May 7, 2019
When it comes to artificial intelligence (AI) and the realm of machine learning, the primary impetus for increased productivity is data. The more high-quality data collected about product production, sales, or any aspect of the wholesale or retail industry, the better demand forecasting becomes.
Intermarché, a large grocery chain in France, pitted its existing data processing system against AI. The contests included both massive warehouses containing 30,000 different products as well as specialty centers with much fewer frozen items. In the end, the product types or numbers made no difference in the function of the AI systems. Expectations were exceeded across the board, and the grocery giant enjoyed 95% accuracy in demand forecasting with a 75% reduction in errors. As powerful as machine learning systems are, the success of their processing power comes down to the amount of in-depth data fed into them.
Any type of automated “smart” system cannot operate properly without data. The main problem with relying on these systems is that the sources for this information may not be sufficient for the AI’s needs. Any company that sources data inefficiently or fails to recognize the top sources of the most powerful information harms the AI’s ability to do its job.
Any business that involves the sale of products needs accurate and comprehensive demand forecasting in order to direct manufacturing to produce just enough and not too much based on how many items business clients or consumers are likely to buy. This is exceptionally important for corporations like Intermarché that has time-sensitive products. No matter what industry the company represents, no one wants to be left with old, unsellable stock.
AI systems need data. The more, the better when it comes to providing the right tools for their machine learning components to devise accurate histories and future models of production, inventory, and sales. In order to provide the most comprehensive collection of information to make this process work with the greatest amount of accuracy, the three top sources of data should be included in every system.
Historical records establish a pattern for machine learning
In the same way that people learn from information gathered in the past, machine learning systems also require historical data to come up with accurate results. The Intermarché experiment included three years’ worth of past sales records in order to achieve the almost perfect accuracy rating. The more information you give the digital system, the better the outcome will be.
One important thing to remember is that manipulating the information or using creative processes to categorize or manage it in any way can skew the AI’s ability to use it successfully. Raw data is best. After all, if you intend to trust your manufacturing and sales activities based on a smart system’s recommendations, it makes sense to leave the fallible human element out of things.
Ongoing contextual data allows for AI maps of market influence
After historical information is fed cleanly into the system, contextual data is the next top source for AI usage. This includes any factors like special events or holidays, seasonal sales changes, or other cyclical things that influence sales in either one or several specific stores. It also includes product categorization and the geographic location of shops within the city or country. This type of data adds another layer of understanding that the AI system can use to create a more comprehensive forecast.
The initial combination of contextual and historical data serves to overlay sales history with factors that could influence it. Most people can figure out that, for example, frozen desserts sell more frequently in the hot season while soup sells faster when it is cold. However, the digitized system uses the actual data collected over time to associate sales ups and downs with logical factors. This allows a more complete and thorough demand forecasts specific to the season and other pattern-changing things.
Product specifications, inventory, and locations help AI systems evaluate the details
Finally, specific product information for every SKU a company offers ensures that nothing is skipped when it comes to demand forecasting. Even the least expensive product with the fewest sales provides data to the artificially intelligent digital processor that creates a more robust future forecast that companies can rely on. Using this data destroys potential visibility gaps caused by limited products, seasonal interest, or new data without an established history yet.
How Intermarché’s test demonstrated the 3 data source importance
The intention of the grocery chain’s AI experiment included a plan to make demand forecasting a more automatic process and ensure higher degrees of accuracy. The first tests included historical data gathered for 36 months for just two of the warehouses used by the company. After these initial tests, 10 more warehouses were added. This part the third type of specific data on trial since all of the warehouses included some products that did not align with logical forecast models in the past. For example, road salt used during icy or snowy conditions fell under this heading. With all three types of data in certain into the system, the resultant forecast is accurate and easy.
When errors went down by 75% and demand forecasting accuracy rose by 15%, no question remained about the power of machine learning when it came to demand forecasts processes. These excellent results continued across product types and locations. Intermarché gained additional important insight that allows them to increase ROI and expand business with less risk and more profit.
The automated information gained by using AI and machine learning demand forecasting methods allow for productivity and profitability gains across the board, they can minimize loss for time-sensitive products like groceries, increase income from event-focused promotions, and make the entire supply chain run more smoothly. However, all of this requires sufficient data from multiple sources fed into the system regularly. As long as you have the ability to transfer raw data to the AI system, your business will enjoy all the benefits that it brings to modern sales.
Kudos for the inspiration and source materials to Symphony RetailAI’s blog
About the author
Cristina Autino is Evo’s marketing specialist. Her previous internship was at Agenzia delle Entrate in Chivasso.
She has two degrees: a Master’s Degree in Economics and Business Leadership and a Bachelor’s Degree in Economics and Business from the School of Management and Economics in Turin.
Her interests are finance, marketing and advertising.