Following my recent blog post on how difficult demand forecasting for promotions is, there is also another key challenge for retailers where new product introductions and heavily refreshed seasonal assortments account for the bulk of sales – particularly during Black Friday and Christmas.
From supermarkets and general merchandise chains, to fashion retailers and department stores, most retailers have to manage hundreds, if not millions of product introductions. Of course, when a retailer starts looking to manage products at individual store level they effectively multiply one factor by the other, in terms of stock keeping units (SKUs) – and the quantities of data they are processing grows exponentially.
So How Does One Create Meaningful Forecasts for New Products When There Is No Sales History at All?
To help them overcome this, one option is to use reference products as a basis for forecasting – a new product’s forecast can then be based on a similar product that has a sales history. Automation can also make a significant difference when handling large numbers of SKUs. References can also be selected quickly based on defined rules, or forecasting could be based on product attributes such as colour, brand, size or genre. It also helps a lot if there is any sales history available prior to the season that can be then used as a baseline, and the seasonal pattern can be adapted from items in the same product group.
One option is to use reference products as a basis for forecasting.
If we first look at the starting point of automating new product introductions, we begin with identification. For this, a database search for products with an introduction date in the near future can be conducted, or if that option is not available, then a simple search for any active products with no sales at all.
Next, it’s important to reference products so that references can be automatically assigned to new introductions. For example, where a product is introduced that’s tagged ‘candle’, ‘red’ and ‘large’ the system might search for existing products with those tags. These reference products are then used to provide sample data, and are thus a reasonable basis from which to calculate a forecast for those new products, that as yet, have no sales.
These reference products are then used to provide sample data, and are thus a reasonable basis from which to calculate a forecast.
In practice, different retailers use different sets of criteria to map the best possible reference product for each new introduction. Most commonly though the search starts by identifying the level in the product hierarchy where there are enough candidates for a reference product. The search can then be narrowed by further product data, quite often, for example, by price. Colours are also often used when searching for reference products. If products have a distinct demand pattern at the beginning of their lifecycle then it is also possible to duplicate across the launch profile from the reference product, by using its launch-period sales figures.
So by implementing a process where forecasting for new items is carried out automatically, retailers are able to crack the new product forecasting conundrum. It can be done by searching the historical assortment for items with similarities to those being introduced and building a forecast from that. The results – availability can increase several percentage points to near perfect and, simultaneously, inventory values and markdowns reduced considerably.
So by implementing a process where forecasting for new items is carried out automatically, retailers are able to crack the new product forecasting conundrum.
Have a look at how product introductions are automated using RELEX Business Rules Engine in our whitepaper: Everything you wanted to know about the RELEX Business Rules Engine.
