While CTOs and CIOs are aware of the challenge, they are typically short on time and resources to drive their ideas to implementation; scaling these initiatives from a proof-of-concept (POC) to a value-generating program comes with many roadblocks.
This article is my first in a three-part series on how to bring machine learning to life, including building the right foundations, choosing the right business problem, and identifying the best approach to solve it.
Build your internal foundations first
At its heart, ML is a tool for solving business problems or achieving business outcomes. However, many organizations are too eager to implement the latest products and features that promise to disrupt the market (and which they can sell to clients), without using ML to facilitate their own operations first. It is often the case that optimizing internal processes with ML will yield significantly more value and position your company for sustainable growth. While this may seem counterintuitive, by focusing on operational efficiencies internally, tech leaders can build the credibility and resources necessary to take on the ‘next big thing’ and deliver it to the external market.
As a first fundamental step, technology leaders should pinpoint exactly what ML can accomplish within their business and how it can help achieve their specific goals. This means looking at ML holistically in both the long and short term. For example, in the short term, where is your company today? What use cases have the capacity to be optimized with machine learning with immediate value realization? And, looking at the long-term, what is your desired end-state in ten years’ time? How might ML help you adapt to changing market dynamics to meet evolving customer needs?
Identify the right business problem – and the right approach to solve it
There are two primary approaches you can take to solve problems with machine learning: supervised learning and unsupervised learning. The approach you take will depend on the business challenge you are looking to tackle. In supervised learning, the algorithm is taught by example with annotated (labeled) data sets. While supervised learning requires accurately labeled data, it is incredibly effective at predicting future outcomes based on unlabeled, or un-occurred, inputs. Some common prediction-based issues we solve with this method include demand forecasting (how many shipments are going to be late next month?), churn prediction for employees or customers, and document identification.
Some business challenges require unsupervised learning; here, the data is not labeled, and the ML algorithm needs to find patterns and correlations that are unknown to us humans. This may involve organizing the data in a grouped cluster. Common examples of finding hidden relationships or patterns include filling in the blanks of missing data or text (imputation), image and video recognition, and outlier detection. We also help clients solve language-based problems with this type of ML, for example, natural language processing, conversational AI, and sentiment analysis.
Regardless of the use case, there must be a strong alignment between the C-Suite’s understanding of the value generated by the program, and the implementation team itself – something I will discuss more in my next article.
I would love to hear your perspectives on this topic; get in touch if you’d like to discuss bringing machine learning to life within your own business.