Scott Godfree

Scott Godfree

Scott is part of the Product Management team and specializes in support and enrichment of Planning related capabilities within SAP Analytics Cloud. Scott holds an MBA in International Management and has an extensive background in Finance and Accounting. He has spent the last 15 years involved in the development and delivery of Financial Planning solutions.

Keep in touch

Subscribe for the latest news, updates, tips and more delivered right to your inbox.

Subscribe to updates



Connect with us

The Predictive Forecast function embedded within SAP Analytics Cloud is a valuable feature which utilizes time series forecasting to either project future values in chart form or to materialize these values back into a stored planning version.

Predictive Forecasting Example

Time series forecasting utilizes historical values to identify seasonality and other trends which can be used to effectively project future values. The dependency on historical values implies that certain source data criteria must be met to produce meaningful projections. In other words, the better the source data meets the standard, the more reasonable the time series forecast will be. To help better understand how to gain the most value from Predictive Forecasts, let’s consider some of these criteria further.

Source data volumes

The most important factor we need to consider when trying to use a time series forecast is the volume of source data available. Here, we are specifically referring to the volume of historical source data based on the time granularity of interest, e.g. year, quarter, month or day.

One of the most common questions we hear about predictive forecasting is “why is there a limitation on the time horizon that can be predicted?”. Any limitation on predictive time periods stems directly from the volume of data available for the predictive algorithm to analyze in each individual scenario.

The predictive forecast must establish a trend, and to do so it requires a certain number of source data points. In general, the minimum viable number of source values is three. Meaning that at least three historical data points are required to project the next period. Specifically, three years of history are required to project the following year (at the yearly level), three quarters to project the next quarter (at the quarterly level), three months to project the subsequent month, and so on.

The more historical data available, the more accurate the prediction will be, and the greater the time horizon you will be able to predict. If you have less than three source periods of historical data available, you will not be able to use the predictive forecasting feature.

Predictive Forecasting settings

Source data quality assumptions

In addition to ensuring sufficient historical data, time series forecasting inherently assumes that past performance is a quality predictor of future performance. So, there is an implicit assumption that historical data was generated under similar conditions to those we expect in the future. Any factor that violates this assumption, whether it be macro-economic or scenario specific, lowers the confidence in any prediction.

Let’s consider this further with a couple of examples:

Scenario 1 (macro-economic): Projected Recession

Let’s assume a consumer good producer is trying to forecast sales for next year. The leading indicators seem to suggest that a recession maybe on the horizon. In such a scenario could time series forecasting be used to project future sales?

While there are certainly cases where products may not be materially affected by a recession (e.g. bread, milk or other necessity goods), for most products this is not true. And, reductions in overall disposable income typically result in restricted spending on consumer goods.

In such a scenario, predictive forecasting would not be a good fit as it only utilizes historical data as an input without the ability to incorporate these additional macroeconomic assumptions (although SAC predictive forecasting does allow us to incorporate additional forecast inputs in the case of time series trend charts).

Scenario 2 (scenario specific): Retail Store Redesign

Here, let’s consider a situation where a retailer has temporarily closed a location (Store A) for refurbishment. The retailer is creating an updated sales forecast for all locations, including Store A. In such a scenario, would time series forecasting be appropriate to predict future sales for Store A assuming we have sufficient sales history?

The answer is probably not. Typically, the goal of a location refurbishment is to attract more customers to the store in order to increase sales. The retailer may also use this opportunity to change the product mix/margin in Store A. If we use a predictive forecast in such a scenario, then we are not considering the potential impact of the refurbishment on overall sales.

Seasonal trends

Beyond source data volume and quality requirements, a more specific comment is warranted regarding the incorporation of seasonality into predictive results.

Time series forecasting in SAP Analytics Cloud incorporates seasonality into projected results if, and when, the source data includes at least three valid seasonal cycles. So, if a retail store has weekly sales trends which are reflected based on daily sales, we would need at least three weeks’ worth of daily sales to project the subsequent week, inclusive of seasonality.

Source data distribution and uniformity

The assumptions discussed above apply to time series forecasts at any hierarchy level in your data. However, additional considerations around underlying data distribution should be evaluated when using Predictive Forecasts on aggregated data.

Consider a situation where a predictive forecast would not be appropriate at a granular level due to some of the data fluctuations discussed above (e.g. a specific project with minimal sales history). In such a scenario can we utilize a predictive forecast at an aggregate level (e.g. product family)? Potentially yes, if (and it’s a big if) the consolidated data adheres to the criteria previously discussed.

In a case where aggregate projects can work, It is important to recognize that any such projection is only valid for the aggregated dataset itself, and the distribution and underlying analysis at a more granular level may not be useful.

As part of the predictive forecast feature, we are given the option to distribute target values based on the existing weightings in the target cells or based on the weightings in the source reference period (which is the most recent period of source data).

In either case, the aggregated projection is distributed to its children based on the weightings of a single source period, and this is true regardless of the number of periods being projected. Consequently, per our earlier example, the distribution to any individual product and time period, has little, if anything, to do with the trend for that individual product. As such, while the aggregate projection maybe reasonable, the detail will not be.


Predictive forecasting is a valuable tool in BI simulations to identify upward or downward trends in future predictions, in helping to establish plan values or to create additional benchmarks against which to compare planned values.

Whatever the use-case, it’s important to ensure that source data is qualified before considering the potential value of the resulting time series forecast. In addition to meeting source data volume criteria, there are implicit assumptions around maturity, stability, and uniformity of the source data. Meeting all of these criteria is important in order to get the most of out of your predictive forecast.


SAP Analytics Cloud earns a top ranking from BARC

See how SAP Analytics Cloud performed in the world’s largest survey of Business Intelligence software users.