Making Sense of Forecasting 2.0 and the Role of AI

AI brain illustration

Listening to current retail technology discussions, it’s safe to say that artificial intelligence is the early favorite for buzzword of the year, with countless taglines promising unprecedented productivity improvements based on AI. Advanced forecasting is often cited as one of the top areas where AI holds great promise – but how do you separate the hype from the reality?

As retailers make big investments in AI technologies that can transform their business, a key focus is increasing supply chain effectiveness and creating more accurate forecasts. However, prior to implementing new solutions, retailers need to have a clear understanding of what advanced forecasting actually entails, how AI will play a role in advanced forecasting, and what their specific forecasting strategy needs really are.

Machine Learning and AI – What Do They Truly Do?

By definition, artificial intelligence means “the capability of a machine to imitate intelligent human behavior.” In retail, a prime example of an AI use case is machine learning, which provides the ability for the machine to utilize a retailer’s data to deliver advanced insights that can continue to evolve based on various variables the retailer is looking to analyze. For example, a retailer can implement a machine learning solution that utilizes cloud computing to expand the number of machines tackling massive computations in areas such as forecasting. Furthermore, in advanced forecasting, the presence of machine learning algorithms can lead to success in managing clean attributes and planning for future demand.

Prior to machine learning, retailers were limited by a lack of hardware and human resources. They simply could not execute the amount of data analysis they needed to effectively create an advanced forecasting environment. However, at the moment, retailers are confused about what AI really is, the capabilities it provides, and concerns that AI will lead to job losses. These fears are unfounded, however; if utilized correctly, AI technologies such as machine learning can serve as advanced assistants to help retail executives realize forecasting insights that were not achievable prior to AI.

Traditional vs. Advanced Forecasting—What is the difference?

Traditional forecasting means retailers analyze their data through a series of Excel spreadsheets to gather historical data on a particular product or SKU and forecast demand based on the history. In this process, retailers also have to manually make decisions on certain categories such as “like items”. For example, in this process, a retailer would ask “since we’ve never had that ‘new shirt’ before, we’ll say it is like this ‘old shirt’.” In this scenario, traditional forecasting takes the sales history for one particular item and utilizes this data to forecast new item demands. The issue here is that retailers are limited to the amount of data they can analyze, leading to a single-sided view of the customer and potentially creating biased forecasting.

On the other hand, advanced forecasting, or “Forecasting 2.0,” provides retailers the ability to combine both historical data and real-time data to quickly understand their history and drive their organization’s future. Advanced forecasting is crucial for retailers who are operating a multi-channel organization as it helps retailers introduce new products into new channels and create an accurate and informed forecasting strategy to anticipate and meet demands across all channels.

For example, by incorporating machine learning algorithms, retailers can gain a 360-degree view of the customer and work to create an environment of insight-based forecasting that brings in and prepares for ‘what-if’ scenarios, such as weather patterns, extreme or complex seasonality or multiple channel exposure, among other factors. Additionally, advanced forecasting can help inform and drive additional merchandising activities such as planning, buying and assortment tasks.

Forecasting 2.0 utilizes attributes of all products, not just ‘new shirt’ and ‘old shirt.’ So instead of having the sales history of an ‘old shirt’ for three years (12 months x 3 years equaling to a total 36 data points), machine learning forecasting can use the entire data set for all items – including all sales history, all attributes, all locations and weather, etc. to create an accurate forecast for one item. Machine learning also removes the need for retailers to pick ‘like items,’ as it decides itself which ‘old Item(s)’ the ‘new shirt’ is closest to. The machine learning algorithms are self-learning, so they improve over time and learn as they take in more data. This results in far greater accuracy and fewer errors – which in turn reduces overstocks, out-of-stocks and markdowns and improves profitability.

Forecasting 2.0 can help take retailers’ capabilities to the next level. However, given the critical importance of forecasting, retailers should have a clear understanding of the underlying technology, ensuring that it truly embraces AI components such as machine learning to help retailers thrive in a complex, multi-channel world.

Max Bruni is SVP of Solutions, PlumSlice Labs

Leave a Reply

Your email address will not be published. Required fields are marked *