Using Next Best Product Recommendations to Determine Inventory Levels

A couple of weeks back, I gave you a set of three hyperlinks that describe efforts to predict the next product that each of your customers will buy. Many companies that initially sprung up around this field have taken Amazon or Netflix and their collaborative filtering approach as the method for predicting what customers will like next. More strikingly, recommendation engine companies have patterned their offerings after Amazon and Netflix as well. That is, the output pieces are built around website infrastructure, meaning that the recommendations live online and are tailored specifically for customer visits. Alas, there are more than a few businesses that still make a majority of their money offline and need a solution that fits the non-electronic areas of their business. What about the merchandising arm that just wants to know how to determine the quantity of each product to buy from the manufacturer or wholesaler for the coming quarter?

How can we use this information elsewhere? That’s the question I get from Marketing when we deliver next best product recommendations to them. Perhaps, it’s just the economy but information reuse is more rampant now than I’ve ever seen (as in: “I’ve just paid you to run this through some fancy predictive models and now I need to wring every last drop of use out of this data that you’ve given me.”) The first thing I tell them is, if they’re not already using inventory prediction models of some sort, that our models will give them a good idea of how much inventory they should consider stocking in each of the product areas.

At the very least, our models will give retailers an understanding of how many current customers are interested in the wares you’re considering buying from manufacturers. When you add the output from these models to an expert procurer’s intuition about which products are - and will be - hot, then you have a good decision-making duo. Having these predictive cues as input is even more important in a bad economy where the problem of distressed inventory is both more likely to occur (because of depressed buying) and more difficult to remedy (because severe price sensitivity will necessitate even more severe discounting).

One idea whose time is quickly arriving is the idea of building predictive models for product attributes. For example, a pair of jeans might be considered “fashion forward” or “painted on”. And you might have 10 such pairs of jeans in each of these categories, as opposed to 20 pairs that are “classic” or 12 more that are “high comfort.” By using these attributes, procurement can get a scientific perspective as to what the company itself is known to sell by the general public, and, more importantly, by the segment that shops there.

Whatever the approach - and I do understand that attribute-based recommendations will have a difficult time gaining momentum because of the time needed to create and implement these attribute categories - using next best product recommendations are a useful tool to build on when your company needs to cut down the error when stocking up.

Tags: Amazon, collaborative filtering, inventory, Netflix, next best product, product attributes

Add Personalized Product Recommendations to Your eCommerce Site

Istobe is a powerful recommendation engine that makes it easy to add recommendations to your eCommerce site. How powerful? Shoppers who click on Istobe recommendations spend 20-50% more than the average visitor.

This entry was posted on Monday, January 26th, 2009 at 7:20 pm and is filed under Merchandising, Predictive Models. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.