Forecasting promotional demand is necessary in order to allocate the correct amount of stock. However, time-series forecasting models are typically not a good fit to address pricing-related demand patterns. More complex machine learning forecasting models are needed to properly take into account past promotions, and to reflect the upcoming impact of those that are planned.
Viral videos of shoppers viciously fighting over cut-price products on Black Friday show the dark side of consumerism. However, today, promotional periods such as these are a key part of the consumer landscape, with some companies making the majority of their profits during these periods.
Yet it can be a double-edged sword, because if promotions are done badly they can reduce the credibility of a company’s products and alienate the customer base into expecting constantly lower prices.
In this episode of LokadTV, we investigate these various impacts of promotions and delve further into why the data behind them is currently so misunderstood.
Forecasting promotions is fundamentally difficult due to their vast variability and the type of vertical they’re associated to. Also, they warp the perceived levels of demand for a product. Many companies are still relying on moving averages on Excel to try and map their promotions, however this isn’t sufficient. Given the recent advances in machine learning technology, it is now possible to create far better statistical models to successfully forecast promotions. Naturally, moving towards a machine learning method is often anything but simple.
We try to explain how these algorithms can predict which promotions actually work and when they work best. We analyse just why enriching data is so important, as it’s not just the price of a product that’s important, but many other vital statistics that should be correctly tracked. Finally, we expand on the processes you should follow to have successful promotional forecasts in your organisation that do not skew your data.
00:50 Promotions are incredibly variable depending on the industry you are in. What types of companies are we talking about today?
02:11 Why are promotions complicating sales? Why is there a difficulty seen with promotions?
03:53 How are companies adjusting these forecasts?
04:47 Why do companies want to change their history?
06:58 Do statistical methods actually work?
08:35 Why are these companies so happy to work with moving averages?
10:25 What sort of data should we be collecting?
13:25 Does this mean that, through your forecasts, you will be able to tell us what promotions to do, when promotions work best, etc. What sort of information would you be able to generate?
17:26 In terms of Machine Learning, how quickly can the machine learn what we are doing here?
20:28 What is the process CEOs should follow if they want to implement promotion forecasting into their organization?