Nilly Essaides, Director & Practice Lead, Financial Planning & Analysis, AFP
Driver-based models express the mathematical relationship between key business drivers and financial performance, e.g., how number of website hits translates into revenue. They are the way more than 80% of companies are (to some degree), looking to advance the accuracy of their forecasting process by tying operational KPIs to enterprise performance.
But how can companies create the right algorithms to capture that relationships. (See more in an upcoming AFP FP&A Guide, in our FP&A Guide series, What’s Driver-Based Modeling & How Does It Work?)
First companies need to identify the right drivers. Next, according to Philip Peck, VP Finance Transformation at the Peloton Group, they can use several techniques to develop the models, which apply to:
1. The determination of the candidate operational drivers
2. Testing the mathematical algorithms and drivers for the effectiveness and accuracy of their predictive value
3. Refining those drivers and related model relationships over time.
Slicing and Dicing
One way to start is by decomposing the end-to-end value chain of the underlying business. Essentially that’s looking at the business and understanding all of the various activities and functions that contribute to making, packaging, selling, delivering, and supporting the goods and services generated by the organization.
“The analysis helps one develop a horizontally integrated view of the organization and highlights the key integration points between various aspects of the business,” said Peck.
The most common approaches include “driver trees” or “Predictive logic diagrams.” Those help determine the driver/model relationships and candidate drivers. This involves looking at the outcome measure in question (dependent variable) and determine all of the contributing drivers (independent variables) that impact the outcome measure. Depending on the domain area in question, the number of levels or layers can be as simple as one or could continue into multiple levels. “Driver models can be quite simple, e.g., A X B = C. Alternatively, the driver models could be complex and incorporate a multitude of independent variables and potentially several levels of driver interdependencies,” Peck explained.
Once the initial driver model relationships and candidate drivers are determined, FP&A professionals can employ statistical techniques using representative data to identify the most applicable drivers and refine their math.
The goals is to find drivers that are most highly correlated with the outcome measure and demonstrate clear causality, vs. mere correlation. That’s how FP&A can ensure the model has good predictive capabilities.
But it’s not all about math, according to Tony Levy, Business Unit Executive, Business Analytics at IBM Software Group. FP&A also needs to work closely with the business to make sure it picked the right drivers, before it develops the complex or more simple models. He encourages finance to get closer to business leadership in addition to do the statistical analysis of the data. “Use both to refine and adapt your models.”