Glossary
What is AI in retail?
In short
AI in retail is software that forecasts demand, manages stock and prices, answers shoppers and watches how goods are displayed in stores.
AI in industries · AI in retail
MMOOZZGG.com editorsChecked: 1 min read
For a child
Imagine a school fair where you sell cookies. If you bake too few, they are gone in a minute; if you bake too many, half are left over. A smart helper in big stores looks at how much people bought before, what the weather will be and whether tomorrow is a holiday, and suggests how much to bring. Then the shelves are not empty, and food does not have to be thrown away.
In plain words
A store solves hundreds of small tasks every day: how much milk to order, what price to set, where to place a product, what to answer a shopper in the chat. This used to be done by experience and spreadsheets. Machine learning programs look at sales history, season, promotions and weather and calculate a forecast for each product in each store. Cameras notice empty shelves, chatbots answer about delivery and returns, and a recommendation system suggests products for the basket. For example, before a heat wave the system raises the water order for stores near the beach in advance.
For business
The main money in retail sits in stock: shortages lose sales, and surplus goes to write-offs and discounts. So most retailers start with demand forecasting and automatic ordering, then move on to prices, reviews and customer support. The project cost depends on the number of products and stores, the data quality in the accounting system and the integration with cash registers and the warehouse. The risks: the forecast breaks on new products and in crises, and personal recommendations need careful handling of shopper data under 152-FZ, Russia’s personal data law. Before the launch, the model’s forecast is compared with the current ordering method in a few stores, and the effect is counted in money. Our review of AI in retail covers the points of adoption with figures.
For a specialist
The basis is demand forecasting on hierarchical time series, inventory optimization, dynamic pricing, recommender systems and computer vision for shelf monitoring. The M5 competition gives a benchmark for methods (Makridakis, Spiliotis, Assimakopoulos, International Journal of Forecasting, 2022): participants forecast 42,840 hierarchical series of Walmart unit sales and submitted 30,490 bottom-level forecasts, whose sums gave the forecasts for the upper levels of the hierarchy. So hierarchical reconciliation of forecasts between levels is part of the task. In practice such series are often forecast with gradient boosting and external features: prices, promotions, the calendar, while statistical methods serve as a baseline. A separate difficulty in retail is intermittent demand, when a product has days without sales, so the error is scaled for each series to keep rare sales from distorting the score.
Example
A grocery chain connected demand forecasting to automatic ordering. In the morning the store manager sees the proposed order for each item and edits only the products where the model flagged low confidence.
Related terms
Questions and answers
Sources
Not sure whether your company needs this?
AI solutions map