The CIO Web Transaction Diet: Shedding

How many webpages can you serve for a penny?

Do you know? Do you care? You should. While your e-commerce operation may not generate the traffic of an Amazon or an eBay, that doesn’t mean you shouldn’t be concerned about how much each visitor costs you.

Because, almost invisibly, those costs may be bleeding you.

In the early days of the Internet boom, “stickiness” was the goal. Drive the masses to your URL, trap them with the digital flypaper of your clever applets and riveting content?and worry later about how you were going to make money off them.

Now it is later.

For certain, the Web has cut a lot of fat from the cost of dealing with customers. Self-service online apps reduce the need for customer service reps. Online sales cut the requirement for expensive storefronts and pricey human help. But companies have already extracted all the easy dollars; the bottom-line battlefield is beginning to move to the pennies.

Spurred by a desire to find any and every way possible to add to the bottom line, many companies have begun to do the math and are making changes designed to decrease the costs they incur per visitor (call them transaction costs) or drive up the average revenue earned from every person who hits their site. From tweaking the user interfaces in an effort to help visitors buy more, faster, to testing ways of filtering out the window-shoppers, to fully reconstructing the back-end infrastructure, smart e-commerce execs are finding ways to wring every extra penny out of what was often a slapdash operation put in place during the boom-boom ’90s, but which now must justify its existence based on profitability, not pizzazz.

The savings are out there.

How you find them depends on the particular challenge your business is facing.

Challenge: Window-Shoppers

Sabre Holdings is the $2 billion corporation behind such travel services as Travelocity, the Sabre Travel Network and GetThere. Few industries have been changed by the Web as much as travel.

“The basic shift is from expert agents?professional agents?to consumers or intuitive users,” says Craig Murphy, CTO at Sabre Holdings. “An expert agent wants targeted information; a consumer wants more options. More options take more data processing.”

As consumers demand more features?everything from search tools to weather reports?they are also more likely to spend time browsing and dreaming about a trip than they are actually buying tickets. Consequently, the “look to book” ratio is going up, Murphy says.

Since keeping costs low is key, Sabre last year began to shift from HP-UX Unix platforms to Red Hat Enterprise Linux AS on Intel running the open-source MySQL database.

The reasoning behind this change is straightforward. With proprietary systems, every time Sabre needed to add a server because of increased load on the site, it had to pay significant incremental hardware and software costs. But with Linux and MySQL, Sabre can duplicate a standard software configuration onto a low-cost Intel-based computer for a fraction of the price. So the daydreaming jet-setters can keeping dreaming, without breaking Sabre’s bank.

Solution: Use open-source software to reduce overall computing costs.

Challenge: Business Activity Peaks and Valleys

For H&R Block, keeping costs under control means finding a balance between having the horsepower to meet peaks in business while not sitting on millions of dollars of idle hardware during lulls (which is an issue for any business with fluctuating activity, whether your cycles encompass seasons, quarters or even certain days of the week). Block found its answer not in cheaper software and hardware, but in letting someone else deal with the problem.

Block has what may be the ultimate seasonal business, with the online peak coming Jan. 1 through April 15. (In 2003, more than 2.1 million U.S. taxpayers used Block’s online tax services. Block CEO Mark Ernst publicly stated early this year that the company expects to serve 20 percent to 40 percent more online clients in 2004.) Tracking the cost of every customer requires modeling every change to the systems not only on the individual level, through user interface research, but also in the aggregate, using scalability testing tools from the likes of Empirix.

And to keep from owning a data center that gathers dust from May to December, Block found a hosting provider four years ago (the company won’t say who) that was willing to ramp up computing power for the tax season, and then unplug CPUs?and not charge Block for them?during the quiet time.

“We scale up for our peaks, and we scale back down for our normal business. And then during the off-season we go to a very minimal configuration. Our cost is variable along with that,” says Roger Zaremba, vice president for enterprise technology at Block. “We’ve crafted what we think is a very competitive solution.”

Block has its eye on the future as well. “Working with various folks like HP, we’re trying to architect all of our systems to provide a technology-on-demand scenario,” Zaremba says. The ideal situation, he adds, will be a time when Block can turn on a service, then turn it off, “and truly be charged for just what we use.”

Solution: Find a provider that will turn off the meter when your business is idling.

Challenge: Different Customers with Different Demands

Recognizing that different types of customers use Rapsheets.com for various purposes, the company has developed a sophisticated infrastructure to avoid wasting expensive processing time on simple tasks. Rapsheets owns the largest privately held database of criminal records in the United States (160 million records covering about 95 percent of the U.S. population) and allows companies and individuals to search those records for, say, the nanny who has a couple felonies that didn’t show up on her rŽsumŽ or the wannabe security guard with outstanding warrants in three states.

Given increasing concern about liability and safety, such background checks are becoming the norm (the service currently has between 10,000 and 15,000 companies as customers, including national contracts with Little League Baseball, the Salvation Army and YMCA), and Rapsheets has watched its transaction volume increase by an average of 20 percent per month during the past year.

The company started in 1997 with a fairly traditional approach to scaling its applications. As transactions increased, the company bought bigger boxes and more CPUs. Unfortunately, it became apparent that the expansion couldn’t continue. The combination of rising hardware and software costs had reached a desperate point by November 2002. The company also found itself sitting on mountains of unused computing power during slow stretches, as customers tend to submit requests primarily in the middle of the week and do multiple searches on Saturdays when they want to run through a list of potential new hires at one time. “We were paying for a lot of depth of performance in off-peak times,” CTO Keith Grimes recalls.

So that month, the company decided to take another look at the situation and seek a long-term solution for keeping transaction costs in check. The result was a complete overhaul. “We rewrote applications and rearchitected our hardware,” Grimes says.

So-called “deep searches,” where Rapsheets looks through records of multiple jurisdictions and even checks for intentionally or accidentally misspelled names or transposed numbers, require more computing power. But Grimes says 60 percent to 70 percent of searches require less exhaustive techniques?perhaps nothing more than matching a name and date of birth to a record. Previously, however, both types of searches ran on the same powerful?and expensive?hardware.

Now that has changed. Custom-built middleware determines on the fly the type of search required and sends the search to systems tuned for the task. Deep searches still go to the more potent multi-CPU boxes. But the middleware layer sends lower priority and simpler requests to low-cost IBM blade servers. And Grimes didn’t stop there. Each blade can run a different combination of operating system and application, allowing him to tune performance?and reduce costs?even further. Some searches run best on Linux, others on Windows. And with blades, Grimes can add new horsepower?or reconfigure the old?in hours instead of days to deal with new product offerings or new customer requirements.

The middleware also lets Grimes add new resources regardless of their platform?including Web services. “I could literally plug in a Unix box, a Sun box or a big IBM AIX box that will handle a hundred times what I’m doing now, and the middleware wouldn’t know the difference,” he says. While one search might use 10 different systems, the user sees the same results. But Grimes?and his bottom line?will certainly notice the difference.

Solution: Install middleware to route less demanding requests to cheaper blade servers.

Challenge: Increasing Revenue Per User

Some people would argue that companies should worry less about saving pennies on the back end when there are still dollars to be extracted from the front. “I’m having a hard time believing that infrastructure costs are where the biggest improvement [in reducing transaction costs] is going to come from,” says Bob Chatham, principal analyst at Forrester Research. “I’m looking at [the cost issue] more from effectiveness and quality of customers’ experience.”

And many companies are constantly looking for ways to do just that.

Wells Fargo’s Wholesale Internet Solutions group, for instance, is certainly investigating infrastructure cost-savers such as blade servers. But according to Executive Vice President Danny Peltz, such savings are secondary to finding ways to increase revenue per user on the front end. To do that, his group is constantly looking for ways to make the Web interface used by commercial customers easier to navigate, as well as searching for new services to add. But generating more revenue per customer isn’t as simple as rearranging menus and adding features. Where some companies would love to put up a website and immediately get rid of human contact points to save costs, Peltz’s group actually takes the time to train customers in how to use the site most effectively.

Training can take many forms, from online tutorials to classrooms to one-on-one sessions with key customers. Wells Fargo is convinced that helping customer companies use site services goes a long way toward keeping them happy?and increases the likelihood that they’ll take advantage of (and buy) more services.

“It’s not really about cost reduction,” Peltz says. “It’s about the ability to scale without adding costs.” To that end, the group has constantly tweaked the site (Peltz claims 14 major upgrades since the middle of 2000).

Self-service tools have been a big part of the changes, allowing customers to modify such things as their policies and access rules without intervention from bank employees. For example, a treasurer could set limits on wire transfers for various clerks, then easily change the limits as needed without help from anyone at Wells Fargo. The approach seems to be working. According to Peltz, the Wholesale Internet Solutions group has grown from serving 2,200 customers at the end of 2000 to more than 22,000 at the end of 2003?all without making significant staff additions and their related costs to his group.

Solution: Train your customers to use your site most effectively.

Challenge: Reducing Number of Transactions Onsite

Peltz’s group deals with large customers spending large dollars with Wells Fargo, so training is a cost-effective option. For other sites, however, simpler, cheaper solutions are in order.

At H&R Block, for instance, the company can’t afford to train each of the more than 2 million users who show up during tax season. Instead, it concentrates on usability testing to make sure tax-filers get what they want (and only what they want), and don’t spend time using computing resources they don’t need.

The company allows users to complete a tax return online, for instance, and doesn’t require them to pay a fee until they actually submit the return. Designed badly, the interview process that generates the return could end up running millions of calculations for thousands of users who never give Block a dime. “We’ve done a lot to optimize our interview [in order to reduce those costs],” says Neal Shaw, divisional information officer for Block’s e-solutions group. He notes that the company tries to identify early on if there are sections of the tax interview a user can avoid (perhaps because she doesn’t have investment income or business activities), thereby saving steps and reducing Block’s overall cost to complete the return. In one case, testing discovered that users didn’t want to see running subtotals as they filled out the return. Block cut the number of times subtotals appeared, increasing user satisfaction while reducing processing time?and costs?on the back end.

Solution: Test your site to find out what your users want and need.

Challenge: Deadbeats

But what about those customers you don’t want at all? Block’s site is full of calculators, educational content and tax information?all available whether you use it for your returns or not. Shaw admits that such features attract some lurkers who never plan to buy, but “that’s the price of doing business,” he says. “We look to that as being the kind of content that helps us offer the most value to our customers.”

Amazon.com feels the same way and doesn’t see stickiness as a problem. “We’ve always sought to be the place where customers can find and discover anything they might want to buy online,” says Amazon Vice President for Technology Infrastructure Tom Killalea. To that end, the company has added numerous processing-intensive features, including online audio samples and a “Search Inside the Book” section, which lets customers search the full text of 120,000 of the tomes Amazon sells. “Increased stickiness has been a happy byproduct of this sort of innovation, with discovery being the main goal,” Killalea says.

But not every site sees it that way. While stickiness can assist a general-consumer retail operation, it may not be an ideal situation for companies seeking a more restricted clientele. “Who are you getting into your website?” asks Jakob Nielsen, a principal at product development consultancy Nielsen Norman Group. “Are you getting motivated customers or the riffraff? It is becoming clear that sheer numbers of unique visitors are detrimental.”

One of the more recent attempts by many e-commerce sites to separate the paying wheat from the deadbeat chaff is to force visitors to register, providing some personal information in return for being able to order from or even view content on a website. But experts say companies must be extremely careful with how they use such strategies or they will end up chasing good customers away as well.

Companies think, “Oh, we’re going to do this fancy marketing” with registration forms, says Nielsen. But using such forms increases the possibility that a customer may never actually use the site. “There’s the risk of browser crashing, of the modem going down, that the user doesn’t understand the question. It gives customers one more page to think about whether [they] should do this or not.”

Salesmen, Nielsen points out, would never do this. At the moment when someone is about to sign on the dotted line, they don’t say, “Oh, by the way here are five other things to think about,” Nielsen says.

Instead, companies wanting to keep unprofitable users off their sites should look for other solutions. Nielsen suggests using search engines as a first line of defense. “If you’re advertising on search engines,” he says, “screen out the keyword free.” Doing so will keep many of the cheapskates from dropping by and wasting your time. Similarly, he says, put the price of your products in the ad or meta tags used by the search engine. If your products cost $10,000, you don’t need to waste time catering to users looking for something under 20 bucks.

Solution: Use search engines as a first line of defense.

First Steps

Which way to reduce costs and increase profitability will work for you? It depends on your needs?and budget. Full-scale infrastructure changes are expensive and can be disruptive, even if you know the ROI will eventually be realized. But don’t think such shifts are just for big companies. Sabre’s Murphy says that most companies could benefit by a change similar to his company’s. “High volumes make the payoff bigger,” he agrees. “But the move to open source will provide benefits across all of computing.”

If you’re not ready for that kind of leap, smaller user interface tweaks are an easy way to get started. “Probably the topmost [goal] would be to develop a mind-set of experimentation,” says Forrester’s Chatham, adding that companies must quantitatively test customer experience and not be afraid to acknowledge that the Web isn’t always the best means of customer contact.

The key is to do something, not nothing. Web usage will only increase in the coming years, and early-mover companies are learning facts about customer costs now that will ultimately benefit them when all the easy dollars have been pulled from the system. Web services, for instance, is just around the corner, with the potential to drive a huge volume of interaction on systems that once dealt with only the meandering clicks of human users. Whether that volume is an opportunity or an onslaught depends on how well you’ve prepared.