Google, Trust, Content Match, Placement Reports

A respected peer, Richard Ball, often writes about distribution fraud, and the waste of advertising funds on Google’s domain parking. I have some evidence to back up his claim that advertising on domain parks simply wastes advertiser funds.

This is a portion of a screenshot from an account with a content match targeted campaign. It shows impressions, clicks, the CPC and CPM, and the conversions. The data is organised by cost. This data includes the earliest advertising efforts in content match for this client - dating back to June, when Placement Reports first allow insight into where the adverts go. You need to check the optional box for “Special Placements” to see Domain Parks and Error Pages.

I’m not willing to reveal the names of the sites, but I can reveal that the number 2 row, with a spend of £186.73 are aggregated Domain Parks. The distribution of advertising to Domain Parks is not controllable by small advertisers. It is rumoured that large advertisers can tell Google that they don’t want Domain Parks and Error Pages included in content match. Over the next year, Domain Parks will dominate placement on Content Match for this client, and I’m pretty confident that there will be only the most random conversions from that placement, at a cost that dwarfs the most effective repeat conversion sites.

Google has recently started efforts to improve advertiser trust. One of the best and most simple things that Google can do to improve trust is to allow all advertisers to opt out of Domain Parks, Error Pages and other Special Placements. Of the few clients we have, that have online sales and are willing to share the conversion data with Google, none show any conversions from a Domain Park or Error Page. That’s simply not acceptable.

Further Analysis

Our CPA target for this client was in the range £50 to £60. The spend on the site in the first row was so fast that although this was the first site added to site exclusions, the spend was achieved in the first few days of the test. It still dominates the overall CPA of the content match campaign, despite not being served for more than 80% of the period shown in the report. From that extra piece of knowledge you can infer that we might have seen more than 60 million impressions for that single site in the report time range.

Other sites, with 100,000 impressions, are showing two conversions in this time range. The CPA metric is clearly achieved.

FWIW, this clients’ performance metric for our agency fee, was that we got a fraction of the spend. Technically, we should include this lousy site, as it will increase our revenue, even if the desired CPA isn’t met. I just can’t watch pointless spend while respecting myself. The client didn’t specify that we shouldn’t use site exclusion. So I excluded this site and told the client… I have limits to my mercenary zeal.

Even with the incentive of a fraction of spend, I can’t get a conversion volume of more than 20% for this client, from content match, of the volume achieved by keyword search. In other words, managing content match is as expensive as keyword search, but the revenue achievable is lower, as Content Match is now working.

What causes the wide difference in CTR and Conversion?

Intent.

The first site is a social networking site. People using this site are communicating with friends. They are not focused on buying anything. Arguably, you could claim that any advertising directed here should be intentionally oriented towards raising awareness, rather than selling goods and services. Arguably, I’d probably still claim that you are insane, unless you can give me long period impression data (that’s a topic for another article) or have something directly related to the communication passion.

The two sites that you can see with conversions - they are both about buying. One has focused on the general area that my client is interested in. The other is one of the current generation of free classified listing sites. Users on both sites are there because they want to buy something. Our adverts were enough to attract the attention of a potential buyer.

So, if you are trying to sell items, then you want sites that help you to sell. Small advertisers, who may not be trying to establish themselves as a major brand, don’t want awareness programs. They want sales, for the most part.

Content match merely offers that the site you reach will have concepts related to some of the keywords that interest you, somewhere on the page, even if the site has been deliberately built to attract users and high value adverts, without offering any real information. Here’s an example of a “Made For AdSense” site, discovered for another client:

You can see the insertion points where the site administrator has to insert a page specific term. Below this paragraph is a spot to select the RSS feeds that will evade the organic ranking page duplication checks (other MFA sites, even if they pick the same insertion text, are likely to select different blogs, meaning that the pages are not identical, so far as Google’s spiders can see).

Sites like this are just a drain on advertising budgets, likely candidates for click fraud rings and low quality resources for users. Extremely difficult for Google to get rid of these, because the evolutionary pressure exerted by PageRank algorithms causes pages like this to be developed - it is information rich (to a spider) and unique. Humans can tell that it is garbage. Bots… not yet, not easily, anyway. There are some ways… Another posting, perhaps!

Disproving Selling Capability

Let’s assume that you are advertising because you want to sell, rather than just contribute to Google’s payroll program and shareholder dividend. If so, you need to identify sites that have a cost efficient revenue potential. That usually means late phase, rather than research and awareness phases. If that last sentence meant nothing, go read up on the Buying Process and Buyer Models - these aren’t an engineering model of buying, but come close enough to representing how users think, that it should change how you think about advertising and what you communicate.

You could try to look at each site, as soon as they appear on a placement report. (Look at my article on gclid to understand why you need this report, rather than using referrer_info.) But reviewing sites manually has a cost, and the frequency of appearance may be low. That is, of the 572 sites that appear in the report, more than 550 have two or fewer clicks. So you could invest five minutes per site to check the quality… a dozen to twenty sites per hour. 160 sites per day if you get fast… almost four full days to check this campaign. At developed economy rates, you’re looking at around £500 to £2000 to get this done (assuming you pick up someone that doesn’t need training). Well, that cost would blow the ROAS out of the water. It’s about 25% of the spend of the campaign, and way over the margin that most clients offer.

Another way - you could either find out that a site is early phase or even in the wrong segment, by simply spending until you see that you couldn’t possibly make enough conversions to recover the cost of advertising. But that means that for 600 sites you have to spend around £150 per site (£90,000, or about $180,000) - and perhaps find that none of them generates a single extra conversion, much less repeatable conversions.

Alternatives - use labour from a developing economy. You need someone with good English language skills, and marketing intelligence. You’ll probably have to train them, over the phone. So factor in your time walking through fifty examples or so… a half day to a day, for a clued up low cost economy worker. Well, that puts the cost in a reasonable range, if you can find that worker. Try something like Elance or Guru.com.

Time Limited Campaigns

What if you have time limitations? If the campaign must be up now, and run for a week, or a month? Under those conditions, you can’t tell that a site is good or bad, unless it is very high volume. Look at the data for this campaign again - more than 80% of the sites had less than 2 clicks in 2 months, but accounted for 66% of the spend. You can’t know beforehand, which sites you’ll be exposed to.

So you need to have quality sites, to be sure that Content Match works for you. That’s not offered by Google. Google’s metrics for quality appear to be a diluted form of their editorial review for advertisers, mixed with uniqueness criteria from organic search. Both of these are breakable by publishers who simply want to generate revenue and really don’t care (because they don’t have to) about helping advertisers to find sales.

I can’t honestly see a sales reason to use content match for time limited campaigns.

Automation?

We tried looking at using information made available to advertisers to control content match spending.

It can’t work, unless you have high volume spend, or many, many clients that share data.

The problems to a technical solution:

About 30-40% of content match clicks do not have referrer_info. Identifying the source is impossible.

Training an AI to recognise signatures takes a few thousand examples. That’s many days of human assessment and months of collecting sites. And that’s an AI trained for one domain of knowledge. It may not work when applied to another product.

So only the largest advertisers could afford to deploy this type of solution, or a large number of cooperating small advertisers. I’ve had too many experiences with VC’s to want to set this business up and run it, and I’m building the Merjis business … I’ll gleefully be a consultant for your business, though. :)

Summary

Well, this was a little more waffly than usual. So I’ll try to get some solid recommendations:

Google should allow advertisers to choose whether to use Domain Parks, Error Pages and other Special Placements. These should be disabled by default, to improve advertiser trust.

Google should better police AdSense sites. Better AdSense publishers will mean that advertisers can afford higher AvCPC from those sites, so you may be able to justify additional human staff to assess AdSense sites.

Google should disable content match by default. Effective techniques for content match AdGroups are completely different from effective techniques for paid search AdGroups.

Google should improve the AdWords Learning Center to better reflect how to build effective Content Match AdGroups.

Advertisers should inspect Placement reports very frequently initially. High volume sites can mop up the advertising spend very quickly - but not always efficiently.

Advertisers should manually review higher volume and higher cost sites, as they approach around half the target CPA, to see if further spend is worthwhile. Otherwise, be ruthless and add the site the site exclusions in all campaigns.

Advertisers should consider using resources such as the Ads Black List.

Advertisers should probably avoid using content match for short period promotions (typical of most small businesses - they usually want to advertise when they have insufficient clients and turn off the advertising otherwise).

Weaknesses

I haven’t considered how the AdGroups are constructed. We don’t put content match together in the same way as keyword search. That affects results - in our experience, we get better conversion than if you simply follow Google’s AdWords Learning Center advice.

I haven’t considered here, latency - how long it takes for impressions and awareness to result in sales - and the Google 30 day cookie. Arguably, longer period cookies would show sales. As would Impression Tracking - not available from Google and probably a subject for another day.

I haven’t considered the speed with which new sites appear and old sites disappear. If Google is weeding out the worst AdSense sites and organic results are improving, then each month will see a new crop of sites. I have not considered the costs of that churn.

I haven’t considered the costs of the AdWords API in tracking and managing content match spending, to accompany automated improvement of results.

I haven’t explicitly considered whether the results would differ for drive-to-offline sites. That is, sites where the goal is to get users to pick the phone and call. You can use adverts with embedded phone numbers in content match advertising, for example…

Material Disclosure

We attempted to develop a product that would automatically improve content match. This article is partially based on some of that experience. We may yet come up with a way to reach the market with tools derived from that research - so some significant details have been held back from this article. However, I have consciously tried to make sure that what has been revealed is accurate and is not misleading. If your results differ, I’d be happy to discuss them with you!

Updates

2007-08-17 Added link to Google’s Domain Parking Service. Added link to my Wiki about the buying process. Minor tweaks to correct typos and improve readability, none intended to change the sense of the copy.