Site Sleuths

Mattress manufacturer Select Comfort was getting results from its paid search listings on Google and Overture; its selection of paid keyword was driving web site sales. Yet the company wondered: Given its investment, was the paid search campaign as effective as it could be? And were online sales the only way to measure that?

The application of new and more sophisticated web analytics provided the answers to Select Comfort`s questions: no and no. Given the high-consideration nature of a mattress purchase, Select Comfort suspected that after finding the site through search and studying it, many consumers were then using a toll-free number to complete the transaction at the call center. But site managers had no proof. The company decided to deploy a system that would provide dynamically generated toll-free numbers for different keyword listings, allowing it to track the keywords that lead to the most calls.

Not just clicks

Factoring in the offline sales driven by keyword buys overall shot Select Comfort`s ROI on its keyword investment up from 460% to a whopping 1165%, providing a more accurate picture of just what it was getting for what it was spending on paid search. And despite ROI that jumped from the triple into the quadruple digits, the findings suggested Select Comfort could do even better. Analytics showed that 58% of Overture keywords and 35% of Google keywords generated fewer than 10 clicks and no sales in either channel.

Apart from how the individual keywords performed on the different engines, Select Comfort`s cost per order on Google was 47% less than on Overture and 25% less than on MSN`s search engine. As a result, Select Comfort is now reducing spending on more poorly-performing keywords and boosting keyword spending on MSN.

Retailers believe they`re doing well when they see sales add up online, but web analytics that illuminate customer behavior in ever-increasing detail now show that sales numbers are only half the picture. The kind of analytics-driven intelligence that wasn`t available even a few years ago today is capturing where customers arrive at the site from, their pathway on the site, their exit points and where they go after they leave a site--even blending that with data drawn from other sources.

Web analytics` new sophistication highlights missed opportunities and points toward improvements in ways that sales figures alone don`t. Online retailers are finding, as did Select Comfort, that problems may lurk underneath even the seemingly satisfactory performance of a site. Digging deeper into customer data with a powerful new generation of analytics, they`re learning several lessons.

They`re optimizing the wrong pages or links. Web real estate may be infinite, but shopper`s page views are anything but. "Retailers really have a very limited amount of real estate to optimize for buyers," says John Mellor, vice president of marketing at web analytics provide Omniture Inc. In a study it commissioned of Jupiter Research Inc., Omniture found that no matter how many pages a retailer put up for its site, the average number of pages viewed on the site per visitor per month was fewer than 20 for more than 50% of sites studied.

"56% of your traffic interacts with your site through fewer than 20 pages over the course of a month," Mellor says. "With three to five of them being checkout pages, you`re down to about 15 pages that you need to optimize so shoppers make a purchase. It becomes very important to understand which pages on your site are producing the most revenue."

That means going beyond using analytics to determine the number of clicks on page links to using analytics to figure out revenue per click. One of Mellor`s e-retailer clients, for example, placed its gift card offer on the lower third of its home page, until analysis revealed that at an average $14, the gift card link was generating higher revenue per click than anything else on the site. After placing the feature more prominently, the retailer saw overall revenue from the gift card link increase dramatically.

Some retailers who`ve done this analysis have discovered that individual product category pages account for as much as 30% of revenue on the site, Mellor says. If so, those pages should get the lion`s share of testing and optimization resources, he adds. "A 2% to 3% gain in efficiency on that page is worth much more than it is on a page that only participates in 5% of your site`s revenue," he says.

Their keywords are off target. "It`s almost impossible for a retailer to know all the different combinations of keywords that match its business or product with what`s in the customer`s mind," says Olivier Silvestre, senior e-business analyst at web analytics provider WebSideStory Inc. For many e-retailers, analytics have shown just how big that gap can be.

One WebSideStory client provides an international directory of information on spas. It focused much of its keyword program around terms containing the word "spa," but analytics suggested it was missing out by not expanding keyword buys to phrases that described the spa concept without using the word itself.

"People may search for `relaxing vacation ideas.` If a spa happens to come up as an alternative in that search, they may realize that while they didn`t think to search for `spa` in the first place, a spa is what they are looking for. By understanding this, you can fine-tune your keyword strategy and get a lot of customers that way," says Silvestre.

Analytics can also show when a keyword buying strategy has become outdated, and when to move into or out of a broad-match keyword buy, he adds. Under a broad-match buy of the term "car," for example, a retailer`s listing would appear in any search for which a user has typed in any keyword phrase that contains that word. A phrase match buy for "red car" would bring the listing up in any search that contains the phrase with the words in that order, even if part of a larger keyword phrase. An exact match buy would bring the listing up only for searches in which the user types in the phrase "red car."

Broad-match is used effectively in discovery, when the retailer is attempting to determine what combinations of keywords people type in that are most relevant to the retailer`s business, Silvestre says. Once that`s established, the retailer can purchase those keywords and phrases on exact match, a more narrowly defined category that connects more directly with interested buyers.

"That`s where web analytics is important," Silvestre says. "You want a web analytics application that will capture exactly what has been typed in by people to get to your site, not just the keywords you bought."

The conversion funnel leaks. The conversion funnel analogy is that the wide end representing all visitors to the site narrows along a set of progressively more specific behaviors to conversion at the narrow end. But not everyone who makes a purchase enters at the top of the funnel, and not everyone who enters at the top winds up making a purchase at the narrow end. Analytics show that people enter and exit the funnel at multiple points. Seeing not only where these points are, but also where visitors were immediately before and after entering or leaving can inform site design to keep traffic in the funnel and moving toward conversion.

Internet Retailer Top 500 Guide

"The home page isn`t a good overall gauge of how many people are entering the conversion scenario," notes web analytics provider WebTrends/NetIQ Corp.`s Jason Palmer, vice president of product marketing and product management. WebTrends analytics can show, for example, that while 20,000 visitors abandoned the site after hitting the home page, 80,000 of the 100,000 who entered a site at a product page abandoned at that point. "Your biggest opportunity to increase conversions is the 80,000 people who exited at the product page," says Palmer. "And without being able to see where people are going, it`s difficult to understand what to improve, because you are simply making guesses based on the number of people at each step rather than basing it on where they actually went."

One WebTrends customer, for example, found a high abandonment rate at the registry log-in. But analytics suggested the reason wasn`t that people seeing the page were choosing not to register. The problem was that the registration page served a number of purposes, such as providing the ability to check order status and sign up for a newsletter. That means a number of visitors were accessing the page with no intention to purchase, though that page was in the conversion funnel.

"By being able to say how many people exited, and if they did, where they went next, we started to see that online order status was a common exit point for the site," says Palmer. "So it became important to understand what percentage of people hitting that page were not interested in purchasing but in checking order status."

While that helped provide a more accurate picture of conversion, high abandonment rates at other points in the conversion funnel can point the way toward improvements. On one WebTrends client`s site, the top exit point in the conversion funnel was a page that listed what sales taxes the site charged in different states. "It was a dry, non-exciting page that didn`t reinforce the value of purchasing from that particular retailer," Palmer says. "By redesigning those pages to reinforce branding messages and make them easier to read and understand you can work to decrease abandonment off the pages. And then with analytics you can truly see whether you are making a difference, based on being able to see the number of people who are exiting the site off that page."

Bad spelling can cost sites big bucks. Fingerhut suspected that its site search function needed work, but didn`t pinpoint the problem until it implemented web analytics from Coremetrics Inc. Finding and fixing what was wrong netted the company an extra $1 million-plus in revenue, according to Mike Sidders, director of e-commerce at Fingerhut.

The Coremetrics analysis revealed that misspellings, plurals, and other derivatives of product words that shoppers typed into the search box weren`t accounted for in Fingerhut`s site search. Compounding the problem is the fact that site visitors are using site search more than ever.

People used to finding information on the web via search are also using site search to find products at retailers, says John Squire, vice president of product management at Coremetrics. "A large number of clients and prospects don`t have good analytics around site search, such as how much of their on-site sales are attributable to site search," Squire says. "Fingerhut didn`t realize that people were using site search as much as they were."

For example, the retailer had several products associated with the word "pool." While the word "pool," typed in exactly, would bring up those products, "pools" and related searches brought up nothing. As a result, Fingerhut was missing potentially valuable traffic that was looking for products using search terms they were accustomed to using and not just those recognized by Fingerhut`s site search engine. "No products were showing up in the search, though Fingerhut had a large number of products that would be relevant. Being able to see null sets and low results sets for specific high traffic keywords pointed them in the direction of where they should fix their search engine," Squire says.

Some campaigns aren`t as profitable as they look. Marketers typically gauge the success of online campaigns by looking at traffic or revenue generated, but profitability provides the most accurate picture of a campaign`s value. By going beyond the basics to incorporate data on campaign performance from external sources--adding external pay-per-click data so as to calculate marketing costs, for instance--marketers using WebTrends analytics have experienced a few surprises in discovering which individual campaigns are really the most profitable, says Palmer.

For example, a comparison of online campaigns A and B might show that B received more click-throughs but got fewer conversions and therefore produced less revenue than A, says Palmer. So is campaign A the better performer? Not necessarily.

Adding in the marketing costs could show that campaign A is costing 30 cents per click, while B costs 11 cents. "So even though B has a lower conversion rate, it`s about a third the cost of A and the marketing ROI percentage is almost twice as high," Palmer explains. "It`s an example of how you can`t just look at revenue. You need to be able to incorporate other costs to truly understand where you should invest your dollars."

Powering up

Powered by the increasing capacity of web analytics , some retailers are taking analysis a step further, incorporating product costs into campaign analysis. That allows them to see that the product they are selling on a data feed through one portal partner, for example, is much less profitable than through another portal partner. "You start to be able to gauge how one campaign or channel performs against another," Palmer says.

And it`s not just online retail`s bigger players that use web analytics at this level of detail to uncover and address their site`s hidden problems. Different pricing models are reducing the initial cost of entry. At WebTrends, for example, the baseline cost of enterprise software is about $10,000, but marketers must pay an annual maintenance fee, have the necessary hardware in place, and assign internal resources for development. Hosted service starts at around $15,000 per year. Palmer says the cost difference between the two models evens out after about 15 months.

"You`d expect that the smaller retailers wouldn`t be able to do this, but it varies by organization," he adds. "You`ve got large retailers with the staff to apply to this level of detail, but what`s interesting is the number of smaller retailers who do it because they are so resource-constrained. They really drill down into the cost of the performance of the campaigns they invest in."

mary@verticalwebmedia.com