Forecasting with Predictive Analytics

• Expense and cost planning:
  • What will be my expenses by location and category?
  • Should we transfer budget between areas?
• Revenue and sales planning:
  • What is the revenue forecast per business unit?
  • How will sales evolve for specific products in the future?
• Headcount planning:
  • How many employees will leave due to attrition or retirement next year?
  • What are the profiles of employees that leave?

To help you make better business decisions, SAP have integrated automated predictive features into SAP Analytics Cloud.

Smart Predict allows you to:

• Make trusted and actionable predictions. Instead of asking how you obtain predictions, you can focus on what you want to predict.
• Transform the way that you analyze data and make plans with machine learning-driven predictions, add results to a story, and share them with the business.

Often, business users and business analysts have no skills in data science. Therefore, Smart Predict makes machine learning algorithms accessible to everyone, and predicts potential outcomes and forecasts with the push of a button. Smart Predicts offers:

• Simple visual experience
• Typical guided workflow for classification and regression models
• No technical terminology
• No coding
• Business oriented questions

A predictive scenario is a preconfigured workspace. You can use it to create predictive models and reports to address a business question that requires the prediction of future events or trends. You choose the one that is relevant to the type of predictive insights that you are looking to create.

Create sorted lists based on expected probability (classification models), estimates about future or unknown values (regression models), or forecasts (time series forecasting models). These predictive models allow you to generate future, hidden, or unknown data.

Smart Predict can be used on various acquired sources (and on HANA Live source), to help you to answer precise questions. There are three main predictive techniques that cover business use cases and are covered in this course:

1. Classification models are used to anticipate the behavior of a customer, their propensity to buy, and the risk of churn. "Give me the sorted list of prospects to focus on for this product."
2. Regression models are used to predict numeric values and identify their influencers. "Give me the estimates of revenue for each customer in the next six months."
3. Time series forecasting models are used to generate time series forecasts with individual forecasts generated for each value of a dimension. "Forecast the revenue for each product and each point of sales daily for the next 30 days."

While there are many smart features in SAP Analytics Cloud, Smart Predict is the focus for this course. We focus on the Smart Predict functionality in building, training, and applying predictive scenarios.

Configuring the predictive scenario is straightforward with Smart Predict. You don't need to be a machine learning expert or a data scientist. The basic workflow is as follows:

1. Select the planning model to be augmented.
2. Select the version of the planning model to learn from. Smart Predict should learn from actual data.
3. Select the measure to be predicted. In this scenario, you want to predict the gross revenue measure.
4. Select the number of forecast periods to be predicted. The granularity of the planning is month, and if you would like to get predictions for 12 months, request 12 forecasts.
5. Select the entities. For example, you can share the predictions and insights with stakeholders responsible for different products in different countries/regions.

You can further refine your predictive forecasts by using the Influencers setting that allows you to select which influencer variables to include before training your predictive planning model.

With Smart Predict, you can go one step further and generate forecasts per entity to get accurate business-oriented insights, not only raw forecasts. You can directly use your planning model as the data source for the predictive scenario, no need to extract data in a dataset.

Smart Predict uses the data available in your planning model to create and train a predictive scenario. You can then analyze predictive forecast accuracy across the combined dimension values and understand signal breakdown in details. Once you're satisfied with the accuracy of your predictive scenario, you can generate the predictive forecasts: they save back directly in the private version of your planning model. It’s then easy for you to augment your story with actual and predictive forecasts.

When you create a predictive scenario, you initially specify a training data source, a target or signal variable, and then define additional training settings. Training is a process that uses SAP automated machine learning algorithms to find relationships or patterns of behavior in the data source. You can apply the result to a new data source to predict with a probability what could be the value of the target or signal for each element of the data source.

When using a planning model, the input version must be a public version, not in edit mode, or a private version. You have at least read access to it.

A predictive model produces performance indicators and reports as a result of a successful training.

Here's a short summary of the different components that you can use to debrief your results so you can verify the accuracy of your predictive model:

• Horizon-Wide MAPE Is the main performance indicator high enough to consider my predictive model robust and accurate? Check the quality of your model performance over the Horizon-Wide MAPE. It evaluates the "error" made when using the predictive model to estimate future values of the signal, where zero indicates a perfect model. The lower the Horizon-Wide MAPE, the better your predictive model performance. For more information, refer to Horizon-Wide MAPE
• Signal Analysis What forecasts are provided by the predictive model? Have a close look at the signal and forecasts. Signals show trends, cycles, and fluctuations in the signal, each with a description. Check if there are outliers in the forecasts and detect anomalies on the signal. For more information, refer to The Predictive Forecasts, The Signal Outliers and The Signal Anomalies
• Signal vs. Forecast How accurate is my predictive model? Use the Signal vs. Forecast graph to visualize the predicted values (forecast) and actual values (signal) for the data source. You can then quickly see how accurate your predictive model is, what are the outliers, the zone of possible errors. For more information, refer to The Forecast vs. Actual Graph and The Signal Outliers

If you're not satisfied with the predictive scenario once you’ve analyzed the results, you can improve it by changing the settings, or if necessary, changing the data source.

Check the quality of your predictive model performance over the Horizon-Wide MAPE. The Horizon-Wide MAPE is the evaluation of the "error" made when using the predictive model to estimate the future values of the signal. A Horizon-Wide MAPE of zero indicates a perfect predictive model. The lower the Horizon-Wide MAPE, the better your predictive model performance.

Analyze the predicted values for the predictive model over a set of known data from the training data source. Check if there are outliers in the forecasts and detect anomalies on the signal.

Use the Signal versus Forecast graph to visualize the predicted values (predictive forecast) and actual values (signal) for the training data source. You can then quickly see how accurate your predictive model is, what are the outliers, the zone of possible errors.

For more information regarding Smart Predict, you can go to Analyzing the Results of Your Time Series Predictive Model.

After running Smart Predict and confirming that the results are reasonable, you can save the results to the planning model in a private version.

You can write Smart Predict results to private planning versions only. The goal is to enforce a validations workflow and avoid unvalidated forecasts to be written to public versions.

The confidence interval (dotted lines) helps to understand the level of confidence we have in the forecast. If the interval is narrow, we have a high confidence in the forecasts.

After running Smart Predict and confirming that the results are reasonable, you can save the data to the planning model in a private version.

After analyzing the results in a story, the values can then be published to a public version.

For more information regarding on predictive modeling, go to the Applying AI-powered Visualizations and Augmented Analytics to Business Data in SAP Analytics Cloud course, where we explore augmented analytics in more detail, including time series predictive models.