Follow ACC Docket Online:  

Making AI a Reality: The Promise of the AI Process

In the second part of the “Making AI a Reality” series, authors Prashant Dubey and Bjarne P. Tellmann discuss using AI to create contract hierarchies. Read the first part of this series here.

I ntimidated by the task of sorting and organizing their pile of contractual data, many companies are tempted by the lure of software tools that purport to intake raw documents at one end, apply AI, and spit out organized contract documents at the other end, fully enriched with metadata. This is a fantastic promise that, at first, seems plausible.

In fact, contract documents are just unstructured text in a user-created file. Identifying key terms in contract documents based on certain characteristics or surrounding terms that typically exist around target terms seems like a plebeian technical challenge. However, what comes in the way of this seemingly utopian approach is contract management itself.

An augmented intelligence approach to a high integrity contract repository

Contracting is messy. Despite best efforts, contracts are rarely straight-through. Unless you are “Ginormous company X,” the probability of sending a counter-party your template agreement and getting it right back, gleefully executed, is rather slim except, perhaps, in the mutual NDA context.

The messiness of this negotiated reality means that the end-state contract that gets signed is often very different from the original template. Moreover, in many cases, the starting template will be on third-party paper, meaning that there will be absolutely no structural similarity between two executed contracts of the same type.

[Related: Beyond Data Collecting: How to Protect and Leverage Big Data]

Complexity further abounds as a result of corporate reality. Industries are consolidating, and M&A activity is resident in almost every industry. This means that many companies will have multiple contracting entities in their organization that maintain relationships — and contracts — with the same counter-party.

How should contract hierarchies be determined in such cases?

Contracting relationships are typically represented by multiple documents in a hierarchy, such as master agreements, amendments, statements of work, data privacy addendums, and other related documents. The inter-relationship between these documents needs to be fully understood and articulated, even in order to determine something as simple as whether the agreement is still active or what the expiration date is. Furthermore, in order to undertake any of this, all documents in the contract family need to be present. In most companies achieving close to 80 percent on this measure the first time around is elusive.

As proficient as technology is in identifying provisions in documents (though the level of precision/accuracy is nowhere near 100 percent and therefore requires human confirmation), the messiness of the reality of contracting creates natural limitations to the utopian vision of “any in, push a button, any out.”

This is where augmented intelligence trumps artificial intelligence, as the ideal technology enabled approach to create a high integrity repository.

What type of augmentation is needed?

At the start of an effort to create a repository of existing material obligations, a company needs to gather all suspected contract documents from the source(s) and put them into one place. At this point, not much is known about these documents other than that they may represent, individually or collectively in some combination, the relationships of the company with counter-parties.

Once gathered, technology can be used to identify the document type and cluster them based on similarity. This is helpful, but human intelligence still needs to augment this process by verifying the clustering into like topics. Is an amendment truly an amendment or is it actually a second master?

After that is done, these documents need to be organized by counter-party. Sounds simple, right? Just run some software over them, extract the counter-party name, and voilà! The contract documents are grouped by counter-party. Not so fast. Counter-parties are often spelled differently. ACME could be Acme Inc., American Corporation for Mellifluous Energy, or other strings of words we haven’t used since the seventh grade spelling bee.

[Related: Realizing ROI from Contract Management Technology]

The technology needs to be guided through inputs defining some sort of taxonomy (or “bag or words,” depending on the technology employed), then there needs to be a human confirmation at the end of the technology invocation. Thus, either the intelligence of the tool needs to be augmented by human input or the human intelligence can be augmented by technology. Either way, there is no big green button.

Once the counter-party has been identified and confirmed, the topic clustered documents need to be combined properly under each relevant counter-party to create intact contract families. Technology can facilitate this, but what if a family has Amendment 4 (implying there are Amendments 1–3), but only Amendment 4 is in the corpus of documents? Even identifying that a document is missing requires human intelligence — because in fact, Amendment 4 is actually Amendment 1, but just mislabeled.

At this point, the case is likely made that (1) an effective process needs to be put in place before technology gets deployed, and (2) a combination of human acquired intelligence and technology augmentation of that intelligence is necessary, simply to counter-act the inherent messiness of contracting.

Given this reality, what’s the future role of AI in the contracting space?


We never purported to be great sages. However, we can safely say that in our carbon-based years left on this planet, we do not see a world where augmented intelligence in the fashion described above will be trumped by an idealistic notion of artificial intelligence, devoid of human intelligence.

The likely route to success for the foreseeable future in the contract space is therefore one that harnesses the power of both humans and machines. Those that will most successfully apply AI will do so by leveraging human experience and best practice to put in place the right processes before and during the application of technology. They will resist the siren song of the simple and sexy, realizing that technology is never a solution. Rather, it is a tool that can help augment the solution.

[Related: The Lawyers of Tomorrow: 5 Trends Changing the Industry]

Another aspect (perhaps for a future article) that needs to be taken into account is that it will be humans, with all their messy emotions, who need to leverage whatever process and technology improvements are put in place. To succeed, the improvements will require humans to change how they behave and work. Changing habits such as these is as much an emotional process as it is a technical one. Change management therefore also needs to be included in the mix to get humans on the right side of the equation. People need to understand why they are being asked to change as much as they need to understand how the change will take place.

About the Authors

Prashant Dubey is president and CEO of Sumati, the leading global provider of augmented intelligence (AI) based Contract Migration services. Dubey is also the co-author of The Generalist Counsel: How Leading General Counsel Are Shaping Tomorrow’s Companies, published by Oxford University Press. He serves on the University of Chicago Alumni Board with Bjarne Tellmann.

Bjarne P. Tellmann is chief legal officer and general counsel of Pearson and a member of its executive team. Until very recently, he authored the ACC Docket “Career Path” column. His book, Building an Outstanding Legal Team: Battle-Tested Strategies from a General Counsel, was recently published by Globe Law and Business. He serves on the University of Chicago Alumni Board with Prashant Dubey, as well as on the University of Chicago Law School Council.

The information in any resource collected in this virtual library should not be construed as legal advice or legal opinion on specific facts and should not be considered representative of the views of its authors, its sponsors, and/or ACC. These resources are not intended as a definitive statement on the subject addressed. Rather, they are intended to serve as a tool providing practical advice and references for the busy in-house practitioner and other readers.