B Appendix: Option Forecast Details

Options Forecast is a module within Oracle Retail Artificial Intelligence Foundation (AIF) services, developed to provide advanced analytical support for retail assortment planning. This module leverages Oracle Retail’s robust AI-driven analytics capabilities to transform vast amounts of historical sales data, item lifecycle information, and product attributes into actionable insights for retailers. By examining which attributes and characteristics are associated with high-performing products, Options Forecast helps identify patterns and preferences that drive successful sales outcomes.

The module delivers a comprehensive set of outputs, including Attribute Weights, Optimized History, Total Option Count, Option Count by Attribute Value, and Sales Potential. These outputs help retailers understand which product features most influence customer demand and how historical trends can inform future assortment selections. In addition, Options Forecast performs granular analysis at the store level, evaluating the number of units sold, pinpointing lost opportunities and sales, and highlighting instances of missed or delayed deliveries.

Through these advanced analytics, Options Forecast equips retailers with precise, data-driven recommendations for assortment planning. By identifying growth opportunities and reducing the risk of missed sales, the module enables organizations to strategically optimize their product offerings for upcoming seasons, ultimately supporting better business outcomes and competitive advantage in the retail marketplace.

The Options Forecast methodology within AIF automates key aspects of the Assortment Planning solution by providing strategic recommendations for assortment strategy. This methodology is structured around a systematic run setup, comprising the following steps:

• Life Cycle: Analyzing the complete lifecycle of products to understand performance trends.

• Optimize History: Leveraging historical sales data to identify patterns and inform future decisions.

• Attribute Weights: Determining the relative importance of product attributes that drive sales.

• Incrementality: Assessing the incremental value each product brings to the overall assortment.

• Sales Potential: Estimating the future sales potential of various assortment options.

Through this structured approach, Options Forecast enhances the effectiveness and precision of assortment planning decisions. This appendix provides details for the parameters involved in the calculations of each of these steps.

Life Cycle

The life cycle of an item is calculated by identifying when the item first started selling (the start date) and when it stopped selling (the end date) using historical sales data for each store or location. By calculating the number of weeks the item was available for sale, this process helps determine how long each item was actively selling, which is useful for planning future assortments.

Table B-1 Life Cycle Parameters

Parameter Name	Parameter Description	Input Specifications (Default, Range, and so on)	Parameter Explanation (AIF)
OPT_LC_EOL_CHECK_WKS	Number of Weeks to Check for End of Life Promos/Stockouts	Default: 4 (weeks)	Specifies the number of weeks to consider when evaluating end of life promotions or end of life stock outs. For example, if it is set to 8, the system will look at the sales data for the last 8 weeks of an item's life cycle to check if the item had significantly more promotions or stock outs than previous weeks.
OPT_LC_EOL_MIN_SO_PCT	Percentage Threshold to Consider End of Life Stock Outs	Default: 0.3 Range: 0.0 - 1.0	Specifies the minimum percentage to consider that week in the end of life window as a stocked-out. An item is considered heavily stocked out when the average percentage of store count with inventory (percentage calculated wrt max store count with inventory over entire life) is higher than this threshold (for example, for default, 30%).
OPT_LC_ITEMS_STILL_SELLING_FLG	Still Selling Filter	Y/N Default: Y	Drop items that are still selling. An item is considered still selling if it has at least one positive sale in the last OPT_LC_EOL_PAST_N_WK weeks from the end of the training period. Y means consider this filter in rule extraction and N means do not apply this filter.
OPT_LC_ITEMS_WITH_FEW_SLS_FLG	Few Sales Filter	Y/N Default: Y	Y: Drop items with less than OPT_LC_MIN_NUM_POS_SLS_WKS sales data points (weeks). N: Do Not drop it. Y means consider this filter in rule extraction and N means do not apply this filter.
OPT_LC_LOW_SELLERS_FLG	Low Sellers Filter	Y/N Default: N Y: Drop items that are low sellers. N: Do not apply filter.	This flag controls whether to exclude items that are low sellers. Specifically, these are items that sell less than one unit per week (fixed value) on average over all weeks in their time series. If Y, items that meet this condition are excluded from the exit date estimation process.
OPT_LC_MIN_NUM_POS_SLS_WKS	Minimum Number of Weeks with Non-zero Sales	Default: 5 (weeks)	Minimum number of non-zero sales weeks an item must have to calculate the lifespan. Items with less than this value are not considered for exit dates rules generation.
OPT_LC_RECENCY_IN_WKS	Item Recency Minimum Weeks	Applied with OPT_LC_RECENT_ITEMS_FLG. Default: 8 (weeks) For example, if an item has had its first sale in the past eight weeks, exclude from exit date estimation.	Specifies the number of weeks to consider when determining whether an item is recent. Items that have their first sale within this number of weeks from the end of the training period are considered recent and may be excluded from the analysis if OPT_LC_RECENT_ITEMS_FLG is Y.
OPT_LC_RECENT_ITEMS_FLG	Recent Items Filter	Y/N Default: Y Y: Drop items that started selling recently, that is, that started selling less than OPT_LC_RECENCY_IN_WKS ago (for example, 7 weeks ago). N: Do not apply filter.	Controls whether to exclude items that have recently been introduced. If Y, items that have their first sale within OPT_LC_RECENCY_IN_WKS weeks from the end of the training period are excluded from the exit date estimation process.
OPT_LC_RETURN_SLS_FLG	Only Returns Filter	Y/N Default: N Y: Drop sales entries of items if they are exclusively returns. N: Do not apply filter.	Controls whether to exclude sales entries that are exclusively returns, as returns can artificially extend the lifespan of an item.
OPT_LC_SLS_AFT_98_PRC_FLG	98% Cumulative Percentage Filter	Y/N Default: Y Y: Exit date is estimated as the week when the cumulative sales reach 98% of the total sales. N: Exit date is estimated as the last week with sales.	Determines the method used to estimate the exit date for an item. If Y, the exit date is estimated as the week when the cumulative sales reach 98% of the total sales. If N, the exit date is estimated as the last week with sales.
OPT_LC_SLS_BFR_FST_RCPT_FLG	Sales Before First Receipt Filter	Y/N Default: Y Y: Sales that occur before the first receipt week ID are excluded from exit date estimation. N: Do not apply filter.	Controls whether to exclude sales that occurred before the first receipt week ID. For example, if an item was first received in week 10, and OPT_LC_SLS_BFR_FST_RCPT_FLG is Y, any sales data for that item prior to week 10 will be excluded from the analysis.

Optimize History

Optimized History is a careful process that helps retailers get a true picture of product demand by correcting sales data for issues related to inventory, such as stockouts (times when products were out of stock) and limited variety on the shelves. This allows retailers to answer important questions such as, “What would our sales have looked like if we always had enough of each product available?”

Optimized History is computed both at the item (stylecolor) and attribute level in two separate algorithms. This allows AIF to produce predictions for both how a specific stylecolor would optimally sell, and how any item possessing a certain attribute (that is, being of color grey, or brand Nike) would sell.

Using this data, a statistical model is trained to understand overall customer demand and how much sales are affected when inventory is limited, for each attribute or item. This model helps adjust both the reported sales and the number of product options (option count), reflecting what would have happened if inventory was never an issue. The result is a much clearer and more accurate foundation for planning future assortments, allowing retailers to better decide how much variety in color, style, or material they should carry, even when planning for products that have not been sold before. This ensures future predictions and assortment choices are based on real customer preferences, not just what happened to be in stock in the past.

Table B-2 Optimize History Parameters

Parameter Name	Parameter Description	Input Specifications (Default, Range, and so on)	Parameter Explanation (AIF)
OPT_OH_ITEMS_MISSING_SCWI_FLG	Missing Store Count with Inventory Filter	Y/N filter for data cleaning. Default = Y Y: Drop items with more than OPT_OH_ITEMS_MISSING_SCWI_PCT% having zero or null store_count_with_inv values. N: Do not apply filter.	This flag controls whether to exclude items with a high percentage of missing or zero Store Count With Inventory (SCWI) values. If Y, items with more than OPT_OH_ITEMS_MISSING_SCWI_PCT percentage of weeks having zero or null SCWI values are excluded, indicating that they may have unreliable inventory data and may not be suitable for optimize history corrections.
OPT_OH_ITEMS_MISSING_SCWI_PCT	Missing Store Count with Inventory Percentage	Continuous filter for data cleaning, paired with OPT_OH_ITEMS_MISSING_SCWI_FLG. Range: 0.0-1.0, Default: 0.2 A higher threshold value (such as 0.5 or 50%) will result in more lenient filtering, including items with a larger percentage of missing or zero store_count_with_inv values.	Maximum percentage of sales weeks where store count with inventory (store_count_with_inv column in pmo_options_activities) can be either zero or null. Default is 20%, which means that items with more than 20% of sales weeks having zero or null store_count_with_inv values will be excluded.
OPT_OH_ITEMS_STILL_SELLING_FLG	Still Selling Filter	Y/N filter for data cleaning. Y: Drop items that are still selling (drop items with sales in the last OPT_LC_EOL_PAST_N_WK weeks). N: Do not drop items that are still selling. Default = N	This flag controls whether to exclude items that are still actively selling. If Y, items with sales in the last OPT_LC_EOL_PAST_N_WK weeks are excluded, indicating that they may not have reached the end of their life cycle and may not be suitable for optimize history corrections.
OPT_OH_ITEMS_WITH_FEW_SLS_FLG	Few Sales Filter	Y/N filter for data cleaning. Y: Drop items with less than OPT_LC_MIN_NUM_POS_SLS_WKS weeks where sales are positive, in which the default is OPT_LC_MIN_NUM_POS_SLS_WKS=5 weeks. N: Do not drop items meeting this condition. Default = N	This flag controls whether to exclude items with limited sales data. If Y, items with fewer than OPT_LC_MIN_NUM_POS_SLS_WKS weeks of positive sales are excluded, indicating that they may not have sufficient data for reliable optimize history calculations.
OPT_OH_MAX_WKS_APPLY_ED_RULE	Maximum Weeks to Apply Exit Date Rule	Default = 4 For example, with OPT_OH_MAX_WKS_APPLY_ED_RULE=4, if the estimated exit week is 9 weeks beyond the expected life cycle (defined here as start week + weeks to exit), the rule will not be applied because 9 > 4.	Maximum allowed difference in weeks between the calculated exit date and actual life of an item. If the difference is higher than OPT_OH_MAX_WKS_APPLY_ED_RULE, exit date rules are not used to define this item’s end of life in the optimize history estimation phase. This parameter controls the tolerance for the difference between the estimated exit week ID and the actual life of the item. If the difference is too large, the extracted exit date rule is not applied, indicating that the rule may not be reliable for optimize history corrections.
OPT_OH_RECENT_ITEMS_FLG	Recent Items Filter	Y/N filter for data cleaning. Y: Drop items that started selling recently, that is, that started selling less than OPT_LC_RECENCY_IN_WKS ago (for example, 7 weeks ago). N: Do not drop it. Default = N	This flag controls whether to exclude items that have recently been introduced. If Y, items that have their first sale within OPT_LC_RECENCY_IN_WKS weeks from the end of the training period are excluded, indicating that they may not have sufficient historical data for reliable optimize history calculations.
OPT_OH_ATTR_CORR_THRESHOLD	Optimize History Model Type	Range: 0.0 - 1.0 Default: 0.9	Threshold of sales percentage to compute Optimize History by Attribute corrections. 90% means that for a given Attribute, top Attribute Values that cumulatively contribute 90% of the total sales will be corrected.

Attribute Weights

Attribute weight refers to a numerical value assigned to each product attribute, representing its relative importance in determining product similarity within a retail assortment. Products are typically described by several attributes, such as brand, color, collar type, neckline, and sleeve length. However, these attributes do not all contribute equally when assessing how similar two products are. For example, brand and color may have a greater influence on perceived similarity compared to collar type or sleeve length. Attribute weights serve to quantify this difference in importance by assigning higher weights to more influential attributes.

Incrementality

An incrementality curve is a curve that models projected sales across a range of option counts within an assortment. This curve is generated by simulating sales outcomes using estimated arrival rates (which reflect seasonality), product attractiveness scores, and store-specific assortment data. The incrementality curve illustrates the incremental sales lift associated with each additional option introduced to the assortment. By depicting the relationship between the number of available options and the corresponding incremental sales, this curve enables retailers to evaluate the marginal benefit of assortment expansion and supports data-driven assortment planning decisions.

Table B-3 Incrementality Parameters

Parameter Name	Parameter Description	Input Specifications (Default, Range, and so on)	Parameter Explanation (AIF)
OPT_IC_MARKET_SHARE	Market Share	Default = 0.5 Range: 0.0-1.0	Specify the retailer market share, which is the fraction of all potential customers expected to make a purchase when all products are available. Lower market share means more customers choose not to purchase; higher market share means more customers shop with the retailer.

Sales Potential

Sales potential is an estimated measure of the expected sales that a product or assortment can achieve within a specific period. This metric leverages factors such as historical sales data, product attributes, and product category to forecast likely sales outcomes. Primarily, it models information on product sales and attributes, as in color, brand, and so on, to understand how well certain attributes, as well as combinations of attributes, sell. Sales potential helps retailers identify high-opportunity products or attribute combinations and support more accurate, data-driven planning.

Table B-4 Sales Potential Parameters

Parameter Name	Parameter Description	Input Specifications (Default, Range, and so on)	Parameter Explanation (AIF)
SP_MAX_ATTR_VAL_DOMINATION	Maximum Attribute Value Domination	Range: 0.0 - 1.0 Default: 0.5	Specifies the maximum percentage of items an attribute value can dominate for a given attribute to be included in training. For example, if all items have a single attribute value, that attribute is not useful for model training. A value of 0.5 means that if an attribute value is present in more than 50% of the items, the attribute will be dropped.
SP_MIN_ATTR_FREQUENCY	Minimum Attribute Frequency	Range: Integer value >= 1 Default: 1	Specifies the minimum number of items that must have a non-null value for an attribute to be included in training. For instance, if only three items have sleeve-length populated, then sleeve-length will be dropped from model training if SP_MIN_ATTR_FREQUENCY is greater than 3.
SP_MIN_ATTR_VAL_DIVERSITY	Minimum Attribute Value Diversity	Range: Integer value >= 2 (recommended) Default: 2	Specifies the minimum number of unique attribute values required for an attribute to be included in training. Attributes with low cardinality (that is, fewer unique values) may not be useful for model training.
SP_MIN_LIFECYCLE_WINDOW	Minimum Lifecycle Window	Range: Integer value >= 1 Default: 1	Specifies the minimum historical lifecycle length in weeks for an item to be included in training. Items with a short lifecycle may not provide enough data for effective model training.
SP_MIN_MAX_STORE_CNT_WITH_INV	Minimum Value for Max Store Count with Inventory	Range: Integer value >= 1 Default: 1	Specifies the minimum value for the maximum store count with inventory for an item to be included in training. Items with low inventory levels across stores may not be representative of the overall sales pattern.
SP_MIN_SALES_UNITS	Minimum Sales Units	Range: Integer value >= 1 Default: 1	Specifies the minimum amount of sales units required for an item to be included in training. Items with very low sales may not provide enough data for effective model training.