3 Major Concerns for Revenue Forecasting

Revenue Forecasting presents a major challenge for companies, especially in finance and marketing. Erroneous forecasts can be disastrous for a companies bottom line, especially when money is spent in anticipation of higher revenues. While the challenges are numerous, there are some actions a department can take to ensure their forecasts are accurate and continue to be accurate.

1. Bad Data

This won’t come as a surprise to many. Every company has bad data, but does this mean you can’t get an accurate forecast, no. First, the data should be analyzed for the causes of errors. Once the causes are known, the errors can either be removed or accounted for, in a model, greatly enhancing the accuracy.

Not all your data, therefore, is bad. The trick is to identify the data that is good, which will act as a core baseline. This baseline can then be used to identify variations and fluctuations to determine the cause. Factors, such as spikes in class center volume, vacations by salespeople, or temperature all can impact results.

Our data scientists, continually find ways to mitigate the issues of bad data through statistical imputation and simulations. These methods can greatly enhance an understanding of the effect of the bad data, as well as provide methods to remediate the issues.

2. Changing Environment

“Change is the only constant in life”, was a famous quote by Heraclitus. Many revenue forecasting models preach “teach it and leave it”, as though it can continue to provide accurate forecasting without ever modifying it. However, this isn’t really the case. Economic environments change, organizations change strategies, and customers change preferences. In some cases, models can pick up the changes, but more complex models require retraining of the model. While many companies say this can occur automatically, it cannot.

Take for example a scenario in which a variable was never included in the original model. If this variable becomes important at some point in the future, the model could never compute its impact, yielding an incorrect result, such as would occur with a new product category or removal of an entire product category.

We think of this as driving a car, always making adjustments. While generally a smooth ride, there are times in which scenarios appear, where traditional training doesn’t work and we need to make a split-second decision. Algorithms cant generally do this, without massive computing power, such as what is needed for self-automated driving.

Change is the only constant in life – Heraclitus

Revenue forecast models need regular and periodic review for accuracy and training.

3. Bad Interpretation

Ultimately, the models produce a number for the forecast, models that are too general have increased errors overall, while specific models have fewer errors for training sets, and increased errors for data it hasn’t seen. Analysts’ interpretations are critical when this occurs since the end-user of the forecasting model isn’t fully aware of what events are causing fluctuations.

One way forecasters combat this is to create an ensemble model, which aggregates a number of models together. The challenge with this approach is that many there may not be an insight into which part of the ensemble is more weighted, thus the ensemble simply tries to smooth out these fluctuations, causing a loss of, most likely valuable information.

Creating multiple models allows for a more granular interpretation of the forecast as well as greater insight into the error. This examination of error is critical to understanding the events that are impacting the revenue. Simultaneously, it creates a better-disciplined process by which analysts, forecasters, and end-users are diligent in understanding as much of the variation as possible.

Creating forecasting models in a vacuum should be a serious concern for the organization. This black-box approach, without truly understanding the effect of multiple data points and methods can lead to a forecast accurate in some instances, but which can lead to failures down the road. We believe these black-box approaches create more problems, and that a solid repeatable reviewable process is the most reliable approach and yields more opportunities.