Mother’s Day is a time for children and spouses to express appreciation for the moms in their lives — and a time for stores to help them to do just that. Whether they’re shopping for flowers and chocolates, or jewelry and apparel, shoppers are looking for the best deals across a variety of categories. Retailers want to capture the potential increase in demand from holidays like this, and often use promotions to do so. But making the right decisions about promotional strategy isn’t always that easy.
For while holidays and special events are a great opportunity for merchants to generate an increase in-store traffic and spark more activity within specific categories, traditional pricing systems aren’t built to help them plan promotions for these events in the smartest way.
Even systems that employ demand forecasting often attempt to align sales weeks on a year-on-year basis, capturing only rough, seasonal effects on customer behavior. But seasonal behavior alone tends to hide the effect of significant events, instead capturing only non-specific and gradually changing fluctuations in sales patterns.
Special events like holidays occur at what appear — at least to traditional forecasting systems — to be arbitrary times. If the system doesn’t account for Mother’s Day, for example, that early May change in behavior would just be folded into the spring season projections rather than used to correctly forecast highly category-specific, holiday-related lift. The floral department of a large grocery store undergoes extreme change just before Mother’s Day. Without the Mother’s Day holiday variable present, a demand model will look for other variables to which it can attribute the change in behavior.
Elasticity estimates, and more general seasonal effects, will be incorrectly skewed by what is uniquely a Mother’s Day effect. So the forecast system will fail to capture the extreme peak of sales in Mother’s Day flowers. Additionally, retailers may be led to erroneously believe that they can dramatically lift floral sales with price reductions and promotions at other times of the year
Just to make the problem more complicated, important events can differ not just by category, but by store. For example, in some areas Passover is a major holiday, while in others, the weekend of the Alabama versus Clemson football game is a significant event, having an impact on many types of items. Each retail outlet likely has approximately five to eight events that have a notable impact on sales in a significant number of categories, and it’s important to know what they are and to integrate them into science-based, data-driven forecasting. In reality, for many retail segments, Mother’s Day may not be remarkable enough across the store for retailers to designate it as a significant event, worthy of special modeling. Another day or weekend may have more importance for them. This is another instance where historical data, rather than gut instinct, must be our guide.
So how can retailers plan smart promotions around these events for maximum lift? It starts with identifying the historical occurrences of these significant events, thus allowing the demand forecasting system to learn their effects. This way, the system’s learning algorithm can associate a particular sales lift with a specific historic event, enabling the retailer to anticipate the rise in sales for future occurrences of the events. More importantly, merchandising and category managers can learn which pricing and promotional strategies can induce the best response among the retailer’s most important customers.
With any promotion, holiday promotions included, it is important that retailers understand whole-category and whole-store effects. Some promotions can cannibalize sales from elsewhere in the category. Other promotions, especially around holidays, can drive incremental traffic into the store. Further, certain promotions may appeal exclusively to lower-value customers who are only on the hunt for bargains and won’t bring meaningful repeat business, rather than appealing to loyal customers whose continued patronage means long-term, sustained growth. So how can retailers tell when promotions are going to cost them rather than reward them?
With a price optimization system that analyzes segmented sales streams provided by the retailer, store managers can see clearly which customers engage with specific promotions and which don’t. This analysis can help forecast the impact of future promotions, and help category managers avoid creating offers that will hurt the bottom line rather than help. When significant events are identified in historic data, their specific effects can be understood on a segmented level.
Visibility to category-level promotional impact can help ensure that, if the retailer accepts a vendor-proposed promotion, the subsidy will be sufficient to make it a financially sound decision. For example, if a certain CPG manufacturer offers to fund a promotion on a brand’s products, the retailer needs to know it will not hurt their bottom line. When the promotion is modeled within the price optimization system, the analysis may reveal that the higher sales on that brand will drain sales from private-label products and have a significantly negative impact on margin. The retailer could then demand a higher subsidy from the manufacturer to run the promotion, protecting their finances in a way not possible with traditional pricing systems that lack such modeling capabilities. This evens the playing field for retailers, making them better judges of how brands’ promotions will affect category and whole store sales, as well as customer engagement.
Ultimately, special event promotions can mean big returns for retailers – if planned methodically and with a basis in science and smart data. Doing anything less risks loss of profit, loss of customers, and loss of long-term growth.
Dr. Paul Helman is Chief Science Officer of KSS Retail. He was previously a founding partner and Chief Scientist of Standard Analytics. Dr. Helman is a professor emeritus of computer science at the University of New Mexico, following a 25-year career of active academic research.