When Autobytel, an Internet automotive marketing services company, ramped up its paid search campaign in 2004 from 10,000 to about 150,000 words, the media and marketing services team became swamped with the work required to manage paid search manually.

Plus, with such a large base of terms, the company was unable to generate enough targeted landing pages, so visitors weren't seeing content that was as relevant as it should be. "We were losing potential," explains Michael Rosenberg, SVP Media and Marketing Services.

While Rosenberg considered an outside consulting firm to manage SEM, he knew their needs were complicated: online revenue is derived both from lead generation for car dealers and from CPM advertising, so calculating ROI is more complex than looking at a single transaction.

For example, he explains, the phrase "Ferrari photos" probably wouldn't generate many leads because few people actually buy a Ferrari, so calculating ROI based on lead generation wouldn't make sense. However, that term may generate 15 page views per visit -- so from a CPM standpoint, "it might be worth it," he says.

"But none of the top consulting firms at the time had a bidding engine that took both those revenue streams into account," he adds.

He needed a way to tie revenue per keyword back into his bidding strategy. Using internal resources, Rosenberg's team created their own engine to automate and optimize bidding based on both revenue streams and to create custom landing pages, eliminating much of the manual work that was killing time.

Here's what he learned about SEM during the process.

-> Best Practice #1. Bid by word, not "bucket"

Whenever possible, track performance of individual words.

"Before, we'd have a bucket around the word 'used' that would have hundreds of different words all relating to it (used cars, used trucks). We'd bid 35 cents per word for every word in the bucket, but then you underbid on words that are performing well and overbid on words that are performing horribly."

By using an automatic bidding engine or looking at individual words manually, you can see what you paid and what return you got and adjust bidding. Be sure to look at several, rather than a single, transaction so decisions are based on a sample size.

Manually bidding on thousands of words daily is, of course, unreasonable. Companies that don't have automated bidding should consider daily management of at least their top 50 words, Rosenberg suggests.

"It's the old 80/20," he says. "80% of your volume is the top 20% of your words."

-> Best Practice #2. Automate your landing pages

It's common knowledge that the more relevant your landing page is to the search term, the better your conversions. Yet again, when bidding on thousands of words, manually creating custom landing pages for each word isn't possible.

Companies get around this by creating landing pages for buckets of words, sending groups of visitors to a single landing page based on similar interests.

However, says Rosenberg, Autobytel has been successful automating the customization of landing pages, which enables them to create dynamically generated, relevant pages for most of their terms.

Here's how:

The company has standard pages for standard terms, with perhaps three or four areas designated on those pages for customization.

For example, they have a standard page for the term "dealer." When someone searches for other similar terms -- Honda dealer, Honda dealer Miami, Accord dealer -- the standard dealer page is prepopulated with Honda information or other information relevant to those more specific terms.

-> Best Practice #3. Customize your bidding strategy

Paid search products are very different from each other, so a one-bidding-strategy-fits-all approach won't work.

For example, on Overture you can buy the number-one ranking, which you can't do on Google, says Rosenberg. "So if it's absolutely imperative that you have a number-one ranking, for a branding play or whatever, Overture's your place."

Look at what you want to accomplish and how to best accomplish it on the various engines.

-> Best Practice #4. Develop tactics that deliver
As search matures, tactics like these will become the norm -- the human factor will distinguish the winners in the space, says Rosenberg.

By automating search, his team has been able to improve in at least five areas:

a. Copy: "It's the little piece of copy under the keywords that drives your clickthrough. You need to find what drives conversions." Write copy that has pull and attracts your customers, but be honest with them.

"It's tough, because you're walking a fine line. You want better clickthrough than your competitors, but it must be qualified clickthrough. So test different copy and see what conversions are like," he suggests.

b. Landing pages: There needs to be relevancy not just between the keyword and the landing page (repeating the word on the page) but between the ad copy and the landing page. What you promise in the ad, you must clearly deliver on the page or consumers sense the disconnect and drop out.

c. User interface: "We put a lot of emphasis on usability and navigation," he says, "and you really need to test those things," particularly through usability studies.

Simpler is always better. "You feel dumb when you find three people who say over and over, 'Why do I have to do this and this and this, why can't I get there easily?'" he explains.

d. Process: Rosenberg now makes sure his team is looking at the right metrics. "They should not just be looking at traffic, but at profitable traffic. If you have a metric of, 'I want tons of visitors from Yahoo!,' that's not useful."

e. Develop a strong keyword list: While Rosenberg's automated engine can take a single keyword and generate "tons" of similar keywords worth bidding on, "it doesn't generate every word under the sun," he says. Humans see "other ways of looking at it."

When he noticed, for example, that the term "pickup trucks" was doing well, he was able to work with his team to come up with other relevant terms not produced by the automated engine.

-> Best Practice #5. Delve into details

Whether you're using an outside firm or managing search yourself, it's important to look at the reports. (This may be more difficult when you're using a consultant, Rosenberg says, because then they're the ones studying the data rather than you.)

"It can give you interesting insights you might not get otherwise," he says. "Like half of our top 10 words were truck-related. That was a learning we had no idea about."