appeal to target customers?
aid with overstock clearance?
optimize margins?
introduce new products or new product categories?
increase average order size (such as through multiple product bundles)?
aid with load balancing (i.e., promote orders at a certain time of day or week)?
conform with typical customer purchases (based on transaction data)?
Step 2: Check to ensure that you have data to support each decision criterion. For example, to test whether there is appeal to target customers, you might refer to demographic or psychographic information or you might use survey data if it is available.
If there's not enough quantitative data available, you can rely on qualitative data. For example, you can use your industry knowledge and common sense to decide whether a certain product would promote orders during the week rather than the weekend.
Step 3: Assign each criterion chosen a relative weight in percentage terms. For example, if a primary company strategy is to attract a different customer segment, the "target customer appeal" criterion may be weighted more heavily than whether the products "conform with typical customer purchases."
Step 4: Construct an analytical model that can evaluate each product (or occasion) using the criteria identified; rank how well each will meet the company's merchandising goals.
Step 5: Launch the algorithm. Run various promotions on the site, using the merchandising algorithm to choose which products (or occasions) to use within each promotion.
Step 6: Once the algorithm is running, test its effectiveness by tracking customer usage. Because of the Internet's unique ability to allow real-time product adjustments (pricing, placement, etc.), it has an advantage over the offline retail environment. Track changes in conversion rate, average order size and sales by category.
