Machine Learning: AI Can’t Walk Before It Crawls
It takes months for a newborn baby to develop the muscles to crawl—even longer before they eventually learn to walk. Similarly, it can take years to train a person to draft, negotiate, and understand a written contract, and we shouldn’t expect machines to do it from the outset.
After using Apple and iOS for the last few years, I recently switched to Samsung and Android. It’s not that I don’t like Apple, but I wanted a change, and I’d like the look of Samsung’s virtual assistant Bixby—which is similar to Siri and Google Assistant.
Based on machine learning, Bixby takes a while to recognize your voice, but when it does, it can put on your alarm and check your emails and messages. If it doesn’t understand you, it says, “teach me,” which is where you need to be patient and give it a written command. But like all machine learning tools, the more time you spend with Bixby, the more your phone gets smarter, and the more accurate are the outcomes of your requests.
Reap what you sow
When it comes to contract management, most people expect machine learning to solve all their problems overnight. But given the complexities and length of contracts, the variety of templates and language used, and the similarity in the documents, the output is bound to vary at the beginning. Only by having patience in the machine learning capabilities of your robots will you obtain the desired results.
Although this might sound familiar, using machine learning tools requires a dedicated investment of time and money. To give an example, one of my clients was looking for specific information from a repository of contracts. After delivering a proposal and configuring the tool, we understood that we only had a few hundred contracts from which to extract data, which wasn’t enough for machine learning to be successful.
Machine learning only happens when thousands of documents with a huge amount of data are fed into the system, and any machine learning tool is bound to fail if you don’t give it enough data to process. In this case, we had to extract the data manually using traditional optical character recognition (OCR) and other methods. But given sufficient data, we could have used it to teach the machine, which would have enabled us to deliver the desired results down the line.
If you’re looking for some advice…
Before you get frustrated and feel your machine learning tools are unable to drive the results you’re looking for, here are a few points to consider when implementing artificial intelligence (AI) and machine learning solutions across your contract management function:
- Identify the outcomes you require from your machine learning tool. This is the basis on which the tool will be set up.
- Set up your machine learning tool based on the data available to you. Access to sufficient data or a number of documents is the key for a successful outcome from any machine learning/AI tool.
- AI tools are equipped to separate your contract documents if they are mixed with other documents in order to maximize the amount of data that can be extracted.
- Standardize the templates or file types you use to ensure machine learning happens faster and with higher accuracy.
- Don’t invest in machine learning tools unless you have thousands of contracts to feed into the tool. Traditional tools, such as OCR, are sufficient for a smaller set of documents.
With AI quickly becoming the norm and the introduction of chatbots into contract lifecycle management, having access to sufficient data can drive excellent AI-based results. Machine learning is a set of algorithms that your implementation partner will customize to your needs, but don’t expect your machines to walk before they can crawl!