Organizations do not live in a vacuum and things happening outside their walls have a direct impact on how they perform. So, it is essential for them to incorporate external data in their forecasting, planning and budgeting, especially for predictive analytics and machine learning (ML) to support artificial intelligence (AI). I use the term external data to include any information about the world outside an organization (including economic and market statistics), competitors (such as pricing and locations), and customers. Until recently, it was adequate for organizations to regard external data is a “nice to have” item, but that is no longer the case. External data is necessary for many functions, including useful and accurate competitive intelligence used by sales and marketing groups. It is also essential for the effective applications of AI using ML for business-focused planning and budgeting and predictive analytics.
Many organizations already collect basic external data such as exchange rates, commodity prices, economic data and competitors’ prices, but few go deeper or gather external data in a way that makes it accessible across an enterprise. Ventana Research asserts that by 2025, 40% of organizations will incorporate comprehensive external measures to enable ML to support AI and predictive analytics, as well as in reviews and benchmarking to improve performance. Those that do will have a clearer picture of how and where to address operational shortcomings and will be able to utilize AI more effectively to support more accurate and valuable predictive and prescriptive business planning.
External data enhances the predictive capabilities of models by dealing with the problem of endogeneity — specifically, the issue of omitted variables — in forecasting and analysis. To illustrate, consider a vendor that is selling ice cream on the beach. On most days, the price does not change, but on cooler than normal days, when there are fewer people on the beach, they often drop the price so that they are able to sell as much inventory as possible. Conversely, on very warm days when the beach is crowded, they raise the price and they are able to sell even more than average because people are willing to pay more. At the end of the season, the vendor brings in a consultant to advise on pricing for the coming year. If that numbers-driven consultant only considers the daily sales totals and the price received each day, the “obvious” conclusion is that higher prices lead to more unit sales so just raise prices. But that illogical conclusion is only possible because the analysis is missing a consequential variable about the external environment: the daily temperature.
External data is also essential for creating robust ML systems that support AI. There are many potential uses of this technology for finance and accounting departments, as I have noted, including enhancing the accuracy and agility of forecasting and planning by automating time-series analysis to rapidly develop predictive models for more accurate project revenue and costs, as well as balance sheets and cash flow. It can enhance the breadth of analytics available to improve situational awareness and decision-making. For these purposes, external data is almost always necessary to create performant models because such data can have high explanatory value in determining demand, supply, prices and costs and, in so doing, enables systems to avoid undermining their credibility. Artificial intelligence and predictive analytics are similar. Predictive analytics use algorithms and advanced statistical methods applied to datasets to make more nuanced, and potentially less biased and more accurate, forecasts than those built around simple rules of thumb or intuition. Robust datasets that hold a large and diverse set of data from which inferences are gleaned will create more useful and accurate forecasts. Predictive analytics can include ML to analyze data quickly. Yet, our Office of Finance Benchmark Research finds that only 24% of organizations use predictive analytics. Simplifying access to external data that supports more robust model creation would make it easier to use predictive analytics.
A robust dataset is also valuable because predictions are almost always inaccurate. When a prediction turns out to be wrong enough, predictive models built around multiple drivers make it possible to identify which part of that model turned out to be wrong — either because an element of the model turned out differently or because the model itself was flawed. This provides useful information about what to do next time to achieve a better outcome and how to refine the model to improve its accuracy. Robust in this case almost always means that the dataset incorporates data about the external world. A robust dataset improves the quality of forecasting and enables better, deeper and more insightful analysis. And robust data is essential for making useful recommendations, particularly in qualifying the degree of certainty that might be attached to a specific recommendation. And while there are a lot of statistical techniques applied here, there will always be the need for experience and intuition in making decisions. External data used in forecasting, analysis and planning is useful in tactical, short-term (that is, up to one year) operational planning by a function or business unit. For example, in demand planning, marketing, manufacturing and supply chain. Especially in tactical planning that explores multiple contingencies, building predictive models using external data enriches the process by identifying a wider range of explanatory variables. External data also improves strategic and long-range planning — planning that typically extends the outlook past the fiscal-year budget — to strategize around market trends, identify opportunities and threats, and develop project plans. This involves identifying the most important macro drivers of demand and supply. A typical application of these forecasts is identifying investment opportunities in the form of projects, capacity expansion and business acquisitions and applying rigorous methods of valuation.
Accessibility is a fundamental issue holding down the use of external data for forecasting, planning and analysis. I recommend that vendors of planning applications help by including data as part of their service, especially those that offer or are planning to offer AI features in their applications. Some basic statistical data in the public domain could be incorporated in the subscription price while more proprietary sets that would cost the vendor money could be billed on an as-used basis. Without external data, predictive analytics and forecasts built solely around an organization’s internal information are fundamentally flawed because consequential information that drives results is omitted. Data services would be a win-win for vendors and users by improving the statistical quality of results and an organization’s ability to enhance performance through more intelligent planning.