We would all like to have a crystal ball, starting with retailers and consumer goods companies looking for new market trends, or logistics companies worried about a possible storm.
Now, with a world of data at their fingertips (or, at least, at their keyboard), all these companies are very close to being able to make real-time forecasts in a wide variety of areas, if they are able to take advantage of the right artificial intelligence and machine learning tools.
For retailers, supply chains and consumer goods companies, accurate forecasting of demand has always been a key factor for good business planning, stock management, logistics rationalization and customer satisfaction. client. Accurate forecasts are essential to deliver the right product to the right place in the right quantity.
Customers get upset when the products they want to buy are out of stock, but overstocking means high and unnecessary costs. Businesses lose more than a trillion dollars a year due to poor inventory management (IHL Group), while a 10-20% improvement in the accuracy of demand forecasts can reduce inventory costs by 5% and increase revenue by 2-3% (Notes from the AI Frontier, McKinsey & Company).
However, stock management is only one of the many activities that are supported by demand forecasting: retailers also need to size the staff of their stores and warehouses in the busiest seasons, plan their offers and evaluate various factors that can influence physical and online sales traffic.
As retailers expand their product offerings and geographic reach, the data available becomes more complex and making accurate forecasts more difficult. Unrestricted activities during the pandemic have only aggravated bottlenecks in supply chains and made forecasting more difficult, as changes have been so rapid.
But today retailers can rely on machine learning to achieve great accuracy in their demand forecasts thanks to Vertex AI Forecast, one of the latest creations by researchers at Google Brain that in a very short time will be available for companies.
Model a million references in half a day
Vertex AI Forecast can support data sets of up to 100 million rows containing many thousands of product lines from BigQuery or CSV files. Its powerful modeling engine will automatically process the data and evaluate hundreds of different models to find the most efficient one, before providing optional settings that are easy to adjust even without advanced computer skills.
The user can then ask the models to target specific characteristics (for example, color, brand, promotions, or e-commerce campaigns) and set budgets to create the forecasts. Given the high volatility of market conditions, retailers need a system that is agile and capable of learning quickly: Vertex AI Forecast is capable of creating an accurate model for 1 million reference/store combinations in just 12 hours.
The key element of Vertex AI Forecast is model architecture search, where hundreds of model configurations and architectures are evaluated. This algorithm allows Vertex AI Forecast to consistently find the most efficient settings for a wide variety of clients and data sets.
Google has built the brains that allow demand forecasting in a contextualized and non-intrusive way, merging the art with the (computer) science of accurate demand forecasting. In benchmark tests with Kaggle datasets, Vertex AI Forecast had the highest 2.5% accuracy in the World’s Top Forecasting Competition.
From weather to leather, Vertex AI handles all kinds of data
Thanks to Vertex AI Forecast’s hierarchical forecasting capabilities, retailers can generate highly accurate forecasts that work at multiple levels (for example, linking demands at the item, store, and region levels), minimizing the problems created by organizational silos.
Vertex AI can ingest large volumes of unstructured data to understand the most relevant “signals” such as weather, commodity prices, shipping and freight costs. This allows companies to obtain valuable information and understand the importance of the main factors that drive demand.
This not only applies to the overall forecast, but also to each of the elements in each of the points. For example, in the apparel category, promotions may be the main drivers of demand on weekdays, but not on holidays. This type of information is priceless when making decisions based on forecasts.
Major retailers have already transformed the way they work and are reaping the rewards of more accurate forecasts.
»For Magalu, Vertex AI Forecast has meant a transformation of our forecasts. Forecasts are now made at the distribution center level and, in addition, we have managed to reduce forecast errors“, it states Fernando Nagano, Director of Analysis and Strategic Planning of Magalu.
“In four weeks of “live” forecasts – adds Nagano – we observed a significant reduction in errors (WAPE) compared to our previous models. This highly accurate information has helped us plan stock allocation and replenishment more efficiently, as well as ensuring that items are in the right locations at the right time, allowing us to meet customer demand and better manage costsyes».
According Amaresh Siva, Senior Vice President of Innovation, Data and Supply Chain Technology at Lowe’s: “With more than 1,700 Lowe’s stores spanning the entire United States, accurate baseline forecasts are critical to making decisions about allocating stock to each store and replenishing items in stores. peak demand periods. Using Vertex AI Forecasting, Lowe has been able to create accurate hierarchical models that balance references and forecasts at the store level. These models take into account our stock at the store, reference and region level, as well as data on promotions and many other factors, thus allowing for more accurate forecasts.”.
Additionally, our ecosystem of partners, including o9 and Quantiphi, are already integrating Vertex AI Forecast to offer value-added services to their clients. For more information on demand forecasting with Vertex AI, you can contact your local Sales Representative or take a free trial of Vertex AI here.