Helen Schepanik
Key takeaways
- 4:03Trade promotions may generate 20% to 70% of retail revenue while repeating weekly or biweekly, making them substantially more operationally demanding than the annual regular-sales cycle.
- 12:15AI-ready promotion forecasting requires data on promo types and mechanics, duration, participating SKUs, regular and promotional prices, discount percentage, locations, category hierarchy, layouts, and sales before and during the promotion.
- 19:31Forecast accuracy should be measured at the level relevant to the decision, such as SKU level for distribution-center replenishment or SKU-location level for store replenishment.
- 27:44A grocery retailer with 90 locations and about 20,000 SKUs achieved 99% product availability during promotions and reduced post-promotion overstocks by half after adopting AI forecasts.
- 28:44Machine-learning tools can deliver up to a 20% improvement in promotional availability and up to 50% fewer overstocks while centralizing promotion history for continuous improvement.
Questions Helen Schepanik answered
Yes. Machine-learning models can use category-level behavior and promotional sales from similar SKUs, which generally produces better results than history-dependent statistical methods for new items.
WEBINAR: MAKING EVERY TRADE PROMOTION EFFICIENT IN THE ERA OF AI31:51
They can be mapped manually when similarity data exists in the ERP system or identified automatically when no such mapping is available.
WEBINAR: MAKING EVERY TRADE PROMOTION EFFICIENT IN THE ERA OF AI32:52
Yes. Historical performance reveals successful and unsuccessful promotion patterns and provides the deep data foundation needed to train accurate machine-learning models.
WEBINAR: MAKING EVERY TRADE PROMOTION EFFICIENT IN THE ERA OF AI33:56
The preferred history is two to two and a half years, but one year can be sufficient. Because COVID-19 changed sales patterns, older history may sometimes be less relevant.
WEBINAR: MAKING EVERY TRADE PROMOTION EFFICIENT IN THE ERA OF AI34:57
Yes. The model considers discount percentage, pre-promotion sales and prices, future promotional price, advertising type, and merchandising layout because these factors affect promotional demand.
WEBINAR: MAKING EVERY TRADE PROMOTION EFFICIENT IN THE ERA OF AI35:59
In Helen Schepanik's words
“The main thing is the right basis of sales. It is the hardest thing for the model to understand the baseline sales.”
WEBINAR: MAKING EVERY TRADE PROMOTION EFFICIENT IN THE ERA OF AI41:07