eCommerce Analytics

How did you read the title of this post? If you read “rules” as a verb, you’re probably getting solid additional revenue out of your on-page cross sells and already know that, yes, cross selling does indeed rule. But if you read “rules” as a noun and want to improve the rules behind your cross sells you’ve come to the right place.

Let’s take a look at 5 common techniques retailers use to come up with cross sell rules, from least to most sophisticated. Which group are you in?

Cross Sell Sophistication Levels

Level 1: Hardcoding Popular Products

There’s an area on your site to show related items and, well, you’ve got to fill it up somehow.

Pros: Easy to do, easy to maintain, better than nothing
Cons: Poor performance, can give sites an amateurish feel

Level 2: Popular Products Within a Category

If a shopper is looking at a pair of shoes you show the most popular shoe styles. If you want to get fancy you can filter the cross sells to match the type of shoe the shopper is browsing.

Pros: More relevant than showing site-wide popular items, reasonably easy to do with just a few lines of SQL or by entering manually
Cons: Fair to middling performance, cross sell stagnation since popular items are likely to rule the roost for long periods of time

Level 3: Manual Cross Sell Rules for Each Product

This is the most common way of producing a “you might also like” area. An expert manually enters the cross sell items for each product and maintains that list as new products are added and removed.

Pros: Good relevance, good performance
Cons: A lot of work, difficult to maintain as new products are introduced and old ones phased out, risk of recommending out of stock products

Level 4: Customers Who Bought… / This Item Often Bought With…

By analyzing sales data to see which products are most often bought together you get a truer picture of how your customers think your products go together.

Pros: Great performance, shows shoppers products that have been demonstrated to go together, easy to maintain (if the analysis process is automated)
Cons: Requires SQL ninja skills, all shoppers get the same recommendations regardless of their behavior or history

Level 5: Recommendation Engine

Software or a service that crunches sales, clickstream, and other behavioral data to produce product recommendations that take a shopper’s particular affinities into account. The functionality of an eCommerce recommendation engine is usually provided by a 3rd party (like us!) in all but the biggest retailers.

Pros: Best performance and relevance, generally automated so little maintenance is required, improves over time, provides a good user experience
Cons: Can be expensive, requires setup, some engines have a “cold start” where data must be collected for a few weeks before recommendations start

Zero to 60, Slowly

Improving your cross selling performance doesn’t have to be all or nothing. If you’re a level 1 you don’t have to jump to level 5 right away. Instead, find the right level for you by improving your sophistication one or two levels at a time as part of an ongoing process.

There is no question that recommendation engines work. If you’re looking for a way to boost order value, items per basket and conversions, adding a recommendation engine to your eCommerce site is pretty low-hanging fruit these days. But how do you evaluate which is the right one for you when all that black magic that goes on under the hood is so complex (and expensive)?

Recommendation engines are not rocket science (though we’d have you believe otherwise)

Most every recommendation engine provider boasts of patented algorithms and legions of MIT grads. Hell, our recommendation engine was created by 3 MIT grads (though (Course XV, not Course VI as you might suspect).  What we don’t tell you, however, is that this is more valuable for marketing than for creating a great recommendation engine.

Why?  The recommendation problem has been solved.  Most recommendation engines use one of a handful of methods that are well understood and detailed in academic literature. We all have our own little twists on the procedure, though, and this is what the legions of MIT grads ultimately patent.  The reality is that, at heart, most recommendation engines aren’t that dissimilar.

So when you’re looking to add a recommendation engine to your site, don’t worry so much about the black box powering it all.  Instead, focus on how well it meets your needs in other areas, chiefly:

• Cost
• Ease of setup and integration
• Customizability

These are the areas that cause the most headaches, though they are often overlooked by potential buyers. If you find a recommendation engine that works for you in these three areas, you’ve got a good one.  Now go add it to your site and count all that new revenue!