered accurately. I believe these should ideally be pulled from other systems via integration.
- Optional sections, where paragraphs (or perhaps just words) are included or excluded based on logic and how questions (or combinations of them) have been answered. Good document automation can handle this really well. Hundreds, or thousands, of sections of text can be manipulated 100% accurately and consistently without any need to go back and see how an AI black box has done things and if it has gone off track. We want the exact same text to be included every single time, particularly for high-volume business processes.
- Larger free-text sections, where there isn’t a pre-canned output and the user has to think and then enter something (such as a description of services in a commercial contract or a job description in an employment agreement). This has always been challenging for traditional document automation because there’s not much benefit (if any) to entering the information onto a screen rather than into the document. At least in the document, you can see the wording in context and can apply formatting etc.
It’s this last section, where I see some great potential for a hybrid approach, and why I’ve been so keen to explore NetDocuments’ PatternBuilder MAX AI solution, as it offers a combination of traditional document automation and AI.
It’s early days in my understanding of this capability, but I’ve recorded the video below, which shows the direction I’m interested in taking.
I think there is an opportunity for this technology to not only speed up the creation of these free text sections but also greatly improve the quality of them.
One thing I’ve observed (and been guilty of in my own career as a lawyer) is that it is really hard for lawyers and contract drafters to come up with things like service descriptions, milestones and deliverables, and expected outcomes. We just don’t know enough about all types of business to know what good might look like. If we did, we’d probably have been able to template more of it into standard options.
The ability to call on generative AI to review:
· A prompt setting out key points; or
· another document(s) e.g. an RFP, a supplier response, or a term sheet,
and to use that to suggest an initial draft could be game-changing. This level of creativity is something that we are often lacking when faced with a blank sheet of paper.
System prompts can work in the background to lighten the load on individual users and lessen the need for them to become prompt experts. The workflow can determine the appropriate AI model to use, and the settings to apply.
Once we have a draft in front of us, it’s often easier to work with and flesh it out.
You can see my initial example in the video. What do you think?
If you have real-world use cases or scenarios you’d like to explore, get in touch. I’d like to focus further efforts and testing on realistic use cases as that will help to tune the approaches and settings to getting the best overall results.
If you’ll be at the ALPMA Summit in Brisbane from 11 to 13 August 2024, drop by the LawHawk exhibit stand, and we can talk about any areas of particular interest.