Contract lifecycle management vendor Agiloft today (9 January) announced new generative AI functionality that suggests redline alterations to contract clauses to speed up the negotiation process. Third party clauses that are substantially different to a corporate’s clause library can be redlined to make them more ‘mutual’, with a further option for more free form generative AI prompts.
The new functionality, surfaced in Microsoft Word, is targeted at users who are performing contract negotiations, whether they be lawyers or contract managers empowered to negotiate contracts in their area of speciality.
It builds on the AI platform launched by Agiloft last August. Agiloft’s chief product officer Andy Wishart told Legal IT Insider: “We know from our product discovery and research with customers that redlining is a pain point. It’s time consuming for users to go through the process of looking at revisions between a third party and the standards that they have set within their CLM system, either within templates or within the clause library. Those pain points can manifest themselves in long cycle times and potentially lead to lost revenue and opportunities as well.”
When faced with an agreement, negotiators can use the AI contract assistant launched last year to automatically tag key terms in the document, which appear on the right-hand side of the document. Users can – again within the functionality launched last summer – select a clause such as force majeure and open it in a task pane, where they can benchmark it against the standard clauses in the playbook and potentially swap it out.
However, Wishart says: “Sometimes the difference between those clauses is great, with significant deviations between the clause in the current contract and the standard clause. So what if we could use AI to recommend insertions and deletions to better align these clauses? That is the new capability.”
In the demonstration given to Legal IT Insider looking at materially different force majeure clause in a services agreement, Wishart said: “It’s going to recommend a change to the clause, so it’s adding in pandemics. It’s adding an additional provision that allows the unaffected party to terminate if the event lasts for more than 30 days and the user can select that change and apply it to the document as a red line.”
He adds: “If I select an indemnification clause and say make that mutual, it’s going to recommend changes in order to switch that clause over to a mutual provision and we can apply that red line into the document. It’s all about using the generative AI capabilities to assist the negotiator in speeding up the red line but they are still very much in control. We provide a summarisation of why it suggested these changes and they are in control of then applying them to the document.”
Some may be confused over the ‘old and new AI’ but Wishart says: “We would use AI to detect clauses to match the ones in the document with the ones in the clause library and provide the ability to do a straight swap out but that is a pretty bold move for a lawyer to say, ‘Ok, I’m not going to accept your clause whatsoever. Here is our standard clause.’ This new functionality is saying, ‘Ok, we understand that these are two different things, let’s align them a little better. We are trying to replicate what a lawyer would naturally do, which is take two things that are different and suggest insertions and deletions that bring the parties closer to their standard and speed up the process.”
If the hype of 2023 taught us anything, it is to ask whether generative AI is really needed. Wishart says: “There’s an area within the negotiation process where the differences are significant enough that going through manually to make those changes would take some time, because the lawyer has got to look at what’s being proposed, look at their standard, and figure out from those two things what insertions, deletions, what revisions, and red lines they need to make to the clause, and that takes time.
“If it is something really simple like their standard payment terms are 60 days, but the contract says 30, it’s going to be quicker for the lawyer to just go in and make that change from 30 to 60 directly into the document. But if it’s a clause that has more significant differences, then gen AI does a good job in recommending the most important aspects to insert and delete. So you’re avoiding the lawyer having to take a look at the differences between these clauses and work through making the revisions. They’re getting a head start on that.”
Wishart stresses that Agiloft is not suggesting we are at the stage where gen AI can automate the redlining process, but says: “What we’re talking about is providing tooling to experienced negotiators who understand the legal language, who understand the commercial language within those contracts, to make recommendations for them, which they can then adjust. So when gen AI makes a recommendation on the red line, they can go in and edit it and then apply that change so they’re always in the loop in terms of reviewing what the gen AI has recommended.”
Agiloft is using a fine-tuned OpenAI GPT 3.5 model, which Wishart says has brought the best performance. “The fine tuning means that we’re able to use our legal knowledge engineering team to provide examples to fine tune the model to say ‘here is an original clause, here is the standard clause, here is the suggested output.’ The fine tuning process has resulted in better performance based on real world scenarios.”
He adds: “We can be agnostic to models, so for further use cases we may look at GPT 4. We may look at Claude, we may look at falcon-40b. We’ve done a lot of experimentation of integrating with multiple models. So our approach will be we’ll use the best model at that point in time that is best performing for that use case.”
Agiloft will be providing early access for a select number of customers over the next few weeks, enabling the product team to gather feedback and make modifications. Wishart says: “They are looking for feedback on the overall user experience.”Agiloft is looking for customers of varying size across the industry, but particularly those that have invested in building out their clause library, where Wishart says: “They’re going to see earlier returns on their investment and see the benefits more quickly.”
While Agiloft is not the first to introduce redline capabilities – other vendors include Ironclad, Casetext CoCounsel, Evisort, BlackBoiler and LegalOn Technologies – what is good is that they have been thoughtful in this release. Wishart says: “We could have brought gen AI into the product just to throw something against the wall and see if it sticks but that’s not what we’re trying to do here. We have spent a lot of time researching with customers across a multitude of use cases before deciding to go down the gen AI route; we spent probably more time on that research than the development itself. So throughout the last year we surfaced around about 18 potential use cases within the contracting domain where we had a hypothesis that gen AI could potentially bring some value. We tested those use cases with real customers to get feedback. And that feedback did indicate that the greatest value seemed to be around the negotiation process because of the pain points and the pressure to be able to turn things around.
“Red line is the first use case but there are other use cases that we’re pursuing. We’ll begin to see more of those throughout the year but it’s so important to do the hard product work of researching with customers to see where the value is.”