July 12, 2004
Should a content site pay search engines for traffic? If you're not selling directly online, how do you set a search budget anyway?
Over the past 18 months Edmunds.com has been testing paid search with Overture and Google. Discover how they set their budget, decided on staffing, tested landing pages, and measured results.
Turns out traffic from SEO, paid search, and direct-to-site visitors is NOT equal in value. More:
Edmunds.com, like most content sites, stayed away from paid search because it was almost impossible to justify whether the resulting traffic would be worth the money.
After all, they weren't an eretailer selling directly. They were an information publisher monetizing traffic in a wide variety of ways from generating sales leads for car dealerships and insurance brokers to selling CPM ads to major automotive brands.
Back in 2001-2002, they invested in overall optimization work to have the site show up well in organic listings (those which are not the paid ads.) As a result, Edmunds.com pages showed up just under 10 million times per month in search results pages.
While 10 million may sound like a lot, VP Marketing Seth Berkowitz was increasingly frustrated.
"The way SEO works wasn't conducive to monetization. The critical make/model pages, such as GMC Yukon, where 40-50% of our income comes from didn't rank very well. We weren't even in the top 30 results. For example, if you were searching for information about Honda Accord, SEO dropped you into our article about Honda Accord instead of our pricing funnel."
While fielding sales pitches from account reps from Overture and Google, Berkowitz also learned the total search universe of consumers seeking car-related information online was as much as five or six times the amount of traffic he was getting.
So, there was enormous room to grow if only he could get up paid listings.
Before testing paid search ads, the marketing and site team first invested in a sophisticated site metrics back-end. Reviewing basic Web logs would no longer be enough. They needed to be able to track and value traffic by source in a highly specific manner, including:
a. Which search term a visitor came from
b. Cost per click
c. Dollar value per visitor based on number and types of
pages viewed, CPM ad clicks, sales leads generated for
site partners, and likelihood of returning to the site
d. Segmented group results for time of day, day of
month, overall type of source, overall value
Once the new metrics software was installed and working well, the team carefully moved into paid search tests starting in February 2003.
Step #1. Test a few high-value paid search ads
Berkowitz decided to test search the same way he'd test postal mailing lists for a new direct mail campaign. He began at the most potentially valuable selects, and then slowly widened the scope from there to include more and more segments until the campaign hit the edges of profitability.
In this case, he chose the 50 most-valuable Edmunds.com pages -- all of which were make/model pricing start pages. Consumers searching for those pages were most likely to be actively researching an intended car purchase, and thus red-hot sales leads for Edmund.com's partners and advertisers.
Berkowitz included common permutations on the 50 make/models, such as typos and search terms. So the resulting campaign ended up being about 1,000 search ads in total, all targeting the 50 pages.
He invested $70,000 the first month, watched results like a hawk and then carefully expanded and optimized the list of terms over the next 18 months.
To keep costs under control, he gave his search team a specific guideline of what each search term was worth to the site, telling them to try for a cost per click within 10% over or under that price. He also specifically did not require the top ranking (which can often cost more than it's worth.) As long as the site showed up in the top three listings, that was fine.
Step #2. Hire experts to run the ongoing campaign
With a rapidly growing number of search terms to manage, Berkowitz knew handling paid search couldn't be anyone's part-time job. He got cost estimates from leading search campaign management firms and also researched salaries for highly qualified in-house staff.
The in-house versus outsourced decision was based strictly on math. Berkowitz says the quotes he received from outsourced firms equaled 10-15% of resulting site income. For the amount of paid search Edmunds.com was planning to do, it would be cheaper to take the project in-house.
This doesn't mean in-house staff were cheap. "We went out and aggressively headhunted for the right people. I wouldn't compromise in this area. It takes time to train someone for this and it's core to our business. You have to be prepared to pay for experience. Unless you can afford to pay for strong staff, outsource it."
Step #3. Test landing page variations to maximize income
Aside from testing specific terms, Berkowitz's team also tested varying landing pages to see which site pages would result in more overall income.
Typically, visitors who come to the home page of their own accord move through a fairly predictable path as they navigate the site. First they pick a make/model, then pick the sort of trim they'd like their car to have, and then after reviewing prices, pick a dealer.
The search team tested landing search clicks in a variety of places along that path. Would they prefer to pick their trim and then click to see a price? Would they prefer to see dealer info immediately? Would they prefer more general information?
"It's had a multi-million dollar impact on our business. We probably would have gotten into paid search earlier had we known that," says Berkowitz.
"We now find ourselves in the top three of automotive advertisers online with more than 50 million monthly impressions together on Overture and Google. Paid search does have a ceiling, and it took us about a year to get to full maturity."
The site's monthly paid search budget is now roughly around $400,000-$500,000, not including staffing.
More results data:
-> Visitors clearly have a different value to the site depending on how they discover it. "The highest performing bucket is paid search; but the problem is paid search visits are episodic and less likely to return to the site."
Visitors who bookmark the site or type in the URL directly are slightly less valuable on their initial visit, but they are more likely to visit frequently.
The least valuable visitors are produced by SEO. "SEO isn't all that directed; we can't pick the URL we land you in." So according to Berkowitz's estimates visitors who find the site from organic search are only 55-60% as valuable as those who find the site from paid search.
-> Although Edmunds.com's traffic from most search terms is fairly steady, they've discovered value per click can dramatically shift depending on time of year and day of month.
"Traffic doesn't move that much, but quality moves around. The return on investment is completely different depending on what's going on in the automotive marketplace. In May there were very strong incentives, the market responded to that. In June incentives were withdrawn and interest rates went up. People were still looking at cars on our site, but they don't execute the same way."
Also, days at the end of the month, when most dealerships tend to run the heaviest TV advertising promoting specials, tend to result in Edmunds.com traffic that's ready to purchase.
-> Overture & Google are less and less different. "Our experience has been that Overture users performed better and were slightly more qualified, but Overture is slightly more expensive. The gap with Google is closing though."
Berkowitz strongly recommends other publishers use both instead of considering Google versus Overture as an either/or proposition. Otherwise you won't reach the total universe you need to get the traffic you deserve.
-> Berkowitz currently has two full-time staff handing search in-house. He notes that due to the intensely competitive hiring marketplace for experienced staff, he has to worry about his own team being hired away, and is constantly on the lookout for possible replacement hires.
-> Landing page tests showed that consumers want a valuable result from their very first page. So, instead of starting visitors at a quiz page asking what trim they want on their car, Edmunds.com drops them into the results page showing the price for the model using the most popular trim.
From there visitors can either go back up the funnel to choose different trim, or they can go click directly onto a partner or sponsor's link to take action on the vehicle they desire.
-> 15-20% of Google ad traffic for Edmunds.com now comes from the contextual advertising AdSense ads as opposed to search results ads. This is growing.
-> Berkowitz notes overall average costs per click are rising, despite the broader pool of sites carrying search ads. The competition for top search terms is intense and not likely to slow anytime soon.
With this in mind, he's carefully limited bids for more general terms such as "Cars" which can cost a great deal but produce too much unqualified traffic. Instead, his team is aggressive only with niche terms that are proven to result in consumers ready to purchase from partners and sponsors.
Currently the team runs paid ads under 60,000 terms, and growing.
Useful links related to this article:
Omniture - the metrics software Edmunds.com relied on to measure their traffic