Reducing Uncertainty with Predictive Modeling

Advanced analytics goes beyond hindsight and insight, providing CFOs with the ability to see the future.
Shaheen DilOctober 26, 2016
Reducing Uncertainty with Predictive Modeling

The business corollary of the popular quote, “The only thing that is constant is change,” by the Greek philosopher Heraclitus, is that the only thing constant in business is uncertainty.

How businesses deal with that uncertainty is everything. It can limit opportunities and hamstring the business, or it can provide a competitive advantage. CFOs who recognize the upside and act on it will have an advantage in their markets.

Drive Business Strategy and Growth

Drive Business Strategy and Growth

Learn how NetSuite Financial Management allows you to quickly and easily model what-if scenarios and generate reports.

Realizing this advantage does not happen in the dark. Not incidentally, the buzzword in the marketplace today is analytics. Many vendors are offering a variety of analytics solutions across a spectrum of techniques, technologies, and applications. These range from traditional ad-hoc reporting to online analytical processing (OLAP) and advanced visualization techniques that lie in the realm of descriptive analytics. Such solutions give you hindsight — telling you what happened so you know what events to avoid or seek moving forward.

In addition, there are predictive software vendors, data scientists, and predictive modelers who can identify cause and effect through a variety of techniques. These scientists and techniques answer the question of why something happened, or what is likely to happen. These modeling techniques provide insight, or the understanding of what might be done differently next time to achieve a different result.

Finally, the field of operations research provides the techniques for prescriptive recommendations that answer the question “How do I optimize the solution to a given problem?” As computing power has become abundant, the discrete solutions of linear, integer, and mixed-integer programs have given way to stochastic optimization techniques that allow the determination of optimal outcomes given uncertain inputs.

Optimizing Results

These methods provide businesses with prescriptive actions likely to result in optimum results. This goes beyond hindsight and insight, providing foresight: the ability, coveted by every CFO, to see the future, to the degree the future can be drawn from what is known today.

Such foresight is critical in many aspects of business, as financial outcomes are often tied to both uncertainty and dependent events. This allows business managers the opportunity to leverage Jensen’s inequality — a mathematical expression proving that the optimum number in a probability distribution is most likely not the mean of the distribution but a different number — for their benefit.

Shaheen Dil

Shaheen Dil

Let’s take the example of forecasting sales and buying inventory in a retail environment. From a business perspective, we have a choice of how much inventory to buy, when to buy it, and where to have it distributed. The challenge is that, unless we get the forecast exactly right (and eliminate all the other variations in the supply chain), we will end up with either too much or too little inventory.

In either case the financial impact to the business will be negative. Too much inventory, and we have unnecessarily increased the need for working capital to cover the additional costs of carrying excess inventory (warehousing, cost of capital, etc.). Depending on the nature of the inventory, it may eventually be marked down, put on clearance, or even written off.

Too little inventory, and we suffer stock-outs. At a minimum, this is a lost sales opportunity. It may drive our customers to our competitors, reducing market share, reducing the frequency or lifetime value of the customer, or even resulting in customer attrition.

Enter a data scientist. To a data scientist, uncertain business outcomes are the same as dependent or target variables, and treated as such. Uncertain values over which the business has no control, but which may affect the outcomes, are independent variables. Finally, variables that the business has direct control over are considered treatments.

From here, several techniques can be applied to help optimize the business and financial outcomes of this type of problem.

First, predictive modeling can be deployed to understand several of the key uncertain dependent variables. The most obvious technique is a time series forecasting of sales. This can be supplemented with additional models to better understand market share, attrition, and the impacts on and costs of working capital.

For example, retail sales often depend on weather, and weather data can be included in the forecasting model as a variable input, resulting in a probabilistic forecast, which generates a probability distribution of possible sales performance.

The advantage of a probabilistic forecast is that it gives the CFO a data-driven analysis that can support the sales and operations planning process. As the sales, marketing, operations, and financial plans are reconciled, the organization gains an awareness of the variability inherent in one of the key numbers that fits into the models, namely sales. In this way the CFO can assist the organization by ensuring that the proper amount of working capital is allocated for the planning period.

Second, a value-of-information model can illustrate how the applied treatments map to core financial variables. In the case of the retailer above, the model can help the CFO understand the impact of buying more or less inventory, based on the uncertainty inherent in the sales forecast.

Finally, data analysts can prepare a stochastic financial model that leverages the previous work, allowing the CFO to examine a set of probabilistic projected financial statements based upon the optimal decision points for the treatments.

Forecasts are inherently inaccurate. They are never 100% correct. Forecasters rely heavily on the control limits of discrete forecasts and the entire distribution for probabilistic forecasts.

Perhaps more importantly, when we use a mean forecast (in other words, one that’s in between higher-confidence and lower-confidence forecasts) to select a treatment (inventory level) to yield a financial outcome (gross margin contribution return on inventory), the result is most often less than optimal. This is due to the nature of the probability distribution combined with the cost/benefit trade-off of too much versus too little inventory.

Results cannot be guaranteed, and each business problem is different. However, in many instances application of these techniques can have significant positive impacts on top-line growth, working capital requirements, and operating expenses.

Wide Application

The example we used focused on retail and supply chain, but the same techniques can be applied across a wide variety of industries and functions. Today, analytics are used heavily in marketing to determine allocation of marketing budgets. The technique can also be applied to operations, research and development, merchandising, supply chain, and human capital.

Due to their insight and understanding of standard financial models, CFOs are perhaps in the best position to understand and benefit from stochastic financial projections. Understanding the range of possible outcomes of a core business decision gives them the ability to make risk/reward decisions with more confidence.

The marketplace today is much more dynamic than in the past, and the price of getting business direction wrong is much greater. With technology advancements and computing power far beyond what was available just a decade ago, there is no reason to forego opportunities to reduce uncertainty and realize the business and financial goals of the enterprise.

While change may be certain and uncertainty constant, advanced tools and techniques can help wrangle this uncertain future and drive your business forward.

Shaheen Dil is a managing director and global leader of Protiviti’s data management and advanced analytics practice.