Every day, retailers are making countless decisions that drive both the top and bottom line.
- Should we carry this new product?
- Which items and in what quantity do we buy for next season?
- Do we need more (or less) inventory at the store?
- What promotion should we run?
- How much do we markdown at the end of this season?
- How many people should we schedule to work?
While these decisions are all made by many different people throughout the organization, they all share a common starting point. Regardless of the question, the answer starts with a forward view of business performance. In short, they all need a forecast.
Because different tools were built to address a specific business problem, it is common for each to have a different prediction of future demand. Imagine planning for a promotion and the marketing expectation is that it will sell 200 units. However, the supply chain only stocks 100 units, and store operations only staffed the store for the regular demand of 50. If the promotion plan is correct, then we’ve just lost at least 100 units of sales, and customer service has suffered. Flip this scenario to a promotion plan of 100 units, 200 units in stock, and having extra staff. Now, if the promotion plan is correct, then we have 100 additional units to markdown and profitability is further reduced because of the extra staff. Forecast misalignments have a material impact on both the top and bottom line.
As you can see from the above example, it is essential that everyone in the organization has the same perspective on future demand. This is where a unified demand signal can bring tremendous value to a retailer. A unified demand signal is a forecast and price elasticity signal that is utilized by all processes that rely on a forward view of demand. With a unified demand signal, an organization can make decisions that are properly coordinated which drives improved performance.
While the idea of a unified demand signal is not new, the limitations of existing technology and statistical techniques have created significant hurdles in its implementation. Traditional statistical models were limited to what factors they could consider (e.g. could not consider non-hierarchal product attributes) and needed long sales histories to perform (i.e. could not predict pre-season and new product forecasting at the SKU level). With these limitations, the models built would not always provide an accurate forecast for every use case.
However, with recent advancements in cloud computing, data platforms, machine learning, and artificial intelligence, technology has finally caught up with the vision of a unified demand signal. The ability to quickly process data and train the latest machine learning models at scale allows for a significant improvement in both forecast accuracy and portfolio coverage. We are no longer restricted to forecasting products with long sales histories and stable demand. Modern data platforms and cloud computing allow us to harmonize and leverage disparate data at scale from across the organization (e.g. POS, promotions, hierarchy levels, product attributes, store attributes, new product launches, inventory levels, marketing spend, planogram layout, customer information, etc.). With the continuous inclusion of additional data, we can empower AI models to produce increasingly accurate predictions at the store/SKU level across the entire product portfolio.
With these advancements, retailers can break down silos that exist within their organizations and allow them to operate with increased agility and precision - a concept that was contradictory only a few years ago. They can now realize the benefits of a unified demand signal to improve customer satisfaction, reduce lost sales, lower inventory carrying costs, and have fewer markdowns by reducing forecast error by 10-20%.