With the preponderance of customer data available to firms these days, it’s no secret that companies want to leverage technology products to mine their “big data” or find hidden strategic gems in their “dark data.” It should come as no surprise, then, that other firms purport to offer data solutions in the form of “machine learning” or “artificial intelligence” products. So, how does one determine a bona fide machine learning or artificial intelligence offering from a marketing operation posing as a product?
The truth is that commercial applications for machine learning and artificial intelligence tools are still in their infancy for all but a few firms. It can be difficult for the newcomer to distinguish between jargon such as “deep learning,” “machine learning,” and “artificial intelligence.” We at Relativity6 pride ourselves on applying machine learning and artificial intelligence tools to simple, real world industrial and commercial problems, such as customer retention and customer reacquisition. One does not need to be a mathematician to determine if a machine learning product is adding value; the proof should be in the pudding. In an effort to differentiate ourselves from other companies operating in our sphere, we thought we’d offer three questions to those firms exploring a machine learning product.
1) Is the company you’re speaking with answering a specific question?
Machine learning is an incredibly accurate and effective technology when answering specific questions. However, machine learning can be an answer in search of a problem. When exploring a machine learning product, look for those companies that focus on either an industry, or on a use case. Having a clear use case, or question to answer, can yield incredible results. On the other hand, machine learning algorithms have trouble answering open-ended questions, so be cautious when a company claims to be able to solve “any problem.”
2) Is the company too focused on their “proprietary” algorithms?
There is value in creating machine learning algorithms from the ground up. However, commercial applications need not be reinvented with every customer. Finding a company that has deployed its algorithms already, and has demonstrated the value of those algorithms over time, is a good way to determine if the company’s products will do what they should. Machine learning algorithms inherently get stronger over time, so working with a vendor that has validated their algorithms with actual customer data is much more valuable than a vendor that has sophisticated proprietary algorithms that are being deployed for the first time.
3) Is the company willing to conduct a proof of concept?
Since there are so many possible solutions and so many claims of high accuracy rates in predictions, a proof of concept with clear objectives is a great way to determine if a machine learning solution can actually solve your problem and provide your firm with the valuable insights you’re looking for. Typically within a predetermined time frame (we at Relativity6 shoot for 4 to 8 weeks), a vendor should be able to directionally prove to you that they have built a model, or redesigned an existing model, that outputs actionable predictions. And, be warned of vendors that tout extremely high accuracy rates in the proof of concept phase. Algorithms need time to predict, validate, and re-train themselves. This isn’t magic, it’s math.
Warren Buffett says “time is the friend of a wonderful business, the enemy of the mediocre.” That quote couldn’t be more apt in the machine learning space. We at Relativity6 love the long term prospects of our products and are focused on the potential of machine learning to continuously provide firms with new insights into their customers — not just a quarterly sales bump.