The generation of automated cross-selling suggestions is useful if your company manages a large product catalog, where manual setup of these suggestions would take significant time and effort.

You specify the settings for the generation of cross-selling suggestions on the Machine Learning tab of the Sales Orders Preferences (SO101000) form. In the Cross-Sell Assistant Settings section, do the following:

• In the Data Period for Analysis box, specify the analysis period for the AR transactions used in generation. Use a shorter period if your set of items changes frequently. Select one of these options:
• 3 Months
• 6 Months
• 1 Year (default)
• 1.5 Years
• 2 Years
• In the Min. Relevance Score (%) box, enter the minimum relevance score items must have to be suggested as cross-sells. You can enter any value from 50 to 100; the default value is 65.

The ML model calculates the relevance score based on several factors, such as how often items are sold together. A score that’s too high may yield a small number of suggestions, while a too-low score may bring weaker suggestions.

• In the Max. Number of Suggestions box, specify the maximum number of ML-generated suggestions the system can add for an item after each generation. These suggestions are added on the Related Items tab of the Stock Items (IN202500) or Non-Stock Items (IN202000) form. For more details about this setting, see Limit the Number of Generated ML Suggestions.
• Select the Add Relations as Active check box to make the system do the following:
• Remove all suggestions of related items created during the previous suggestion generation. The system does not remove related items that were added manually.
• Make new suggestions immediately active.

To further refine the generation of cross-selling suggestions, you can also do the following on the Machine Learning tab:

• In the Generated Suggestion Notification box, select the notification the system sends when it generates suggestions.
• Add to the Excluded Item Classes table any classes whose items you need to exclude from the generation.
• Add an order type to the Excluded Order Types table if you need to exclude the AR transactions generated by this order type from the set of records used to generate cross-sell suggestions.

As mentioned, you can limit the number of ML-generated suggestions the system can add to the Related Items tab of the Stock Items (IN202500) or Non-Stock Items (IN202000) form after for each item during generation. You specify this number in the Max. Number of Suggestions box on the Sales Orders Preferences (SO101000) form. The Add Relations as Active check box on the same form affects this number as follows.

• If the check box is selected, the number limits the total number of suggestions—both approved and unapproved.

A suggestion is unapproved if the Accepted ML Suggestion check box is cleared on the Related Items tab of the Stock Items or Non-Stock Items form.

• If the check box is cleared, the number limits only unapproved suggestions. The system does not take into account approved suggestions.

Suppose that a stock item has 3 approved and 2 unapproved suggestions. In the Max. Number of Suggestions box, you have specified 3. On the next generation, the ML model returns 4 suggestions. The resulting number of suggestions on the Related Items tab depends on the state of the Add Relations as Active check box:

• Selected: The system deletes all existing suggestions and adds 3 of 4 suggestions of the latest generation: those with the highest relevance score. The total number of suggestions becomes 3. If any of the new suggestions coincide with the deleted suggestions for which the Accepted ML Suggestion check box had been selected, they remain approved.
• Cleared: Because the item already has two unapproved suggestions, the system adds 1 unapproved suggestion: the one with the highest relevance score. The number of unapproved suggestions becomes 3, the total number of suggestions is 6.

You can refine the suggestion for any item class on the Machine Learning tab of the Item Classes (IN201000) form. These settings are copied to new items of the class.

For greater precision in the generation of cross-selling suggestions, you can refine generation for any item class. These settings affect all new items with the selected class. You do this on the Machine Learning tab of the Item Classes (IN201000) form.

In the Cross-Sell Assistant Settings section, you specify the following settings:

• Preferred Item Classes: An option indicating the item class or classes whose items are preferred—that is, will receive a higher relevance score—when cross-sell suggestions are generated for an item of this class.
• Items of the same class get a higher relevance score (Same Class).
• Items of different classes have a higher relevance score (Other Class) .
• The system doesn’t take item classes into consideration when it calculates the relevance score (All Classes, the default option).
• Price of Suggested Items: An option indicating how the system considers the price of analyzed items while generating cross-selling suggestions for an item of this class: It will suggest items with lower default prices (or the same price) only, or it will suggest items with any default price. For these comparisons, the system uses the default price of items on the Price/Cost tab of the Stock Items (IN202500) and Non-Stock Items (IN202000) forms.

You can also adjust these settings at the item level on the Related Items tab of the Stock Items and Non-Stock Items forms.

To generate cross-selling suggestions, you click Process on the Generate Cross-Selling Suggestions (ML504000) form. The system analyzes stock and non-stock items, along with sales documents that include them. Based on selling patterns, the system creates cross-selling suggestions.

You can monitor the process stage in the Progress box. Also, the system displays informational messages with the about the progress.

You can use the Frequency box to schedule weekly or monthly automatic generation or keep it manual by selecting On Demand.

To review the data that is sent to the ML model, you can use two generic inquiry forms:

• IN-CrossSellWithML: Lists the AR transactions used in the generation of suggestions.

You can modify a copy of this generic inquiry to filter the data that’s sent to the model. The conditions on the Conditions tab of the Generic Inquiry (SM208000) form can be modified to include or exclude certain transactions. For instance, you might exclude a customer whose unusual purchasing behavior shouldn’t influence suggestion results.

• IN-RelatedItemFeedback: Records user feedback. Each time a user approves or deletes a suggestion, information about it appears in this generic inquiry.

Non-administrative users can’t open these generic inquiry forms through searches or workspaces; they can be opened only by first opening the corresponding inquiries on the Generic Inquiry (SM208000) form.

When the system completes the generation of cross-selling suggestions, you approve or delete them on the Manage Cross-Selling Suggestions (IN503500) form. Suggestions for each item also appear on the Related Items tab of the Stock Items (IN202500) and Non-Stock Items (IN202000) forms.

You can approve or delete any number of suggestions by selecting the appropriate option in the Action box of the Selection area. To process only particular suggestions, you can select the unlabeled check box for each suggestion and click Process on the form toolbar. Alternatively, you can process all suggestions at once by clicking Process All. If you approve or delete a suggestion, it won’t be shown on this form after future generations. The ML model reduces the score of the deleted suggestions.

When you approve a cross-selling suggestion, the system selects the Accepted ML Suggestion and Active check boxes for the cross-sell item on the Stock Items or Non-Stock Items form. You can also approve or delete a particular item’s suggestions directly on these forms.

If you’ve selected the Add Relations as Active check box on the Machine Learning tab of the Sales Orders Preferences (SO101000), you don't need to perform manual approval; the system selects the Active check box automatically. You approve the suggestions only to send positive feedback to the ML model.

When you approve a suggestion on the Manage Cross-Selling Suggestions (IN503500) form, the system removes it from the list. If you need to review all non-deleted suggestions—including approved ones—you can find them all on the Cross-Selling Suggestions (IN409500) form.