00:30 Can we forecast for new products?
01:55 Why the “time series approach” does not work?
03:41 Can you actually forecast for something that is completely new?
07:16 With the advances in Deep Learning technology, is there anything that can be applied to look at these attributes in more detail?
11:41 How can we have confidence in these new product forecasts?
14:23 When we produce our forecasts, we are looking at business decisions across a whole catalogue. Will new products not skew our whole forecasts?
18:02 Is there any way of producing multiple forecasts to work out what that sensitivity of price will be?
21:50 What can we see in the near future in terms of technology advancements for forecasting new products?
New products do not have a sales history that can be represented as a time-series. As a result, time-series forecasting models don’t work for new products. Forecasting demand for new products requires alternative forecasting models capable of leveraging data, such as product attributes, that do not come as a time-series.
Due to the amount of unknowns and the lack of data that can exist, forecasting product launches is inherently difficult. However, it is a vital part of the product lifecycle and must be carefully planned in order to capitalise upon the spike in demand that often follows a new product release.
In this episode of Lokad TV, we discover that forecasting new products is possible but explain in detail why traditional methods, like Time Series Forecasts and moving averages, will simply not work. We illustrate the Attributes Approach, something that isn’t typically utilized by most forecasting software programs, which is a method that consists of leveraging a series of product attributes such as size, colour or previous launches.
This can help you to fill in the unknowns with your forecasts by linking your new products with older products to better predict trends, as there will usually always be similar characteristics shared between products that can be analysed to produce reliable statistical models for future demand.
We then explore in more depth just what the various advances in Deep Learning technology can bring to reduce the levels of uncertainty and debate if we can have any confidence in the results of these forecasts.
One of the reasons companies can have high stock levels is due to launching a new product that was unsuccessful. This results in poor cashflow and drains valuable resources. We attempt to understand the impact of product cannibalisation and how new products can in fact skew the results of whole entire forecasts.
Finally to finish up, we discuss the future impact of more sophisticated forecasting techniques, such as those that utilise images and product descriptions to group the attributes of various products.