4 Recommendations for Recommendation Engines

I just got finished perusing ReadWriteWeb’s series on recommendation engines since, as a recommendation engine company, we have a keen interest in the subject. One thing that is abundantly clear is that recommendation engines still reside in the realm of technology, as opposed to business. Despite all of the success of Amazon and Netflix in using their recommendation engines to drive revenue and customer satisfaction (both companies are in the top 40 in customer satisfaction), the idea of recommendation engines still hasn’t quite caught on. One of the reasons is that old channels die hard and recommendation engines are still perceived as a purely ecommerce play. But they don’t have to be. And following these four suggestions will make recommendation engines more palatable to multi-channel retailers who need to take more time migrating online.

1. Make it work on data they already have

   This is my biggest pet peeve. Why don’t recommendation engines work on data that retailers already have? I was listening to a Coremetrics webinar about four months ago on personalized product recommendations and Sucharita Mulpuru from Forrester was saying that it can take a while for recommendation engines to collect enough data to turn out good recommendations. I’m thinking, “Why the hell is this?” After all, so-and-so retailer likely has several years of sales data sitting around. Why doesn’t the recommendation engine just get started on that? Well, the truth is that this is pretty common. Many recommendation engines don’t look at any of the sales data a company has. Nope, the recommendation engine just assumes that time 0 is the day that it’s installed. Seems like an excellent way to kill any instant-ROI arguments that might otherwise have been made. And all that “old” data - all that purchasing behavior - just goes to waste. All recommendation engines should be able to work from basic sales data. In fact, your client shouldn’t even really need to do any “setup” to see benefit from that. That is, they shouldn’t have to tediously tag pages for days on end before waiting two months to see an inkling of a glimmer of monetary return.

2. Make it cost-effective with simple pricing

   This speaks for itself. Everyone loves revenue sharing until the terms of sharing are discussed. Then, whatever hint of ardor is still left after term talk, is totally extinguished when it comes time to split up the pie. Revenue sharing isn’t the way for a recommendation engine company to show that it’s willing to put skin in the game in order to prove that its technology works. Cutting the price until such time it is proved to work is. And then just price the recommendation engine per 100,000 customers/mo. or something like that. That lets everybody do simple math and move on to actually assess the efficacy of the product without worrying about whether or not admitting to efficacy - or inefficacy - is going to cost them.

3. Make recommendations a part of a marketer’s decision-making not the marketer’s decision-making

   Sometimes, a product can just be too intrusive or too ahead of its time. I’m thinking that most recommendation engines fit into this category. They assume the height of sophistication when it comes to their customers’ web architecture. And the truth is that many retailers have other channels that they have to keep an eye on as well. It’s not that multi-channel retailers don’t care about their websites. Just the opposite. They have high hopes for their websites. But they can’t simply ignore other channels such as catalogs and brick and mortar that, taken together, may still source most of their revenue. And that’s not to mention direct marketing sources. In these other channels, the automated decision-making of a recommendation engine isn’t so easy to stomach. When it comes to these channels, recommendation engines shouldn’t be so closed-loop; they should provide recommendations that can be ported beyond the web. Recommendation engine companies should realize that ecommerce is still tiny compared to…well…commerce.

4. Be part of the procurement puzzle solution

   This closed-loop piece also puts me in mind of the procurement puzzle that is virtually left out when an engine begins pushing product. What happens to those product dogs that merchandisers said were going to be hot but which never even approached warm? Now our poor retailer is carrying inventory that it can’t sell, paying to store it while watching the potential price erode ever further. The recommendation engine in a closed loop isn’t going to help solve this problem. The procurement puzzle is better solved by a more simple recommendation engine where the recommendations can answer a question like, “Who is going to buy all of this overstock?”

Tags: Coremetrics, Forrester, product recommendations, Recommendation Engines, sales data

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 Tuesday, February 17th, 2009 at 9:56 pm and is filed under Recommendation Engines. 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.