In a recent article MIT Sloan Management Review identified three specific implications for companies wanting to create impactful and valuable machine learning applications:
- Differentiated data is key to a successful AI play. You won’t uncover anything new working with the same data your competitors have. Look internally and identify what your organization uniquely knows and understands, and create a distinctive data set using those insights. Machine learning applications do require a large number of data points, but this doesn’t mean the model has to consider a wide range of features. Focus your data efforts where your organization is already differentiated.
- Meaningful data is better than comprehensive data. You may possess rich, detailed data on a topic that simply isn’t very useful. If your company wouldn’t use that information to help inform decision-making on an ad hoc basis, then that data likely won’t be valuable from a machine learning perspective. An expert machine learning architect will ask you tough questions about which fields really matter, and how those fields will likely matter to your application of the insights you get. If these questions are difficult to answer, then you haven’t put in the thought needed to produce practical value.
- What you know should be the starting point. Companies that best use machine learning begin with a unique insight about what matters most to them for making important decisions. This guides them about what data to amass, as well as what technologies to use. An easy place to begin is to scale and grow a piece of knowledge that your team already has and that could create more value for the organization.
In finance, alternative data reaches beyond the traditional Securities and Exchange Commission reports and investor presentations that influence investment decisions. Alternative data, such as social media sentiment or number of patents awarded, is essential for two important reasons.
First, traditional data focuses on traditional assets, and that isn’t expansive enough in the age of intangible assets. Second, there’s no reason to bother using machine learning to study the same data sets that everyone else in the market is analyzing. Everyone who is interested has already tried to correlate industry trends, profit margins, growth rates, earnings before interest and taxes, asset turnover, and return on assets — along with the more than 1,000 other commonly reported variables with shareholder return.
Looking for connections among the same sets of material that everyone else has isn’t going to help companies win. Instead, organizations that want to use AI as a differentiator are going to have to find relationships between new datasets — datasets they may have to create themselves to measure intangible assets.