Sergey Nivens/Adobe Stock
In the last two installments of this AI in CLM — Artificial Intelligence in Contract Lifecycle Management — series of considerations for procurement practitioners, I introduced the topic and then dove into the concept of knowledge representation which discussed the importance of building a contract domain knowledge model in the form of a rich repository of contract clauses and related data (e.g., risks) and metadata – not just contract document artifacts.
In fact, more broadly, you can think of concepts such as contracts, clauses, obligations, risks, remedies, milestones, suppliers, etc. as classes of “objects” that represent the knowledge of those physical/logical entities.
These objects are increasingly richly classified, attributed, and interconnected (beyond traditional relational database models used in ERP-type systems), but making sense of them is where reasoning comes in. You can have experts build out specific rules to guide their interaction and eventually even begin to give them goals/objectives to help themselves.
But, you can also teach / “supervise” the software to do things like classify/attribute/relate unstructured or semi-structured contract data into these higher-level knowledge constructs.
From Reasoning to Planning (and Action)
Teaching the machine to learn (i.e., “machine learning”) is also key to enabling predictive analytics in areas such as risk management and prescriptive analytics that help guide you in decision-making. All this knowledge modeling is great, but then what?
Well, expert systems, broadly defined, use domain-specific logic (i.e., rules) to act on these “objects.” Such business-logic in CLM applications (which like most other systems are not typically implemented using object-oriented data) are increasingly allowing business users to define and extend business logic themselves.
For example, an emerging best practice is “guided contracting” which presents end users with automated questionnaires that help flexibly configure a ‘standard’ contract from atomic-level clauses in a clause library. This is functionality that can be built with traditional configurator-like business rules to emulate an expert that guides you to utilize appropriate clauses to meet the commercial/business need.
Such a system isn’t intelligent in the sense of using deep learning like IBM’s Watson or Google DeepMind (these systems use, in part, machine learning — which I’ll address momentarily), but it can help support flexible contract authoring to develop a bottoms-up draft contract built from atomic-level contract clause. This can greatly reduce or eliminate lengthy legal contract reviews by attorneys.
In the future, I expect that expert systems in CLM will have planning and reasoning logic to enable agent-based behavior where contracts become goal-seeking entities that are self-monitoring and can interact with designated humans (and other systems). Such human interaction will require natural language processing (NLP) which I’ll discuss later.
In terms of process automation, some people use the term “robotic process automation” (RPA) to denote such intelligent process automation that also includes many other traditional workflow/integration techniques, but I find this buzzword terminology to be somewhat misleading and confusing, and try not to use it.
As a side note, even if you completely ignore the topic of AI, the concept of embedding contract-related activities and triggers into the workflows of other stakeholder processes and systems is still valuable. Flexible integration and intuitive user interfaces are key here.
Discuss this: