height=”100″ hspace=”5″ vspace=”3″ width=”100″>

If you buy search ads to drive sales, you are a direct response advertiser. Welcome, today’s short column is for you.

If you buy search to increase awareness of your brand, you’re a brand advertiser. Sorry, but this column is for the direct response gang, us red-headed step-kids of the advertising world. Save us a canapé from the awards banquet; we’ll be back at the office with our spreadsheets.

Just us direct response folks? OK, here’s today’s suggestion: we search marketers can learn a lot from the direct mail wizards.

Putting it more crudely: if Claude Hopkins were alive, his Google campaigns would kick ass.

Claude Hopkins. Albert Lasker. John Caples. David Ogilvy. Leo Burnett. Maxwell Sackheim.

If you don’t know these names, you should.

They’re the guys who invented direct response. It was called “direct mail” back then, but the rules are the same. Their ideas for getting an envelope opened, an ad read, and a check written are still effective today for getting a link clicked, an Add To Cart button pushed, and a VISA number typed in.

Long before the Digg crowd discovered the importance of headlines for linkbaiting, Claude Hopkins had that idea covered. Check out Hopkins on headlines — that was published in 1923!

The early mail guys knew what mattered: tracking your results, knowing your profitability metrics, marketing to strong lists, testing different versions, writing compelling headlines, crafting copy which sells.

To learn more, head to the library or Amazon. These dusty tomes are pure web marketing rocket fuel:

• Scientific Advertising, Claude Hopkins, 1923 (free online)
• Tested Advertising Methods, John Caples, 1932
• Ogilvy on Advertising, David Ogilvy, 1985

Another classic direct mail idea relevant to search: the List-Offer-Package Rule.

The List-Offer-Package Rule states that when you are trying to sell something remotely, the list (who you are communicating with) is more important than the offer (the details of what you are selling, the item, the pricing, the guarantee), and the list and the offer are more important than the package (how it looks, the copy, the artwork, color and typography).

Years ago a consultant gave me this example.

“Say you are selling sets of collectible china plates with the state birds on them. If you had a list of people who recently bought collectible plates, that would be a good start.

“Now if you had a list of people who had recently bought collectible plates with animals on them via the mail or web, that would be even better.”

“If you had that targeted list, you could practically send them a handwritten scribbled postcard saying ‘Hi I have some plates that would interest you’ and you’d pull a 5% response rate.

“Suppose you had a totally killer direct mail piece for these plates, with a powerful long letter written by an top DM copywriter, beautiful pictures and optimized response card, buck slip, the whole shebang tested and proven in the mail with solid split tests.

“Say you mailed this killer package to a generic list, perhaps, women aged 50+. You’d be lucky to see 0.5% response rate.

“That’s the critical idea of the List-Offer-Package Rule: even a plain note to a highly targeted list will outpull the perfect package sent to an untargeted list by 10 to 1.”

I personally learned the List-Offer-Package Rule a decade ago, and it has held true in everything I’ve observed since.

How does List-Offer-Package apply to search?

• List trumps everything. That means getting terms, bids, and match types right matters far, far, far more than perfect ad copy.
• Offer trumps copy. That means great site design, usability testing, and MVT improvements won’t help all that much if you’re out of competitive position on price and shipping.
• Response lists trump demographic lists. Microsoft and Google are excitedly promoting demographic bidding. We red-haired step-kids just yawn and turn back to our spreadsheets. As an RKG client insightfully commented regarding Microsoft’s demographic data: “I don’t care who they are, just so long as they want to buy our stuff.” If you wanted to see the direct response crew sit up and take notice, offer us the option to bid based on a recent online purchase flag. (More on this: PurchaseMatch: How GOOG Could Hit $750″ on rkgblog.)

Goto.com launched in February 1998. Google opened its doors in September 1998. The modern search industry is just 10 years old. Yet, many important ideas for making ads relevant and effective predate ‘98 by over 50 years.

To get a leg up on your competition in 2008, try spending an afternoon with your nose buried a classic direct mail book from the 1930s.

Alan Rimm-Kaufman leads the Rimm-Kaufman Group, an agency helping online retailers manage large-scale paid search campaigns and improve site conversion. The Paid Search column appears Mondays at Search Engine Land.

height=”100″ hspace=”5″ vspace=”3″ width=”100″> With Google’s recent announcement about page load time influencing Quality Score, now is a good time to discuss site speed.

Speed matters. People rate snappy, responsive sites as more usable, even when the user interface itself doesn’t change. If your architecture or design aren’t that great, your users will give your site more chances before abandoning if the site responds quickly. Google knows that speed supports usability. I’d suggest much of the credit for Google’s rise to industry dominance goes to their ongoing obsession on making search results blazingly fast.

So, how does site speed relate to PPC?

To begin, read Google’s announcement about speed and quality score. If you serve Google’s Adbots slow loading pages, you’ll eventually get hit with QS penalties and drive up your AdWords click costs.

Bots experience your site differently from browser clients, because bots can skip images and javascript. I don’t have inside information on this, but it is a safe assumption that the Google code evaluating AdWords landing pages is simpler than the code assessing natural search relevance, and likely isn’t pulling in and analyzing external resources.

To serve AdWords landing pages to bots more quickly, use server profilers to determine which components of web pages takes the most time to build. Don’t optimize before profiling; odds are you’ll apply your efforts in the wrong place. (One of my favorite software quotes is from Sir Tony Hoare: premature optimization is the root of all evil.)

Two likely systems which could be slowing your page build times are your database and your templating engine. To speed up your database, bring in the best database administrator you can find, check your slow query logs, invest in more memory and faster disks, and build smart database indexes to support common requests.

To speed up your templating engine, minimize trips to the database, bump up the memory on those machines, and cache common page components or entire pages. For example, if your SKU pages provide customer ratings and reviews (a very good idea), consider generating and stashing the review block each hour—conversion wouldn’t suffer if a product review was 59 minutes delayed getting on the page.

Other ideas to increase speed: gzip your pages, remove whitespace from your source, remove needless CSS markup, use shorter class and id names, and use the cascading properties of CSS wisely, applying styling at the highest level of the DOM possible, rather than repeating styles ad nauseum on child elements. Be sure to benchmark your pages before and after, because some approaches (like gzipping) could, depending on circumstances, cause slowdowns.

To speed up your load times, make sure any redirects between you and Google, or any redirects on your site, are blazingly fast. Avoid tracking redirects which query databases to determine where to send traffic. We know a significant retailer whose home-built e-commerce platform grabs every new visit, teleports it off to a tracking system which queries databases to place cookies, then returns the session to originally desired page. Slow for humans, slow for bots. Not good.

Of course, make sure your site lives on fast servers, hosted in top-notch data centers with fast pipes. Memory prices continue to fall, so again, make sure your page-building servers and database servers have generous amounts of RAM.

I’m confident Google’s speed checking will be quite liberal, with the penalty applying to only really slow pages. If you get hit with a page load QS penalty, Google says the AdWords dashboard will tell you so. If you don’t get these warnings, should you pat yourself on the back for a “fast enough” site and consider speed a non-issue?

Nope.

It is worth investing some additional effort to take your site from “reasonably fast” up to “impressively fast.”

Why? One word: conversion.

Faster sites sell more. As we’ve often said, conversion is the ultimate “secret weapon” in PPC. Sites which sell more per click can afford to advertise more per click. Google understands this principle quite well. That’s why they provide conversion tracking, rudimentary bid management, and tools like the website optimizer for free—Google wants you to sell more so you buy more ads and increase their profits and valuation.

Speeding up sites for browsers is somewhat different from speeding up sites for bots (though every improvement for the later also helps the former). To display pretty pages for humans, browsers need to load images, render HTML, apply CSS, and execute javascript.

There are some amazingly powerful and simple ideas to speed up page rendering time in Steve Souder’s excellent book, “Speed Up Your Site.” Souder held the title Chief Performance Engineer at Yahoo, before he recently jumped to Google. His short book gives practical advice on CSS sprites, using page caching and expires headers correctly, and being smart about DNS. Buy Steve’s book for your engineers—I promise the $30 you spend ($30 at O’Reilly, $20 at Amazon) will be the highest ROI you get all year. You can basically read the whole book online: instructions are on our blog, but it is worth buying the entire book.

Make sure your team is using the amazing Firebug plugin for FireFox, including the YSlow extension written by Steve.

Much older, but also decent, is Andy King’s Speed Up Your Site. Many of his tips are also on his site, websiteoptimization.com.

Don’t go crazy obsessing about site speed. Many of these techniques are simple—often just server tweaks and template changes.

Reasonably fast sites avoid Google QS penalties, saving you advertising money. Really zippy sites are more usable, increasing your site sales. Got it. But what’s the carbon footprint angle mentioned in the headline?

Steve Souder wrote a fascinating post last week titled How Green Is Your Web Page. He suggests if we all followed his simple advice and optimized our sites, the reduced load on our servers would save energy and so reduce carbon emissions. A few less CPU cycles or hard disk rotations might not sound like much, but Steve calculates if the entire industry went along, the net energy savings could add up.

Lower Google bills. Happier users. Increased sales. And helping the planet a bit, to boot. You have your marching orders: get out there and speed up your site!

Alan Rimm-Kaufman leads the Rimm-Kaufman Group, an agency managing large-scale paid search and multivariate site testing for online retailers. The Paid Search column appears Mondays at Search Engine Land.

height=”100″ hspace=”5″ vspace=”3″ width=”100″> Last Monday, in a SEL post discussing the click quality by engine, I mentioned in passing an 89% drop-off between ‘phrases tested’ and ‘phrases actively generating good clicks’. Specifically, we posted 176,903 terms en route to discovering 20,152 active good terms for a client.

Today I’ll revisit that drop-off and analyze similar data from two other retailers.

I enjoyed the Q&A format with Danny last time, so I’ll do that again. I don’t expect any hardballs, as today I’m answering my own questions!

Q: It seems like a lot of work to post 176K terms just to find 20K good ones. Why not run fewer terms and use broad match?

A: Our experience shows PPC advertisers get better results by handling the long tail themselves. It can be dangerous to rely on broad match on head terms. Broad match and phrase match can make sense for long tail terms, when used judiciously. And explicitly handling the long tail gives you finer control of landing page and copy. Q: “Head terms”? “Tail terms”? What do you mean?

A: Chris Anderson introduced the Long Tail concept in a 2004 Wired article, then later expanded it on his blog and in his book. The Long Tail argues that virtual retailers with lower inventory holding costs can carry more SKUs, and that the aggregate sales from many non-best-sellers can top sales from best sellers.

The Long Tail is a merchandising idea, but the metaphor has been stretched to other areas (”paid search is the long tail of advertising,” “terrorism is the long tail of war,” “microbrews are the long tail of beer”). In linguistics, Zipf’s Law says the frequency of word use follows a power law, with a tiny set of popular words seeing huge usage and a huge number of words seeing small usage. In paid search, high search volume single- or double-word phrases are called “head terms,” and three- or four-word phrases with lower traffic are “tail terms.” Q: If that advertiser found 20K good terms, why run the other 156K “dud” terms?

A: Great question!

Yes, why run terms which haven’t generated orders?

As I wrote last week, “if you wanna to catch alotta fish, you gotta keep alotta hooks in the water.”

The fish are the orders. The hooks are the advertised phrases.

Looking past the head terms, many of the tail terms which generate sales this month won’t generate sales next month, and many of the tail terms which generate sales next month won’t have generated any sales this month.

Yes, high volume head terms enjoy consistent clicks and sales. But which tail terms generate sales in any given month is highly variable, given their low click volume.

Let’s look at data from two clients from last summer. They are different clients than my post from last week because the client in last week’s post started with our agency just last fall, and I wanted four clean summer months of data without the holiday surge. For this post, given Google’s dominant share in the industry, I’m just showing Google data. And for this post I’m aggregating ad-level results up to phrase-level results. For bidding and tracking, we call an “ad” each unique combination of client, engine, phrase, match-type, copy, and destination URL; the aggregation here summarizes away the details of copy and destination URL tests.

OK. Meet Client #1, a small-to-medium size direct retailer with no stores, web pure play, a small website, small SKU count, and 4K active phrases.

[enlarge]

For client #1, only 4% of their phrases running in May generated orders (165/4356).

Now, what would have happened if, on the last day of May, we turned off the 4192 “dud” phrases which didn’t generate orders in May, and rolled into June with only the 165 “winner” phrases?

Sales in June would have dropped to $84K, down from the $118K that actually happened, a reduction of 27%!

Zounds. Let’s keep on playing this bad strategy.

What if on the last day of June, we turned off the 86 “dud” phrases within the remaining 165, rolling into July with only the 79 winners?

Had we done this, sales in June would have been $75K, 37% below what actually happened.

Ouch!

Here are those data for Client #1, recomputed with the assumption that at the end of each month we turned off all terms which hadn’t generated an order in that month.

[enlarge]

What’s the takeaway here?

Tail terms have low click volumes. Typical PPC phrase conversion rates are a few percent. A phrase with a handful of clicks and no orders in a given month hasn’t proven itself bad. Turning off these phrases shrinks your program, leaving significant sales and profits on the table.

Q: Retailer #1 is only spending $50K a month on 4K terms. Does your claim hold for larger advertisers too?

A: Yes, even more so.

Meet Client #2, a large direct retailer with no stores, multiple call centers, multiple catalogs, a large website, large SKU count, and 83K tested phrases.

[enlarge]

For Client #2, only 4% of their phrases running in May generated orders (3424/83655), coincidentally matching the rate for Client #1.

As before, what if we ended each month by turning off phrases which had zero orders that month?

[enlarge]

We see the same phenomenon as before.

June sales would have down 42% versus what actually happened. July sales would have been off 51%. August would have been off 48%.

Again, ouch. Even bigger ouch, as the absolute numbers are so much larger.

Q: So, you’re saying large term lists are good.

A: Yes.

Q: I’ve heard of two methods some SEM agencies use to bulk up term lists. Method one is prepend “BUY” to all your phrases, and then post-pend “ONLINE.” Shazam, that triples the number of phrases on your list. Method two is to use a product feed to cram all your product titles into the PPC engines. For example, advertise on phrase, “Nikon D40 Kit 6.1-megapixel digital SLR camera with 18-55mm zoom lens.” Shazam, that can add another few thousand terms. You said large term lists are good. What do you think?

A: No, no, no.

The goal isn’t to bulk up your term list just for the sake of bulking up your term list. The goal is to add relevant profitable ads.

Prepending meaningless words doesn’t help. Running absurdly long phrases doesn’t help. The game is won through quality, not quantity.

If someone is bragging about the size of their term list — “We have 189 gadzillion terms running on Google!” – the right next question is, “In the last 90 days, how many of those have had clicks, and how many of those have had orders?”

Q: What’s the right way to come up with extensive list of good phrases?

A: Here’s a secret: don’t start with phrases, start with URLs.

Take every selling page on your site – that’s the home page, every product category page, every subcategory page, and every product page – and for each page, develop a good set of unique phrases suitable for driving traffic to that page. Unless you’re testing destination URL, no phrase should go to to multiple landing pages.

Aim for testing 5 to 10 unique phrases for each URL you’re advertising on your site.

Q: If I understand your Client #1 and Client #2 examples, it makes economic sense to keep running phrases that have no orders, right?

A: No. Don’t waste your money like that. Bid down poor-converting phrases, and turn off the stinkers.

Q: If a typical phrase conversion rate is 1%, do I need to see 200 or 300 clicks before I turn off a phrase for no orders, as it takes that many clicks to prove it a loser?

A: No! You can use statistical techniques to make inferences about the conversion rate of low volume terms, and bid them or kill them appropriately. With good stats, you can be far more nimble.

Q: Ok. While the long-tail phrases are still collecting clicks, should I bid them all the same?

A: No, keeping all the low-volume terms at the same constant bid isn’t smart. Might as well send a signed blank check to Mountain View.

Even though they’re low traffic, it is very likely these terms are of different quality, and thus should be bid differently. Again, you can use statistical techniques to estimate the sales-per-click of low volume terms, and bid accordingly. Estimating SPC for low-traffic terms is the secret sauce of an effective bid management platform.

Q: This all sounds complicated. You’re saying that running large term lists poorly can be costly. Which is better:

1. Having a campaign with a smaller number terms, focusing on high-volume terms, and bidding them carefully, which is easier because there are fewer details to manage and because the performance of high-traffic terms requires no statistics to understand, or

2. having a campaign with a large number of terms, including many terms with low click volume, and bidding as best I can, by the seat of my pants?

A: Option 1 clearly beats Option 2.

Having a big term list and bidding poorly gives you that many more chances to waste your money. I’d strongly recommend Option 1.

But, if you can pull it off, the best choice would be Option 3:

3. Have very large term lists and bid well.

Q: Do bidding algorithms matter? Aren’t they all about the same?

A: Yes. No.

Q: Final question: Do you often conduct Q&As with yourself?

A: Sometimes.

Alan Rimm-Kaufman leads the Rimm-Kaufman Group, a direct marketing services and consulting firm founded in 2003. The Paid Search column appears Mondays at Search Engine Land.

Earlier this month, in a blog post reporting our clients’ aggregate ad spend across the major search engines, we noted that Google’s slice of the pay-per-click pie increased across 2007. I wrote:

“We never sat down and had meetings about moving budgets. Rather, our systems noticed that ever-so-slightly better clicks could be had on Google and so shifted spend there. ‘Better’ in this context means “more likely to generate sales dollars or margin dollars for our clients.”

This shift from Yahoo to Google was imperceptible on the day-to-day scale. The search wars are fought one click at a time. One bid at a time. A penny here and a penny there. But looking back over ‘07, the trend becomes clear: across our client base, Google won 5 points of share at Yahoo’s expense.

Danny Sullivan, illustrious Editor-In-Chief, emailed me some follow-up questions about how we buy clicks for clients (answer: carefully!), and how we allocate budget across the engines (answer: we don’t!).

At Danny’s suggestion, I’ll respond to his questions about that post more fully here. Danny’s words are presented in bold. I’ll also show an Excel analysis you can run on your own PPC data to explore your own efficiency vs. volume trade-off curve.

Danny Sullivan: OK, so I wanted to understand more. Did you have a set budget?

Most of our clients tell us to buy as many clicks as we can for them that meet their economic goals. Those goals could be a ROI target, an A/S goal, a CPO goal, a margin goal, whatever.

So, in general, no, we don’t have fixed dollar budgets. Instead, we have efficiency budgets. If the clicks are converting, we increase the spend.

That post was about our base of 100+ clients in aggregate. A small number of our clients do set hard dollar budgets, but they’re in the minority.

Digression: I recall one client, a well-known national store retailer, who in early December of 2005 reached their annual search advertising budget. They instructed us to turn off their ads and go dark, right before Christmas, right in the thick of their hottest selling of the year, when their ads were printing money, because they had hit the budget number they had set months earlier. Ouch! Happily, this retailer changed their instructions to us for ‘06 and ‘07, and now budget by efficiency, rather than by hard dollar amount.

Even when we manage to absolute dollar budgets, we get better results managing spend by managing bids and matchtypes. We often avoid the engines’ budgeting mechanisms. In their well-intentioned desire to serve advertisers, the engines’ ad-serving algorithms sometimes do dumb things to campaigns when advertisers set daily caps.

DS: Usually, people complain they can’t get enough traffic.

There’s ample traffic. What’s is scarce is good traffic. For each retailer, there’s a finite supply of clicks which meet their performance metrics.

When we optimize campaigns for new clients, we often see big improvements by adding terms, refining match types, and bidding rationally.

But at some point there’s a plateau.

That plateau is a function of how many qualified humans are searching for your goods and services, and it imposes a limit on how large your campaigns can profitably go. Some retailers don’t like to hear that, but it is true.

Busting through that plateau requires lowering your profitability goals, improving the conversion of your site, or changing your business fundamentals (merchandising breadth, pricing strategy, shipping rates, etc.) Bidding and terms and match types and copy can take you far, but only so far.

DS: You hinted at this [not enough traffic] with Microsoft.

Yes, we reported that in 2007 Microsoft clicks performed better for our clients, on average, than Google clicks.

However, for each Microsoft click we purchased for clients, we bought 13.6 clicks on Google(!). That’s a huge difference in scale, and this makes the Google vs. Yahoo performance numbers incommensurate.

Suppose there’s a private high school with a handful of students and a low student-to-teacher ratio, and they send 95% of their graduates to 4-year colleges. And let’s suppose down the block there’s a gigantic and diverse inner city high school with thousands of kids, and the big school sends 70% of their graduates to 4-year college. Even with a lower college matriculation rate, the public school is demonstrating far greater teaching excellence, because they’re achieving their results across a heterogeneous population and at a much larger scale.

That’s what’s impressive about Google: they’re delivering quality and quantity.

DS: So if Google is converting so much better, why are you spending at all with the other players?

Just because Google has on average better click quality than Yahoo doesn’t mean that there aren’t great clicks on Yahoo. There are. Most advertisers will find some great-performing phrases on Yahoo. Most advertisers will find some great-performing phrases on Microsoft too. And perhaps some on Ask. And maybe even some on TinyEngineNobodyHasHeardOf.com.

The issue is how many great clicks can you get.

There’s a bottom cut-off: if the inventory of good clicks on a particular engine isn’t above some minimum threshold, there’s a point at which that engine isn’t worth management attention or robot attention or tech integration costs or accounting hassle. And so advertisers just ignore that engine altogether. That is the barrier faced by TinyEngineNobodyHasHeardOf.com and NewStartupSearchEngine.com. And that’s one of the reasons Microsoft wants to buy Yahoo — as the #2 engine, Yahoo always gets considered. Sometimes, even at #3, Microsoft doesn’t.

DS: My assumption is that you can’t spend all you want on Google or you feel you need some visibility in these other places.

Yes, our clients would like to spend boatloads more on each engine, if they could do so profitably. Sadly we/they can’t, due to decreasing marginal returns on additional advertising.

For our clients buying search to drive profits — and such advertisers comprise the bulk of our client base — we don’t have any a priori need to spend any amount of money on any of the engines — we just buy what works.

To make this more concrete, here’s an Excel analysis anyone can do to explore the PPC quality versus quantity trade-off. I’ll do this analysis using data one of our clients, a well-known specialty retailer with a website, nationwide stores, and catalogs.

To date, we’ve tested 176,903 phrases for this client on Google. (More on Yahoo later.) Some of these phrases didn’t pan out, either because they had poor performance or they lacked impressions. Of the 176,903 phrases tested, currently 11% (20,152) are active and regularly generating good clicks on the engines.

If you’re scratching your head over a 89% drop-off between ‘tested phrases’ and ‘phrases regularly generating good quality clicks’, stay tuned for the second installment of this post next Monday, where I’ll talk about the long tail and the comprehensive term list approach using data from this same retailer. If you want to catch alotta fish, you gotta toss many hooks in the water.

OK, here’s that spreadsheet analysis for you to try on your data.

For each each engine, make a spreadsheet with these columns:

• Engine
• Phrase
• Brand Phrase Flag
• Impressions
• Clicks
• Adspend
• Resulting Sales
• Resulting Orders

If it’s too many rows for a spreadsheet, move up to a database.

If you assign different tracking codes to distinguish different ad copy and destination URLs — that is, if your tracking is more granular than phrase, which is a very smart idea — then for this spreadsheet, roll up your performance data by phrase.

As for time period, take performance over the last month. The time period over which you evaluate ads, both in days and in clicks, matters a great deal for effective PPC bidding, and differs between head and tail terms. Let’s gloss over those details here and just take last month to keep it simple.

Now, add these derived columns to your spreadsheet:

• CPC (cost over clicks)
• SPC (sales over clicks), and
• A/S (adspend over sales).

For this example, we’ll use A/S as proxy for profit. This is a decent first-order approximation, and a pretty accurate if your various product categories have similar margins.

Sort this sheet by A/S ascending (primary) and sales descending (secondary). This orders your phrases from best to worst.

Now add in columns for cumulative sales and cumulative ad cost. By “cumulative sales and cost,” I mean the total sales and cost downto and including the current row on the spreadsheet. Here’s an example spreadsheet showing the formulas for computing cumulative sales and costs.

Compute cumulative A/S by dividing cumulative ad spend by cumulative sales.

Still with me? You should have a sheet that looks something like this:

I scrolled down a thousand rows in that screen image to reach below the handful-of-clicks-one-order ads, as these have essentially zero A/S. For example, the single best A/S phrase for this retailer during this period was a two word phrase, a manufacturer’s brand name with a SKU model number, which consumed 42 cents of ad cost, two clicks at 21c each, and generated one $2899 order. That’s a mind-blowing A/S of 0.017%. Sweet!

Now, plot downto ad spend against corresponding downto sales.

You’ll get a trade-off curve, or trade-off horizon, something like this.

Every point on this graph corresponds an active search phrase running on Google for this client. There’s 20,152 maroon points on this graph, so close together they blur into a line.

I marked three regions on this graph: “I”, under the curve; “E”, on the curve itself; and “U”, above the curve.

You’d never want to be in region “I”. “I” stands for “inefficient.”

Suppose this retailer had a portfolio of search phrases generating $800K in sales via $80K in ad spend. Not so good.

Done right, the graph shows you could generate $1.4m in sales from the same $80K in advertising. $1.4m is much better than $800k.

The curve “E” denotes the efficient trade-off curve. If you’re optimizing your portfolio reasonably well, the curve represents the best sales you can get for each level of ad spend. You can’t do better than being at some point along the “E” curve.

The region above the curve, labeled “U” for “unreachable,” is beyond the horizon if the curve is optimal.

So, which point along the “E” curve is best?

A trick question!

There’s no right answer for all retailers. It depends how advertiser values the top vs. bottom line, sales vs. profits.

You can compute the point on the curve that maximizes profits (see How Much Advertise and accompanying Excel model). But not all retailers seek to maximize their bottom line, and that’s not wrong. Many place a higher premium on top line, or new customer acquisition, or whatever.

Here’s that curve again, w/absolute ad spend replaced with A/S.

This graph highlights the trade-off between the retailer’s choice of advertising efficiency and resulting sales.

It always comes down to the same choice eventually: you can have a super-efficient small program, or a less efficient larger program.

OK, back to Danny’s question. Here’s the same graph again, showing phrases on both Google and Yahoo.

The 20,152 maroon points again represent Google. They haven’t changed from the prior graph. There are an additional 2,634 little blue circles, most clustered and overlapping on the left side of the figure. The little blue circles represent Yahoo ads, placed along the same trade-off continuum.

See how for this advertiser the blue circles thin out pretty rapidly as the curve heads right.

It isn’t that we didn’t try the same 176,903 phrases on Yahoo as on Google. We did.

It is that, compared to Google, many fewer of these 176,903 on Yahoo had both (a) search inventory and (b) sufficiently high performance. And sadly, many of the leftmost points are retailer brand phrases, which often aren’t incremental.

Back to Danny.

DS: But also, is traffic a factor? I mean if Google has more traffic, you’re getting more searches there, more clicks there, and that produces a spend change too, doesn’t it? Even if you don’t do anything?

Again, it isn’t just traffic; it is traffic that converts.

Of course, it’s also your systems that are making these moves.

Yep. You need solid technology to manage campaigns at this scale. Systems and algorithms do matter a great deal.

For example, it isn’t smart to “explore the trade-off spectrum” by bidding each ad up and down the page to determine its response frontier. That’s an inefficient and costly heuristic to explore the trade-off curve.

But is part of it due perhaps to Yahoo getting cheaper? I mean, they have an entire new system. If it’s costing less to be on Yahoo, that stinks for Yahoo, but it’s not necessarily a sign that Yahoo isn’t working well, is it?

Great points.

We don’t see Yahoo getting “cheaper.”

“Cheap” or “costly” clicks are more than just low or high CPCs. Some phrases are overpriced at a penny; others are a fantastic bargain at ten bucks. You can’t just compare CPCs, you need to compare click quality (A/S) too.

I’d love to see you discuss these types of things. cheers, danny

Thanks for the opportunity to respond today!

Alan Rimm-Kaufman leads the Rimm-Kaufman Group, a direct marketing services and consulting firm founded in 2003. The Paid Search column appears Mondays at Search Engine Land.

In this post, I’d like to introduce a handy little tool for paid and natural search called RKG Duck. I wrote the app a few years back, and we find it useful in our agency’s work in paid search and site conversion optimization. We’re releasing RKG Duck under the GNU Public General License, so you can use and extend it relatively freely.

RKG Duck is a Windows GUI to manage filters which intercept the Windows cut-and-paste clipboard. The GUI lets you select a filter, turn a filter on and off, and even make changes to the filter code. The filters are written in standard Perl. You can easily extend the tool by writing additional filters. When you do, you have the full power of Perl, Perl regular expressions, and the awesome depth of CPAN on your side.

What’s new here, you might ask?

Convenience, power, and speed.

To munge data without RKG Duck, you’d save the data to a text file, make sure you have Perl on the machine, write a short Perl script, pull up a Windows command line, debug and run the script, redirect the output to a text file, open that file, copy the data, and paste the transformed data where you need it.

With RKG Duck, you turn on the filter, highlight the data within the Windows app, type “control-C” to copy the data to the Windows clipboard, then type “control-V” to paste back the transformed data, replacing the original. Done!

Because the data flows through the clipboard, you can grab data from any Windows app and paste the data to any Windows app. To see this in action, check out the video demo on YouTube.

Included with the RKG Duck distribution are a dozen or so generic filters. These help you do things like extract URLs from documents, pull SKUs from documents, parse parameters from URLs, and splice data in Excel.

These are useful features, but the real power of RKG Duck lies in extending it with your own additional filters.

Suppose you have an in-house method for adding tracking parameters to URLs. RKG Duck can assist, and it can handle complicated situations, like assigning the correct previously-assigned URL id if some URLs appear multiple times. [1]

There are many places in paid and natural search projects where fast data manipulation staying inside a Windows app can save you considerable time.

Our agency runs Windows on many desktops, but we’re a Linux IT shop. Because the RKG Duck filters are honest-to-goodness valid Perl filters, we could swap the exact same filters between the Linux command line, from cron jobs, from production pipelines, and our Windows desktops. That cross-platform angle is handy. It is also nice that you can run RKG Duck without installing Perl on each machine. [2]

The downsides of RKG Duck? Well, you need to know some basic Perl to write your own filters. Also, you can get some real jumbles when you forget to turn off a filter and keep working, as RKG Duck intercepts the Windows clipboard across all Windows apps. (“Hey, I just URI encoded a paragraph in a client email!”) But no worries—such jumbles are quickly reversed with a simple “control-Z,” the Windows Undo shortcut.

Our firm has benefited greatly from the Open Source movement, and so we try to give back to the community where we can.

Related links

• RKG Duck

• Demo video
• Perl regular expressions: introduction and perldoc
• ActiveState Perl

Footnotes

[1] Our proprietary paid search platform handles all the issues of embedding our own and our client’s analytics tracking parameters into engine-bound URLs automatically. However, we still find RKG Duck useful in reverse: we sometimes use filters to strip our tracking and additional client tracking from URLs obtained from clients and the engines, giving us “clean” URLs to dedupe and subsequently reprocess.

[2] To use additional CPAN modules, you do need to add directories to the RKG Duck folder. But that’s less involved than installing Perl and downloading modules, and you don’t need to do this machine by machine. The RKG Duck folder can live on a shared file system, allowing the whole organization to share a common set of filters.

Alan Rimm-Kaufman leads the Rimm-Kaufman Group, a direct marketing services and consulting firm founded in 2003. The Paid Search column appears Tuesdays at Search Engine Land.

We’re entering the home stretch: Christmas is about six weeks out. For many of our online retail clients, the next month and half of revenue will make or break their year. PPC impressions will soar. CPCs will rise. And SPCs (sales-per-click) will skyrocket. Most clients will see big profits from their paid search campaigns, even with higher CPCs.

For clients who’ve been with our agency for over a year, we review prior years’ data to get a better sense of when their holiday surge kicks in. Looking one or two prior years of sales data helps us stay on top of the shopping tsunami, reacting even more quickly than our automated predictive statistical algorithms would alone. Prior data also helps our robots from overshooting the holiday, avoiding all-to-common industry problem of bids being kept high longer than appropriate after holiday.

So, being knee-deep in our databases looking at year-over-year data, it seemed a good time to share some comparisons of 2006 versus 2007.

In the data which follow, I’ve restricted my focus to those clients for whom we handled PPC in both ‘06 and ‘07. That way, the data reflect changes in the PPC industry and/or changes in our clients’ marketing strategies, rather than reflecting our firm’s growth. I’m reporting averages (vs. medians), but the distributions aren’t horribly skewed, so these averages are reasonably representative.

Year-to-date 2006 vs. 2007 (that is, comparing January through October last year versus January through October this year), our clients were, on average, up 24% in search impressions. This isn’t a function broader keyword lists—we were obsessive about large term lists last year and are just as much this year—or of more aggressive bidding. Rather, I think this increase reflects the intrinsic growth of search, as well as the engines serving more ads per query.

Impressions went up 24%, but clicks only increased 5%. In other words, CTRs dropped, from an average of 2.1% in 2006 to 1.8% in 2007. I’m not sure if other advertisers had similar experiences. I’d love to know Google’s overall CTR from ‘06 to ‘07—are CTRs up, suggesting better relevance matching from Google’s ad serving engine, or are CTRs down, indicating searcher burnout on paid ads? (I’m not holding my breath waiting for Google to share the answer.)

CPCs rose 5%, from an average of $0.51 in 2006 to $0.54 in 2007. Again, our absolute CPCs depend on our agency’s advertiser mix, but I’d hypothesize this 5% year-over-year increase is in line with the overall industry. Of course, CPCs in highly competitive sectors increased more dramatically.

Conversion rates rose 8%, from 1.34% in 2006 to 1.44% in 2007. We attribute this happy situation to improvements on our clients’ websites. Strong site analytics and strong site testing are becoming more the norm among the IR100, and I think the conversion bump shows that, on average, these efforts are paying off.

Digging deeper, we characterized our long-term clients as “online pure-plays,” “catalogers,” or “brick-and-mortar store retailers.” These categories are fuzzy, as many catalogers have retail stores, and many national retail chains also have websites and catalogs. Nonetheless, we characterized each based on how they generate the bulk of their revenue.

Of the three business types, our catalog clients typically saw the smallest year-over-year change in key metrics like CPC, CTR, and cost-per-order. We’re taking this as a measure of the maturity of catalogers using paid search—many have been using paid search effectively for many years, and so on average catalogers aren’t making revolutionary shifts in budgets or strategies.

As one might guess, our national retail clients saw the greatest change in many of these year-over-year metrics: CPCs up 21%, sales up 30%, conversion rates up 35%. We watched many of our store clients overhaul their websites and increase their web marketing budgets in 2007, making their online stores competitive and moving budget from national store branding into their online groups. These aggregate stats reflect these big changes.

One of the key metrics we manage is A/S, the advertising to sales ratio. A/S describes how much of revenue an advertiser invests in marketing, so A/S is a measure of advertising efficiency and/or marketing aggressiveness. The “right” A/S target depends on a retailer’s business strategy, margin structure, and cash requirements.

Across our web pure-play clients, A/S targets increased from an average of 18.6% in 2006 to 20.8% in 2007. This is a much, much larger increase than for catalogers and retailers. Again, this statistics is computed client-over-client, year-over-year, so margin isn’t a factor. What this big change means is that, on average, our online pure-play clients bid more aggressively for sales in 2007 vs. 2006, matching my perception from client strategy meetings.

For anyone interested, we’ve also sliced our year-to-date results by Google vs. Yahoo vs. Microsoft, here on SEL in OctoberGoogle, Yahoo, Microsoft: Year-To-Date PPC Report Card), and again on our blog last week (October 2007 PPC Ad Spend: Google Jumps, Yahoo Slumps, Microsoft Steady).

Stores beefing up their sites and shifting budget to the web. Catalogers holding the line. Pure-plays getting more aggressive. And all of us poised for the upcoming holiday surge.

The next six weeks will come fast and furious, with consumers deciding the winners and the losers click by click. Buckle your seat belts, hold on tight, and best of luck!

Alan Rimm-Kaufman leads the Rimm-Kaufman Group, a direct marketing services and consulting firm founded in 2003. The Paid Search column appears Tuesdays at Search Engine Land.

The nice folks from Google were in for a visit a week or so ago. One of the topics on the day’s agenda was Ad Quality. If you read extremely closely, much of what they presented in their briefing is also described at the AdWords help center (see What is a ‘Quality Score’ and how is it calculated?). They also presented some new angles and dispelled a few myths.

Here are some of the key takeaways on Quality Score (herein abbreviated “QS”):

• QS determines minimum bids. If an advertiser’s maximum CPC < minimum bid, the keyword is "inactive for search"
• Myth: There is only one QS. Nope. One QS sets minimum bid, a different QS sets rank. There are different QS for search and content.
• Myth: Match type impacts QS. Nope. QS is calculated only from queries which exactly match keyword.
• Myth: QS improves with higher position on the page. Nope. QS is normalized to account for higher CTRs higher on the page.
• Myth: High reported CTR implies high QS. Nope. QS depends on CTR on Google.com alone, whereas reported CTR includes other Google properties.
• Myth: If you restructure your account, you lose QS history. Nope. History of keywords, copy, and destination URLs are maintained, as long as that combination is unchanged. (Even in a different account!) However, if you change either keyword, copy, or landing page, yes, QS may change.
• Myth: Pausing an ad harms QS. Nope.
• Myth: QS only matters on competitive words. Nope. The min bid is set based on QS, not on the number of advertisers in that auction. Even if there’s no competition, poor QS could dictate higher CPCs.
• Myth: QS are updated daily. Nope. It depends on volume and statistical significance. High volume terms could have intra-day QS updates, while low volume terms could mean multiple days between recalculations.
• Myth: Any Flash on landing page harms QS. Nope.
• Myth: QS only depends on you. Nope. Some factors involved in setting QS are computed keywords system wide, across the performance of all advertisers.

While we greatly enjoy and appreciate our Google visits, the entire QS topic leaves me confused and a bit saddened.

I don’t think anyone would disagree with the statement that QS is a highly complicated and opaque algorithm. That statement saddens me. QS is too important, too central to modern advertising, to be such a black box.

Google would probably argue that QS must be complicated and must be hidden, because the problem of determining ad relevance is a fundamentally hard problem, and because openness would enable the Bad Guys to game the system. The latter argument smacks of security through obscurity. Far better would be an open secure system which provably couldn’t be gamed.

I’ve posted before on my belief that opacity is Google’s Achilles’ heel. As the wonderful machine built by Page and Brin et al controls an ever-growing portion of the media landscape, transparency and openness become ever more critical. Google isn’t regulated like a public utility or a financial trading exchange—at least, not yet. I’m not a Google stockholder, but I am a fan of paid search in general and of Google in particular, and I want the channel to succeed and grow.

In the short term, I’m confident that following Google best practices for terms, copy, and landing pages will yield high QS ads for our clients, which will help our firm buy clicks for our clients efficiently. That’s what we’ll continue to do in our shop, and we’d recommend that follow-Google’s-suggestions approach to others looking to buy search efficiently.

In the long term, however, I hope Google reconsiders QS. It is time the search industry grows beyond the “just trust us” black-box approach for ad relevance. I humbly suggest that Google users, Google advertisers, and Google stockholders would be better served by some yet-to-be-discovered transparent and open model for ranking ads and charging for clicks.

Alan Rimm-Kaufman leads the Rimm-Kaufman Group, a direct marketing services and consulting firm founded in 2003. The Paid Search column appears Tuesdays at Search Engine Land.

What is the “long tail” of paid search, and why does it matter?

Chris Anderson coined the long tail concept in a 2004 Wired magazine article. Anderson’s original argument applied to online merchandising. Because web-only merchants (think Netflix) should have lower inventory carrying costs than traditional retailers (think WalMart), web-only merchants can afford to offer a broader catalog of items. Anderson describes a graph where the x-axis represents SKU sales-rank and the y-axis represents corresponding sales by SKU. Best-selling items appear on the left side of the distribution. These are hit products with large sales, the products carried by traditional retailers. This graph drops off steeply, then heads rightward with an nearly flat long tail. The tail are the niche products, individually unimportant, but—and here’s the big insight from Anderson’s article—collectively significant.

The shape of Anderson’s curve is a power law. Power laws pop up in natural science, chaos theory, computer networks, and languages. In linguistics, Zipf’s “Law” states that the distribution of word frequencies in a text follows a power law.

Let’s check out Zipf’s Law for ourselves. If you take the text of Moby Dick and count each word, you get this graph:

You can see the most prevalent word by far is “the”, at about 15k occurrences in Melville’s novel. The second most popular word, “of”, is far less popular: under 7k occurrences. Note how the graph drops very rapidly and basically levels out. (The graph continues far far far rightward, not shown.) This plot shows that Moby Dick—like any text—contains a tiny number of super-high frequency words, and a huge number of words with very low frequencies.

Of course, Moby Dick isn’t a novel about “the”. In aggregate, the long tail of low frequency words comprise most of the text.

Also, note that the 21st most frequent word in Moby Dick is “whale”. “Whale” isn’t a particularly common word in normal English, but it is highly prevalent in this whaling novel. This reminds us that notions of “common” and “uncommon” depend highly on context.

By now, you might be wondering if any of this matters to paid search marketers.

It does.

If you take the phrases in a search portfolio and order them by decreasing clicks or cost, you’ll quickly see a tiny handful of terms comprise most of your clicks. Try it out. The distribution of click frequencies in a well-designed paid search phrase list approximately follows a power law.

I pulled sample data for one of our clients, an online specialty retailer ranked in the Internet Retailer 100. We are running about 45K active phrases for this client on each major search engine.

These 45K terms aren’t created equal. The top 10 phrases comprise 64% (!) of their clicks. And phrases 11 through 40 comprise 27% of their clicks. That is, the first 40 terms—less than one tenth of one percent of their total term list—account for 91% of their clicks! Phrases 41 through 45K, which comprise 99.9% of their search phrases, collectively account for a mere 9% of clicks. Wow.

So is the long tail important to this client? You bet. Phrases 41 through 45K generate 16% of their total PPC sales. And because these phrases have, on average, lower CPCs than the high-traffic head terms, they comprise 19% of the client’s total PPC profit.

The lazy search marketer might look at these numbers and reason: “Hmmm…. I can get 91% of the traffic with less than 1% of the effort. I’ll run 40 phrases, get most of the bang for not that much effort, and head out early for the day to do some golfing.”

The savvy search marketer views the situation differently: “Hmmm…. running good ads on 45K terms will take a boatload of work, getting all those bids and match types and landing pages and ad copy done right. Ouch. But not doing this would mean walking away from 16% of revenue and 19% of profit… I guess it is time to roll up my sleeves and dig in.”

Is the long tail right for all search advertisers?

Perhaps not. Big campaigns with tens of thousands or hundreds of thousands of terms require much more care and attention than smaller campaigns of a few hundred top terms. Big campaigns offer more dark nooks for inefficiencies and errors to hide. You need good systems to generate and maintain relevant and targeted ad copy. You need advanced statistical approaches to cluster ads to compute bids intelligently (because almost every phrase in the tail lacks sufficient traffic to assess its conversion individually with any degree of statistical certainty.) In short, embracing very large term lists offers both opportunity and risk. No doubt: a small campaign executed well will out-perform a large campaign executed poorly.

Executed well, the long tail terms offer modest gains in sales and earnings (say, low double digit percentage increases). Managed poorly, the long tail can easily stab you in the back, generating significant cost overruns and earnings losses (say, high double digit percentage decreases).

Eight suggestions regarding large search phrase portfolios:

• Don’t apply sweeping simplistic bid rules (for example: “Any phrase with less than 10 clicks / month, bid $0.50″) to the long tail. This is a sure-fire way to torpedo your efficiency.
• Don’t use generic ad copy with title substitution. These shortcuts hurt performance. Write targeted relevant ad copy which matches the actual phrase.
• Don’t use generic landing pages. Match each phrase with a highly specific targeted and relevant deep-linked destination URL.
• Don’t confuse quality with quantity. It is easy to bulk up phrase lists in dumb ways. We’ve seen agencies seeking quantity over quality add useless additional words to phrases (prepending “buy” to every phrase doubles your list!) or run extremely long detailed phrases which will never see impressions or clicks (for example, full SKU names like “Extra Strength Oscillating 45mm Titanium Widget”).
• Don’t manage campaigns of more than a few hundred phrases “by hand” using spreadsheets.
• Don’t manage all phrases identically. The left and the right sides of the distribution require different strategies. Success in the head is about copy testing, day-parting, and match-type optimization. Success in the tail is about aggregation, smart copy management, and aggregation.
• Don’t update bids on all terms with the same frequency. Be nice to Google (and save yourself API fees) by bidding long tail phrases only as necessary.
• Don’t fall back to a “just run the winners” approach, turning off long tail phrases prematurely. Most long tail phrases generate zero sales in a typical month. The low-frequency phrases which generate sales this month won’t, on average, be the low-frequency phrases which generate sales next month. Use strong statistical portfolio clustering approaches to bid low-frequency phrases smartly.

The long tail. Done well, a great place to find reasonable performance bumps in PPC campaigns.

Alan Rimm-Kaufman leads the Rimm-Kaufman Group, a direct marketing services and consulting firm founded in 2003. The Paid Search column appears Tuesdays at Search Engine Land.

Here are eight observations about writing and testing pay-per-click ad copy. They’re broad generalizations, so there’ll be situations where these don’t hold, but we’ve found them generally accurate.

1. Copy is a 2nd-order effect. Writing good copy for your ads does matter, but copy quality is a 2nd-order effect on your PPC success. That is, copy matters far less than PPC’s two first-order effects, which are (a) getting your term list right, and (b) getting your bids right. Using a car analogy, the term list is the engine, bids are the transmission and tires, and copy is the paint job.
2. Bad copy hurts more than good copy helps.The following graph presents the results of a laborious and top-secret scientific study of copy quality vs. copy effectiveness.

   

   Actually, I just made those data up. But the graph does capture our many years of experience experimenting with PPC copy. We’d suggest there’s significant benefit in fixing bad copy (moving up from a grade of “D” or “F” to a “C” ), some benefit from improving decent copy (”C” up to “B” or “A”), but the additional benefit that accrues from perfect copy is often small.

   What do I mean by “bad” copy? Bad copy is overly generic, relying too heavily on title slugging, or is poorly matched to the search phrase. (Because of how the engines serve ads, getting copy right is closely tied to term list, match type, and negatives).

   Back to the car analogy, a sloppy amateur paint job can really torpedo the sale value of a car. Decent paint would get the car back to normal value. A perfect paint job would increase the sale value somewhat more, but likely by less than its cost.

3. Ad copy is disproportionately interesting to senior management. Copy is more visible than terms, bids, or campaign performance reports. As such, ad copy can attract more attention from senior management than is sometimes warranted. This isn’t a bad thing, unless an over-emphasis on copy testing diverts attention from higher value analyses.
4. Copy should sell you, not the SKU. Specific search terms indicate the searcher has a clear idea as to what she or he is seeking. In these cases, don’t waste precious characters selling the item. Help the searcher choose your ad from amongst the columns of competitors; use copy to sell your company. Searchers want to know, “Why should I buy this from you?”, and good copy answers this question.
5. Click-through isn’t conversion. You can load multiple copy versions in the search engines and instruct them to favor the ad with the highest CTR. (”Please Brer Rabbit, don’t throw me into the briar patch!”) But click-through isn’t conversion, and your highest-costing copy may or may not be your highest-selling copy. Understand when you want to maximize CTR and when you want maximize conversion, and set up your test accordingly.
6. When testing ad copy for conversion, use duplicate-ad design. We’ve seen our top Google agency contacts change their minds a few times over the last few years on Google’s recommended way to test copy, intra-adgroup versus inter-adgroup. Because click-through rate influences ad-serving, we’ve always advocated for using duplicate adgroups to test copy. As of this summer, Google again agrees.
7. Use stats to distinguish signal from noise. The difference between copy versions can be small. Don’t base decisions on random fluctuations. Use statistical methods to ensure performance differences between versions are significant. More often than not, they aren’t.
8. Test shouts, not whispers. Given how hard it is to detect performance differences between different copy versions, don’t waste your time testing nearly-similar copy variations against one another. Test boldly different messaging – that’s the only way you have a shot at discovering a new presentation that provably sells better.

Those are our eight recommendations for creating PPC copy that will blow your competitors away. As always, your mileage may vary.

Alan Rimm-Kaufman leads the Rimm-Kaufman Group, a direct marketing services and consulting firm founded in 2003. The Paid Search column appears Tuesdays at Search Engine Land.

As 2007 is now half over, it seems a good time to compare the performance of the Big Three search engines year-to-date.

The punch line: Google reigns dominant, providing the lion’s share of clicks, and Google clicks convert well. Microsoft offers strongly converting clicks at a lower-than-expected cost—good efficiency, but sadly almost no volume. Yahoo commands 22% of ad spend, but lags in click quality.

The Sample

My agency provides paid search management for about 100 sites, some from the Internet Retailer 100, most from the IR500. About 85% of our clients are B2C; about 15% are B2B. Our client base is largely direct response advertisers—catalogers and pure-plays who buy clicks to sell product, in contrast to advertisers who buy clicks for branding or traffic.

Nearly all of our clients instruct us to run their paid search campaigns to achieve their economic goals. That is, none of our clients establish a priori budget levels by engine. Our portfolio bidding platform optimizes ad budgets, buying the most effective clicks first. Thus, an increase in ad spend on one engine, relative to the others, reflects an increase in click quality relative to the others.

We’ve aggregated our results across all our clients for these observations.

The Methodology

Our agency has grown steadily during 2007. Because of this growth, graphing absolute spend and clicks says more about our performance than about the performance of the engines.

To remove the effects of our own growth and of seasonality, we’ve normalized the data by dividing each engine’s data by the monthly total of the three engines, thus expressing each engine as a percentage of the monthly whole.

To simplify this analysis, we’ve excluded data from Ask, paid inclusion, and the shopping comparison engines.

In all the graphs which follow, Google is represented inblue, Yahoo inred, and Microsoft ingreen.

Ad Spend

As we’ve previously reported, Google comprises about 73% of our total agency pay-per-click ad spend that goes to the Big Three engines. Yahoo comprises 21%, and MSN at 6% takes up the rest.

That graph looks effectively flat to me.

If our data is representative of the industry, Google is by far the dominant engine, and that’s been the case all year, despite Yahoo’s Panama launch and Microsoft’s increased focus on AdCenter.

PPC-driven Sales

Let’s look at tracked client sales driven by paid search, by engine.

We see Google’s driving about 73% of our clients’ aggregate PPC sales. Makes sense, as they comprise about 73% of our client’s ad spend.

Again, the three sales lines are relatively flat across the last six months.

As would be expected, other primary share metrics—monthly impressions, clicks, orders, items, etc— show the same pattern. All those graphs show three relatively flat lines, stacked BLUE / RED /GREEN, indicating Google in first place by a large margin. Not interesting.

Whatis interesting is to divide these monthly share metrics by other metrics. For example, click share divided by impression share gives a click-through rate index. Sales share divided by click share gives a sales-per-click index. And ad spend share divided by click share gives a CPC index.

So, looking at these derived indexes, 100% translates to “at expectations.”

Monetization

Wall Street cares about how well each engine does at translating searches into revenue. Dividing monthly ad spend share by monthly impression share yields a CPM index.

This graph shows Google and Microsoft do equally well monetizing their search traffic, while Yahoo has been about half as effective. Basically flat year-to-date, perhaps with slight gains for Microsoft.

Bid Competitiveness

Dividing monthly ad spend share by monthly click share yields a CPC index. This index provides a measure of the competitiveness of the bid landscape on each engine. Advertisers prefer fewer competitors in their auctions, of course. Search engines want to bring in more advertisers to drive up CPCs. CPCs are also influenced by click quality—smart advertisers use sophisticated tools to align their bids with click quality.

This graph shows Google enjoys higher CPC index than Yahoo and Microsoft.

Lower CPCs are good for advertisers. But CPC doesn’t tell the whole story. It isn’t how much a click costs that matters; it is how much that click costs relative to how much that click sells that matters.

Ad Relevance

Dividing monthly click share by monthly impression share yields a CTR (click-through-rate) index. This gives some sense of how well each engine does in serving relevant paid ads on search results pages—if the ads aren’t useful to searchers, they won’t get clicked.

The graph shows Google performing as expected, with click share nicely aligned with impression share. Microsoft does slightly better than expected. Some of this could be noise—Microsoft provides our clients only one-thirteenth the click volume that Google does—but the green line is consistently above the blue. Yahoo lags in distant third, with clicks coming in only at 60% of the level to be expected from their impression volume.

Click Quality

Dividing monthly sales share by monthly click share yields a SPC (Sales Per Click) index. The SPC index provides a measure of the quality of the clicks being sold. SPC reflects both the average likelihood a click converts (conversion rate) and the average value of a conversion (average order value).

This graph shows Microsoft quality has increased to Google’s level over the course of the year. Yahoo has held steady at lower quality of traffic. That is, across our clients, Yahoo clicks provide less sales than would be expected, relative to performance on Google and Microsoft.

We can break out SPC into its two components, conversion and average order. Here are both of them:

These graphs suggest that, among our clients, Yahoo’s weak sales-per-click stems from both lower conversion and lower average order values.

Most retailers care about sales dollar volume, not sales item volume, so this graph of average items per order falls into the interesting-but-not-important category. It does show that Google and Microsoft searchers are more likely to purchase multiple items per order than Yahoo searchers.

Advertiser Efficiency

Direct response advertisers don’t give a whit about how well the engines do monetizing their SERPS or encouraging searchers to click. Online retailers typically care about driving site revenue, and lots of it, and doing so cost-effectively.

One measure that nicely captures the “what-did-I-spend-and-what-did-I-get” trade-off is the A/S ratio. Diving ad spend by resulting sales indicates the efficiency of the advertising.

So, dividing monthly ad spend share into monthly PPC-driven sales share yields an A/S index, which measures the effectiveness of each engine for our clients. Note that lower A/S numbers are better than higher, as a lower A/S indicates higher efficiency.

Here’s that key graph:

We see that Google comes in at 100%, which is baseline. This is to be expected, given Google’s dominance, and given the portfolio optimization performed by our proprietary bid management platform.

Microsoft, with typically lower CPCs and higher SPCs, enjoys better than expected click efficiency. Microsoft’s efficiency is attractive to advertisers—the only fly-in-the-ointment is their paucity of traffic. (Microsoft has had trouble on their side getting clients launched in a timely manner—one wonders if this efficiency is due to an as-of-yet incomplete pool of advertisers.)

Yahoo comes in with the highest (worst) A/S index, resulting from their lower-than-expected click conversion and lower-than-expected resulting order size.

Your Mileage May Vary

These data represent our agency’s clients’ experience in aggregate. Our client base may not be representative of the search industry as whole. In particular, Yahoo has made strong progress selling search to traditional brand advertisers, who are under-represented in our client base of direct marketers.

So, your mileage may vary. Take all these observations with a grain of salt.

We’ll update these graphs at year’s end to see what changes.

What All This Means To Advertisers

Not much, I’d suggest. Engine-by-engine share results shouldn’t change how advertisers approach paid search. The fundamentals remain the same, regardless of where you are buying clicks:

• Use smart tracking and strong bid management to buy as many high-converting clicks as you can, regardless of engine
• If you are interested in profit, bid by your economics, not by position
• Use large term lists to exploit the long tail, paying particular attention to how you bid low-volume terms
• Build well-organized campaigns with sensible adgroups
• Test copy, landing pages, and match types
• If you are interested in profit, avoid content
• Segregate cost and sales from sales on brand terms from non-brand terms
• Obsess on site conversion—better conversion allows you to get more volume while maintaining profitability

Talk Back

OK. Your turn now. What’s your experience with the relative cost of clicks on Google, Yahoo, and Microsoft? What about the effectiveness of those clicks? We welcome your perspectives on your experience with the performance of the Big Three so far this year in the comments.

Alan Rimm-Kaufman leads the Rimm-Kaufman Group, a direct marketing services and consulting firm founded in 2003. The Paid Search column appears Tuesdays at Search Engine Land.