In this special guest feature, Kevin Gidney, Co-Founder and CTO at Seal Software, explores four main factors that go into creating advanced machine learning technology. There is a lot of required training and work that goes into developing successful machine learning solutions and not all ML is created equal. Kevin, a founder of Seal Software, has held various senior technical positions within Legato, EMC, Kazeon, Iptor and Open Text. His roles have included management, solutions architecture and technical pre-sales, with a background in electronics and computer engineering, applied to both software and hardware solutions.
It’s no secret that machine learning is dominating the enterprise, across a wide variety of industries. From retail, financial services and legal spaces, the technology provides necessary algorithms for companies to produce predictive accuracy and more efficient results. As a powerful artificial intelligence tool, machine learning is helping these businesses solve previously unsolvable problems. However, what many people don’t realize is that not all machine learning is created equal; having machine learning technologies doesn’t automatically mean it’s as intelligent as you may think.
In particular, this is the case when it comes to the Contract Discovery and Analytics market. While machine learning is an essential tool to extract intelligence from contracts, it is not a magic technology. Years of training and applying different techniques are required to turn the technology into a successful solution. Instead of the common fear that “the robots are coming,” it takes human interaction to develop a scalable, usable, and accurate system. The robots are actually joining us.
As more companies enter the contract discovery and analytics space, it confirms that contracts hold critical information to help organizations better manage their M&A activity, regulatory compliance initiatives, procurement and sales functions, and make better decisions. What many companies don’t take into consideration is that it takes years for any vendor to build an effective data extraction and normalization engine for contracts.
In order to implement the most advanced and successful machine learning technology, enterprises need to focus on several main factors: trust-ability, accuracy and precision, model and method, and usability and scalability. Here’s why:
- Trust-ability:Though many companies often disregard the “trust-ability” of any model or method to extract information, it is an incredibly important factor when training a machine learning system. When trusting a model, there needs to be an appropriate amount of data in the teaching process to smooth out any potential errors or mistakes, before executing the technology. Machine Learning (ML) algorithms evolve over time, by allowing it to automatically learn from the data you provide. Similar to a learning curve, it needs time and ample data to properly assess various situations and produce the best results. Adding more data before deploying the technology will only make the system better and, consequently, increase your trust-ability of the results.
- Accuracy and Precision:Accuracy depends on the project and requirements, and various levels of accuracy are necessary to achieve a particular information objective. These objectives can range from migrating contracts to a business system, running broad scale analytics, kicking off a regulatory compliance initiative and performing due diligence, to name a few. In order to achieve the highest degree of accuracy, the system depends on the precision and recall scores. In plain terms, precision is how useful the results are and recall is how complete the results are. Those scores are dependent on the trust-ability, as stated above. The system needs time to learn before deploying, which will produce the right level of precision and recall, thus resulting in the highest degree of accuracy.
- Model and Method:Another important note is that different events need different technology. For example, when working with an M&A, you may choose to use techniques that will yield different levels of accuracy. Other situations may require a slower method, with more generalization. Lastly, there could be an instance when both techniques would be required. By using the right model and method for each task, you will increase precision in a critical time.
- Usability and Scalability: If the system is only appropriate for trade-specific professionals or data scientists, it is harder to extract the business value. The data extraction system should be usable by the people who need the data. The challenge is creating a system sophisticated enough to achieve the project’s objectives, but also has features and a UI designed for business users. These users need the power and flexibility to provide examples (effectively teaching the system) and interact with the interface to achieve the desired outcome. It also is key that the system is scalable. Large organizations can have millions of documents, and by combining the number of contracts with specific precision and recall requirements for accuracy (as discussed above), there is a higher degree of scalability possible out of a ML system.
Before integrating a ML-based system into your organization, it’s critical to consider a system that is powerful, yet usable. Moreover, it’s important to keep in mind that a ML-based system is not a magic robot — it’s an advanced technology that needs to be trained and tuned, with goals for precision and recall to meet both an accurate, specific objective and a positive business outcome.
Sign up for the free insideBIGDATA newsletter.
Speak Your Mind