Contract AI is No Slam Dunk

Contract AI Slam Dunk Basket Ball

In high school, I played offensive line in football. I also played basketball. When you combine the two, it shouldn’t take you long to realize that I couldn’t jump–but I really, really wanted to. One day, I was watching TV when a commercial came on advertising “plyometric shoes.” They promised a better vertical immediately.

I was sold and somehow finagled a pair. I put them on expecting to leap small benches in a single bound, but nothing happened. Nothing. I wore them every day for weeks. Still nothing. Then, a cousin let me in on the catch. That to get the benefits, I needed to commit to a whole set of special exercises. I was crushed. My dream of magically becoming the next poster boy for Air Jordans went out the window.  Instead, I was left with an (at the time) expensive investment and a long road ahead if I wanted to see any reward.  

In recent conversations with people in the contract management industry, I’ve had flashbacks to my brief (and imagined) slam dunk dream. Artificial intelligence (AI) continues to generate buzz among the contract management community as an answer to lagging processes and obscured visibility into key documents and mounds of data. Unfortunately, a large percentage of the people I talk to underestimate or fully discount the work they need to put in to realize the returns they are looking for. 
 

Don’t underestimate the difficulty (and importance) of clean data

Just like I underestimated how hard it was to jump higher, most people underestimate how hard it is to extract accurate, actionable and useful data from unstructured contracts. They also overestimate the “cleanliness” of their current data. At Pramata, we’ve seen customers with error rates of up to 49% across a swath of customer contracts on something as simple as ‘document type’ after deploying a contract AI tool.  Imagine what it was before!

For all its promises of contract management transformation, AI is only as good as the data you’re using to tune it. And what we’re finding is that most companies don’t have good data to start. Plus, most contract AI tools leverage a limited set of contracts to develop their built-in algorithms, making it even harder to create meaningful intelligence across a diverse set of documents.

Just like me and my magic shoes, most people are looking for a “silver bullet” that just doesn’t exist. Because of this, there is a significant, often understated effort that must be put into 1) building a representative data set, 2) tuning the system against that data set, and 3) closing the many practical gaps that technology alone cannot deliver.

To effectively apply AI to its broadest business value, it takes work.  That work comes in the form of eyeballs—skilled ones—providing accurate, personalized data normalization and superior data quality. This is because extracting and normalizing accurate, actionable, and relevant data from unstructured contracts is hard, for many reasons including:

  • Highly negotiated and amended language that materially affects downstream document families
  • Multiple document forms and naming conventions by department or acquired business units 
  • Multiple ways to interpret the same thing (subjectivity)
  • Multiple clauses combining to produce one data point
  • Lack of data entry standardization 
  • Missing documents
  • Poor OCR quality

Without those people and an integrated process, most AI tools essentially become glorified text search or data scraping tools with limited insight into important commercial relationships.
 

The rewards are worth the investment

My point isn’t to dismiss the impact of AI – I’m not a total curmudgeon.  It’s also not to diminish the value of the underlying commercial data sought by companies.  I’ve been working with large enterprises to solve this problem for my entire career.  I’ve seen the cross-functional, organizational impact that having a handle on this data can have on customer retention, revenue, and risk management.    

I firmly believe that unlocking core commercial data should be a top priority for companies; and advances in AI have made that more achievable than ever.  As pressure on margins and operating efficiency increases, it’s a great source of untapped value.  That being said, it’s critical for everyone to be realistic about what it takes to realize this significant opportunity.

At Pramata, we consider ourselves practical idealists.  Because of that, we continue to take an integrated, human-assisted AI approach to unlocking critical data about a company’s commercial relationships. We're straightforward and practical about the underlying challenges of relationship complexity and data inconsistency, and we help our customers with the slam dunk they are looking for.

 

{ }