November 18, 2009
Case Study

Revamped Recommendations Lift Order Value 15%: 5 Steps to More Relevant Suggestions

SUMMARY: Ecommerce marketers need to optimize their product recommendations strategy. Simply offering suggestions isn’t enough to lift order value if those recommendations aren’t personalized and relevant for each customer.

See how an online wine retailer built increased relevance into their product recommendations by considering users’ browsing and buying habits as well as logistical considerations, such as geographic region. Today, 10% of the site’s sales come from these recommendations, and the average value of those orders is 15% higher.

Amy Kennedy, VP, Marketing,, wanted to bring a key element of the in-store wine buying experience to their ecommerce customers. She and her team knew competes with neighborhood wine shops, which have one distinct advantage over online retailers: on-site wine experts.

Kennedy’s team wanted to offer product recommendations, just like an expert wine clerk. The website already offered product recommendations, but they were not as relevant as they needed to be, she says.

"We had some recommendations that we had powered, but the data that fueled those was limited."

They needed their site to make better recommendations to attract customers who would otherwise visit a local wine shop for more personalized service.


The team revamped’s product recommendation system to make more relevant suggestions.

Here are the five steps they took to test and integrate the system.

Step #1. Assess current system

The team’s product recommendations were on the right hand side of the product details page for each product. The recommendations were manually set for each product based on:
o Top sellers
o Wine region
o Wine type
o Price

The team felt this system had several drawbacks. First, the recommendations were not customized for each visitor. Also, the system relied heavily on best sellers, making less popular -- but perhaps more relevant -- products less visible to customers.

Step #2. Outline new system

The team worked on a new system that made recommendations based on two pools of information:
o Customer data
o Product data

The system analyzed customer browsing and purchasing behavior to find trends in related products. Additionally, it examined each product individually, and compared it to other products they were commonly paired with in purchases and in browsing.

The system made relevant suggestions in real time as customers browsed the site. The types of recommendations included:
o Customers who viewed this item also viewed...
o Customers who bought this item also bought...
o Products ultimately purchased by customers who viewed this item...
o Top sellers in this price range...
o Customers who searched for "xxxx" ultimately bought...
o Items related to search "xxxx"...

The product suggestions display included:
o Image of the wine’s label
o Title
o Price
o Wine rating
o Add to cart button
o Link to product page

As customers clicked on recommended products, the system identified which suggestions worked best to increase order values and adjusted later recommendations accordingly.

Step #3. Integrate with regional availability

Selling wine online is subject to legal restrictions. Some states, such as Utah, do not allow wine to be shipped to residents, which is why prompts arriving users to enter the state to which they are shipping. Consumers visiting from these states are shown only non-alcoholic products.

Also, the team sells different products to different states, based on product inventory. They operate different warehouses for different regions of the country, and therefore might offer a 2004 vintage of a particular wine in New York and sell its 2005 vintage in California.

These two issues complicated the automated product suggestion system. The team had to ensure that the system would not recommend products that were unavailable in a customer’s region. To do that, they set up separate data feeds for each state.

Step #4. Test on high-impact pages

The team planned to add the new recommendation system across the site, but did not want to immediately launch full scale. Instead, they enabled recommendations on two high-impact pages in May and watched the results.

"We didn’t want to put this everywhere in the event that something catastrophic happened," such as crashing pages, slower load times or declining average order values, Kennedy says.

The areas they tested (see creative samples below):

- Search results pages

Four suggestions were featured above the search results, and four suggestions were made on the lower left of the results. The top four suggestions also featured the percentage of customers who ultimately bought the suggested items.

- Product details pages

Four suggestions were made on the upper right of the page, and four suggestions were made on the lower left. The team monitored metrics around the suggestions for one month.

Step #5. Expand to other areas

After the first month,"we saw a lift in average order value as well as units," Kennedy says.

The team decided that the new suggestions were working, and expanded them across the site.

Having already built and integrated the system, adding it to additional pages was almost turnkey, Kennedy says. They added suggestions to almost every page where visitors browse wine and gifts, including:
o Homepage
o Product category pages
o Wine Shop browsing area
o Gift Center

The suggestions were not added to’s community, blog, corporate gifts or wine club pages.

Kennedy and the team were happy to see that customers used the new suggestions when making purchases from the site.

"Baskets were bigger in terms of units and dollars, which is what we wanted to see, [and] which was great," she says.

- 10% of’s sales are now made through recommendations.

- Orders that include recommended products have a 15% higher average order value than the site’s overall average.

- In September, conversions on recommendations were 52% higher than the site’s overall average.

Next, the team plans to expand its recommendations into its email marketing.

"We need to dig in to learn more about what else we can take away," Kennedy says. "We’ve just scratched the surface."

Useful links related to this article

Creative Samples from’s product recommendation project

Recommendations Lift Site Conversions: Test Results

RichRelevance - powered the team’s recommendation engine

Improve Your Marketing

Join our thousands of weekly case study readers.

Enter your email below to receive MarketingSherpa news, updates, and promotions:

Note: Already a subscriber? Want to add a subscription?
Click Here to Manage Subscriptions