Describing Predictive Analytics Model Training - Sales (2YJ)

Predictive analytics aims to forecast future events by using data-driven methods. This approach leads to more accurate decisions compared to those based on intuition alone. The field combines various statistical techniques, including modeling, machine learning, and data mining. These methods analyze current and historical data to make predictions about future or unknown events.

Predictive models typically receive input in the form of a dataset, that consists of transactional data or a set of variables. The output, known as the dependent or target variable, is what the model aims to predict. The inputs, called independent or explanatory variables, are used to make these predictions. Before developing a predictive model, it is often necessary to explore and prepare data, considering both the target and explanatory variables.

SAP S/4HANA Cloud provides the following predictive apps for Sales:

• Quotation Conversion Rates - Valid/Not Completed: This app helps you see how well your sales quotes are turning into actual orders. It shows you the percentage of quote values that become sales. If you're a sales rep or manager, you can use this to check how many of your quotes are becoming orders before they expire. The app also uses machine learning to predict future quote conversions, letting you compare what's really happening with what might happen next.

  You can find the app reference documentation in the SAP Fiori Apps Library at Quotation Conversion Rates - Valid / Not Completed.

• Sales Performance – Predictions: This app lets you compare your actual sales results, like total sales volume, with predictions made by the system. You can choose a sales plan you've already created to see how close you are to reaching your sales goals. The app uses smart prediction tools to give you insights about how well you're performing right now and what to expect in the near future.

  You can find the app reference documentation in the SAP Fiori Apps Library at Sales Performance – Predictions.

• Predicted Delivery Delay: This app helps sales reps spot possible delays in open sales orders before they happen. It predicts when deliveries might be late getting to customers. By showing you these potential problems early, you can take action to prevent delays. For example, you can quickly see issues and start fixing them right away.

  The app looks at two main things:

  1. When the delivery might be created later than planned.
  2. How long it might take to process the delivery.

  This way, you can stay ahead of problems and keep your customers happy.

  You can find the app reference documentation in the SAP Fiori Apps Library at Predicted Delivery Delay.

Every predictive app is based on a predictive model. Before using a predictive model in SAP S/4HANA Cloud, you need to train it properly. Here's how:

1. Gather Historical Data: Make sure SAP S/4HANA Cloud contains a good amount of past transactional data. Based on the model, for example Sales Quotations, Orders, Deliveries.
2. Train the Model:
   • Open the Intelligent Scenario Management app
   • Find your specific scenario
   • Start the training process
3. Check the Results:
   • If training is successful, you'll see a new model version with "READY" status.
   • The system calculates quality indicators during training.
4. Evaluate the Model:
   • Look at the quality indicators
   • Decide if the model is good enough to use
5. Retrain if Needed:
   • If the quality isn't satisfactory, train the model again
6. Activate the Model:
   • Choose the best version of your model.
   • Activate it.
7. Use the Model:
   • Once active, you can use the model in your Predictive Application.

Remember:

• Each predictive app has its own scenario and model.
• You must train and activate a model before using it in an app.
• You can have multiple versions of a model, but only activate the best one.

By following these steps, you'll be able to prepare and use predictive models in SAP S/4HANA Cloud effectively.

The performance of a predictive model can vary based on the data in the input dataset. For example, it depends on data quality, quantity, relevance, distribution, feature engineering, noise levels, temporal changes, and preprocessing methods.

SAP S/4HANA Cloud trains predictive model based on the current system data base (sales quotations, orders, deliveries, billing documents). During training, the system measures some quality indicators to establish the performance of future predictions.

Let's explore the proposed quality indicators:

1. Predictive Power: Predictive Power shows how well a predictive model works. It tells us how good the model is at explaining what we're trying to predict when it's used on the data we trained it on.
   • Ranges from 0% to 100%:
     • Predictive Power ≈ 0%: Pure random model
     • Predictive Power ≈ 100% : Perfect ideal model
   • The indicator is specific to SAP, but it is strictly related to standard model quality indicators of general use.
2. Prediction Confidence: Shows how robust a model is. It measures how well the model can maintain its accuracy and performance when used on new data.
   • Ranges from 0% to 100%
   • Recommended threshold: Prediction Confidence ≥ 95% before using the model for predictions.
3. AUC (Area Under Curve): shows how well a classification model can tell the difference between different classes. Higher AUC values mean the model is more accurate.

   Ranges from 0% to 100%

   Note

   Classification Model: A classification model is a type of machine learning model that is designed to categorize or label data points into different groups or classes.

   Imagine you have a bunch of emails and you want to sort them into "spam" and "not spam." A classification model can learn from examples of spam and non-spam emails and then predict whether new emails are spam or not.

   In simple terms:

   • A classification model learns from existing data (training data).
   • It identifies patterns and uses them to categorize new data into predefined classes (like "yes" or "no", "spam" or "not spam").

   So, a classification model helps you automatically sort or classify items based on the patterns it has learned.

You enable this predictive functionality by using an active version of the default model, which is in the predefined predictive scenario for quotation conversion rate (SLSQTANPREDICTION).

You create versions by training the model with different sets of data using the Predictive Models app. After it is trained, you apply the version that you want by setting it as the active version.

You can list quotations with details such as customer, net value, actual and predicted conversion rates, percentage of elapsed time over validity period, and remaining days before expiry.

You can display the net values in your preferred currency. For the currency conversion, exchange rate type M (average rate) is entered by default. You can maintain exchange rates from your transactional currencies to your display currency using the Maintain Exchange Rates app.

You can compare the actual and predicted conversion rates dynamically while drilling down by different dimensions.

To see the predicted conversion rate, you must show the mini tile (hidden by default) that displays the predicted rate. Choose Show Mini Charts.

You can compare the converted value with the prediction by different dimensions.

For further information, you can access the reference App documentation at SAP Fiori Apps Reference Library: Quotation Conversion Rates.

This app allows you to compare your current sales performance with model-based predictions and sales targets. By analyzing these comparisons, you can gain valuable insights into your sales performance and make data-driven decisions.

1. Compare actual and predicted sales volume
2. Analyze achievement of sales targets using existing sales plans
3. Visualize comparisons across various dimensions (e.g., sales organization, customer, material)
4. Compare current year and previous year actual values month-by-month
5. Save and manage filter settings as variants

• Currently, the app only supports sales volume as a performance KPI.
• Sales volume is defined as the total invoiced sales during a specified period, based on non-canceled invoice items or debit memo items.

To display predicted values, enable prediction in the Intelligent Scenario Management app:

1. Search for the predefined scenario for sales volume (SALESVOLUME001).
2. Create versions by training the default model with different data sets.
3. Activate a model version with satisfactory prediction quality.

1. Use relevant and current business data.
2. Ensure sufficient data volume (recommended: at least 2000 records per training)
3. Train and activate the model at least once a month
4. Conduct training during non-working hours

For further information, you can access the SAP Fiori Apps Reference Library at Sales Performance – Predictions.