Artificial intelligence (AI) is everywhere. At least, the label is. Across industries, in B2C and B2B, the AI label is put on services and products without much regard as to whether they’re powered by artificial intelligence. It’s a phenomenon known as “AI washing.”
The procurement industry has not been immune to AI washing. For example, in an assessment of 21 contract life-cycle management (CLM) software providers, many of which touted their AI capabilities, most were still at the very earliest levels of AI maturity according to a report by Forrester, the Cambridge, Massachusetts-based business, marketing and technology research firm.
The Power of Real AI
A contracting solution that is truly powered by real, mature AI has the potential to be a game-changer for procurement professionals. For example, real AI has enabled British multinational oil and gas company BP’s procurement team to:
- Reduce time to contract by 87 percent for software-as-a-service (SaaS) contracts
- Save an estimated 80 percent procurement and legal time
- Gain visibility into current contract risks without manually combing through PDFs
- Control negotiation terms while reducing friction with suppliers.
The solution leverages real AI to create clause options within contract templates that the procurement and legal functions vet. That makes the procurement-legal-supplier dance much more efficient, as well as eliminates redlining. In this case, the suppliers received self-service contracts that were based on the templates and clause options BP pre-approved.
Identifying Real Versus Fake AI
AI washing makes it difficult to discern which products leverage real AI to deliver real benefits. By developing a basic understanding of what separates real AI from imposters, you’ll be able to identify AI washing — and find the right software solution to deliver the kinds of benefits you want to achieve.
So, what is real AI? As artificial intelligence researcher Charles Isbell says, “AI learns, then acts on that learning.”
In the context of a contracting solution, there are six markers of real AI — the technical underpinnings that enable a contracting solution to accelerate time to contract while reducing contracting risk. Procurement leaders can use these markers as a checklist for potential solution vendors, to assess the true AI capabilities of the solution.
The six real-AI markers are:
Automated ingestion of all contract language from historical contracts and templates in any format. With real AI, this is completely automated, and the system reads in unstructured data from any type of file (like Microsoft Word documents, PDFs and even scanned images). This is what enables the AI to learn the types of contract clauses you’ve used most often in the past. It is essential to the software’s ability to make the kind of useful recommendations for template enhancements and supplier clause options that enable you to reduce time to contract.
Automated extraction of any key data. Leveraging natural language processing (NLP) for completely automated key data and contract language extraction enables full visibility into your historical contracts. “ ‘(R)eading’ legal and contractual documents to extract provisions using natural language processing” is one of the key use cases for AI, according to analytics experts Thomas H. Davenport and Rajeev Ronanki. In contrast to real AI, many of the tools claiming to be AI require you to manually tag key data.
Along with the ability to automatically read contract data from all types of files, fully automated extraction of key data is what gives procurement leaders confidence that the software is presenting a holistic view of all potential sources of risk.
Human-like language comprehension and a semantic comparison. Real AI leverages NLP and automatically generated knowledge graphs (ontologies) that learn over time to enable human-like contract comprehension. That’s essential in a world of historical contracts written by human beings — where clauses that mean substantively the same thing could be written differently. AI can compare clauses to identify missing terms and prerequisites, such as negotiation boundaries, policy transgressions, material changes, clause and template similarity, and obligation exposure — even if the words it is comparing are not the same. That way, you can be confident in the accuracy and completeness of the software’s risk-rating recommendations.
AI-powered NLP and knowledge graphs also enable context-based search, so you can find contracts and provisions without searching the specific term. If you wanted to assess, for example, the force-majeure clauses in your contracts, searching ‘force majeure’ without real AI would reveal only those clauses with that exact search term or under a specific force majeure section. With real AI, in contrast, you’d also see the clauses with terms like “pandemic” or “natural disaster,” thus painting a much more complete picture of actual risk.
Analytics gives you insights into provisions suppliers have redlined and contracts you’ve signed. AI-powered analytics on historical contracts reveals such insights as provisions that have been most redlined and those most accepted as is. The system should analyze all historical contracts, templates, clauses and supplier contracting behavior — for example: clauses most changed by suppliers, suppliers making the most changes to contracts and contracts with the highest risk — and present insights accordingly. AI-powered analytics is a reason BP shortened average SaaS contract-processing time from 92 days to 8 days while also reducing friction with suppliers.
Analytics with a purpose recommends ways to improve the template and negotiation process to improve efficiency and reduce risk. Real AI goes beyond pure analytics and leverages “analytics with a purpose” to take insights from signed contracts and recommend ways to shorten negotiation time and reduce risk. It’s another of the CLM solution features that Forrester identified among six that deliver business value: “Linking contracts to results to improve contract language and contracting processes.”
At BP, for example, suppliers were required to have a minimum value in liability insurance. Yet AI-powered analytics revealed that most signed contracts allowed less insurance. In addition, analytics revealed no correlation between allowed insurance amount and contract value — a potential risk if high-value contracts had low insurance amounts. With those insights, BP (1) improved contract efficiency and mitigate risk by including supplier clause options for different insurance amounts and (2) better aligned insurance value to the value and business criticality of the contract.
Machine learning incorporates your data, teaching and suppliers’ behavior into its recommendations. A real-AI solution leverages machine learning in four ways to improve decision-making:
- The major source of learning initially is the analysis of historical contracts. The system learns from previous agreements to make template recommendations and initial risk ratings without heavy manual intervention.
- When you alter a risk rating or the risk points assigned to different clauses and free-form a reason, the system learns from that and will change its assessment of those clauses in the future.
- The system also learns from manually tagged clauses. For example, say you read a force-majeure clause and determine it applies in a pandemic, but the word “pandemic” is not in the clause. You can add “pandemic” as a tag, and the system will understand that the force-majeure clause covers pandemics and will apply that tag to similar force-majeure clauses in the future.
- Suppliers’ clause-option selections are taken into consideration by the AI model to improve the process. Only with supplier feedback can the efficiency of the process be optimized.
As the BP example shows, real AI can drive tremendous results for companies. It’s no wonder some providers are tempted to put the ‘AI’ label on their products. Fortunately, when you know what to look for, it’s not too difficult to differentiate real AI from its imposters.