As Michael himself says, this can’t be taken too seriously – the identities and goals of testers are varied and it is perfectly possible that the system may be easy to game, resulting in perverse conclusions. I’m not taking the stats collected for other testers as very serious – but I am extremely interested in my own preferences and the implications for my clients.

If you do a lot of search and search comparison, you can actually recognise the results pretty well from the length of Titles and Description and the way in which links are presented. I tried to use the tool “fairly” – that is, selecting the one that presented what I thought was the best set of results. I used searches like:

• gclid
• search engine optimisation
• seo
• google money
• google scam
• msn adcenter
• yahoo!search marketing
• google adwords
• adcenter
• overture
• adwords
matt cutts
• danny sullivan

and, of course, the names and products of many of my clients, because I’m very familiar with their pages and competitors, so I can assess with some expertise.

When I started using the tool, there were about 11,000 searches so far conducted. At that point Google was in the lead at about 40%-45%. During my testing, I saw the count crank up radically as new people discovered it. By the time I finished, there were about 15,000 search tests – and Google was still in the lead.

What surprised me? The amount of spam in various list results. I’m aware of Matt Cutts and his teams role at Google and that there are similar teams at the other SEs, of course, and I’m aware of many types of spamming. I normally use Google, so seeing the misindexed results on low search volume keywords on Yahoo and Bing, where spam and accidental inclusions seem to be more prevalent, was a reminder of *why* I often use Google.

The search “google cash” was interesting, with Google apparently having removed a lot of scammers listings that were left in Bing and Yahoo results. I’ll guess this is largely manual by Google – and somewhat of a dereliction by Yahoo and Bing, IMO. “Best Search Result” should also mean “least likely to lead to a ripoff”, not just “best indexed” – care about the impact on users should be significant. If someone promotes drinking Drano as a cure for the common cold, it shouldn’t be number 1, even if heavily linked and referenced.

“gclid” – an arcane topic about AdWords automated advert tagging and web analytics – where the non-Google systems showed quite a few listings in which the content was not about gclid, but there were embedded links including gclid. I infer that document scanning is somewhat better at Google, or user feedback on appropriate results is given more weight.

When I used my own name (vanity search) and variations, I was hard put to identify which search engine to pick. They had substantially similar contents, changing the order of entries rather than what was shown.

Yahoo showed some search results much better than I expected. Bing, too, every so often shone clearly better results.

This parallel testing allows much better comparison than the usual sequential testing I’ve done. I’m hoping that the Blind Search tool will be kept running and improved. This may be a nascent business for Michael – identifying the best search engine for different geotargets, languages, categories of search, perhaps by crowd-voting – there’s a career in that :)

What did I personally observe? About 45% of my votes were in favour of Google, roughly equal for Yahoo and Bing at 25% and 30%; so I prefer Google almost twice as much as Yahoo, and about 50% more than Bing. That roughly fits with my preconceptions of what I thought about the search engines. Bing is major leap over Live. But still has too much spam, and too many irrelevant results *for the small and specialised sample of searches that I used*.

Enhancements I’d Like To See

Use of local indexes. The current tool obviously uses the “.com” indexes. So any UK-only clients fared fairly badly, sometimes being pipped by US-specific responses, when I know they top the UK results.

Record the percentage for each SE *by date* so that emerging trends in improvements by SEs can be observed. Also graph the number of searches by date, so that some level of reliability can be assigned, or use error bars on the vote chart.

A better URL. :)

Personal stats. I consider myself a “fair” tester. I want to see what I assess as the best – my personal record of B vs G vs Y.

I’d also like to see what searches were done, and whether there are patterns in the searches such that some of the SE’s excel at, say, travel, or automotive, or financial service results.

Really interesting tool, Michael – thanks!

Measurement

If you don’t have website statistics now, ask your website management company, or your staff, to install the free “Google Analytics” package. There are many free packages and you may already have one, but a big advantage of Google Analytics is that it integrates well with Google’s advertising systems and reasonably well with many other advertising systems.

Make sure to ask that “error pages” – the “404 page” and any “500 pages” – also have the Analytics code on them. You should investigate and repair your website urgently, or your advertising, if large numbers of visitors reach these dead and useless pages.

A big weakness of Google Analytics is that it doesn’t do quite the right things, until it has been configured. Configuring it can be difficult for beginners – so here’s a good book with detailed configuration guidance, for technically oriented staff – Brian Clifton’s “Advanced Web Metrics with Google Analytics“. If you are non-technical, I suggest that you flip through the book, note the types of information you can get with postits on each interesting page, then ask your technical staff to implement the techniques on the marked pages.

As a minimum, you should ask for Google Analytics “profiles” that give you “Error Pages”, “New Visitors” and “Returning Visitors”, with “filters” that eliminate traffic from your own staff, and ignore traffic on any sites that have stolen your web pages without changing the Analytics code.

If you expect to use AdWords or other paid sources of traffic, you should have each of those paid sources of visitors broken into at least two separate profiles for each – for example “AdWords Visitors – New” and “AdWords Visitors – Returning”.

You can also ask your staff for reports to be automatically emailed to you weekly, so you don’t have to login and work out how to get what you want. There’s also a free iGoogle Analytics “plugin” so you can have stats on your Google Home Page – if you use that.

Understanding Visitors and Your Site

You should be able to answer the following questions. If not, you or your staff probably need to develop an understanding of how your website and your analytics works.

• How many people visit your site? Per day, per week and per month? And when, during the day, do they visit?
• How do new users find your site – are they coming from directories or from search engines or other sites?
• What search queries are visitors asking the search engines?
• Are visitors bouncing – leaving the site as soon as they arrive – in large numbers?
• How long do visitors stay on the site?
• What are the most and least popular pages with users – the ones most and least visited and the ones that users stay on longest and shortest? Is it what you expected?

You should be getting most new visitors from search engines – is that happening?

You probably should be getting return visits from customers – are you getting returning visitors?

If the pages for your best products and services aren’t getting visitor attention – why not?

See if you can improve your performance, by making changes and watching the statistics.

Customer Engagement

Think about the searches that prospective customers should be using to find your business. When you talk to them directly, what do they ask you? Not in your professional jargon, but with their use of language?

Now go to the “AdWords keyword ideas” tool. Put in your user’s important phrases and ask for suggestions. These are usually words and phrases that your site should be using and may even be used as the names or titles of web pages on your site. If you use language differently from your likely clients, you are less likely to attract them to your site and keep them engaged with your offers.

When you have a conversation with a client, lead or prospect, think about whether your website clearly describes the same things in the same way. If your website message is different from what you tell people, try to make the web site reflect what you are doing and saying.

If someone lands on the web page for a product or service, without going through the home page, would they know what you are offering and how to get hold of you? If you have a Unique Selling Point on the website – is it only shown on the home page? People who use search typically jump deep into your website, and will miss messages that appear only on the home page. So if you offer free shipping, or a free first order, or whatever it is – make sure that offer is visible on all likely landing pages.

Make sure that you have a call to action on each solution page, with prominent contact information. We’ve come across a surprising number of sites without phone numbers, inquiry forms or addresses… At the point when visitors should have decided they need to talk with you or place an order – make sure that they can do so. A rough rule is that “users will do what you ask them to” – so ask them to buy.

Before you make changes to the web site, here’s one more review step… does your website tell customers about you, or do you talk about them and their problems? Your website should demonstrate that your first priority is solving your customer’s problems. A swirling-in logo of your company, or the letters of the company name coalescing from a cloud, may look pretty neat, but does it answer a prospects’ first question – “do these guys have anything that solves my problem?” Focus on what your prospects need to know to become clients, and don’t delude yourself that they want to sit through the same 40 second monologue from you at the start of each visit, going on about how great your business is… They won’t like it. Really.

Get to the point quickly, and describe the problems that your customers used to have, and what you do to fix them. If you solve several problems, offer one web page for each, and use the home page to reference the different solutions. Each solution page should link to related pages, so if visitors aren’t in the right place, they can easily find where they should be.

Being Found – Search Engines and Other Visitor Sources

There are two main ways to appear in search engines. For the main search results, you use “Search Engine Optimisation”. Do read free advice from major search engines about selecting a SEO business before you pay for services.

• Google’s SEO advice
• Yahoo SME – SEO advice

Paid search advertising is the alternative. Smaller businesses shouldn’t just pick the market leader, Google AdWords, but consider the smaller players, MSN AdCenter and Yahoo!Search Marketing. There’s usually less competition, and the smaller audience doesn’t make much difference to the volume of visitors when the budget is small. They also attract slightly different audiences – so failure on one of them doesn’t mean they’ll all fail.

The AdWords Help Experts blog has a video showing how to set up advertising on Google AdWords. Do *NOT* use Google’s “Starter Edition” – it is poorly designed for new advertisers. Exercise great care – Google does not help you to control costs and improve ROI – their eye is definitely on their revenue, not yours. Do not expect to carry to AdWords, the same trust that you have in Google’s search engine, or you will lose money.

SEO is slower to return results, but often at a lower cost. If your website is pure Flash, gaining visitors through SEO will be harder and more expensive. Paid search advertising is faster, and it is easier to identify the return on your investment. We recommend that if you can, use Paid Search first, to quickly find which keywords & pages work, and then use SEO to target them.

You can also appear in business listings, such as Google Maps. If you sell products, you can list them, for free, with Google Product Search. You may have images and pictures that interest people, and you can get listed for those. You may make videos and publicise those for Video search. There are planty of ways for a little imagination and creativity to get your your business noticed.

• Qype – expanding worldwide, offers reviews of local resources and interaction with local and interested members of the community. See the Merjis listing showing that we offer internet marketing in Bedford. Accepts adverts, too.
• Gumtree – classified listing directory, accepts adverts
• Craigslist – massively popular classifieds
• eBay Stores – if you have products to sell, eBay has buyers.
• Yahoo Directory – paid listing, human moderated, the original large scale internet directory
• Open Directory Project – free listing, human moderated and a contributor to the Google Directory.
• LinkedIn – link to professional colleagues – see my profile for example
• Twitter – status feeds for customers, personal conversations, all sorts, Twitter is flexible
• FaceBook
• MySpace

Do *NOT* pay for “Search Engine Submission” tools – if your site is properly linked to the rest of the web using the services and sites that I’ve given you, the major search engines will find you anyway. Do *NOT* list your site on free and unmoderated directory.

Internal Response

Leads received from the website are often left in email inboxes for a few days. Web leads are as urgent as a phone call. Make sure you and your staff respond promptly.

Finally, you probably have news about your business sector. Can you write it up, even if in only a few short sentences? A business blog with short, or even long articles like this one, can provide a way to communicate with new leads to show them how you help your clients, and it helps your SEO efforts. There’s lots of free blogging software that you could host on your own site, or you can use one of the public blogging services.

About Us

The author, Jeremy Chatfield, is a Google-recognised Top Contributor in the AdWords Help Forum and manages the AdWords Help Experts blog; he is a Google AdWords Qualified Individual. Merjis, in Bedford i-Lab (+44 1234 834760), helps clients to improve their web site performance.

Material Disclosures

The link to the Google Analytics book above is an “affiliate” link. If you click and buy, we may make a small amount of money. It’s a demonstration of another common technique for attracting visitors (using “affiliate programs”) and may be a way for you to additionally monetise your site.

None of the other links are affiliate links, even when an affiliate program exists for that organisation.

This article is an expanded version of an advertorial in our local business newspapers, “Business to Business” and “Business MK“.

Well, when I say “wrong”, I suppose I really mean “less than complete”. I think the reason is quite subtle – and not explicitly recognised in the white paper. That doesn’t mean that they don’t know, and it doesn’t mean that they haven’t worked it out… it just got me thinking about some work we’ve done. That, in turn, means a brief look at recent history.

Google’s dominance of paid search, especially in the UK, means that many users won’t be aware of the original innovator in the arena. “First mover advantage” (yeah, that’s irony for you) went to GoTo.com, who eventually became “Overture” and then “Yahoo!Search Marketing” (Y!SM).

Bidding Models

GoTo’s original model for an advert auction (pre-Panama) was that bids would be compared and the longest standing, highest bid, would appear first. A simple auction, easily understood. Also, an auction in which automated bidding would be hugely important. Why so? Because the majority of clicks go the highest ranked listing, but the value of the listing would depend on conversion (or affiliate reward). Getting your advert to the right position would be a matter of fast paced juggling.

Automation can help reduce costs and mistakes in an auction like that, for all advertisers. Tactics, such as choosing when to bid, and how frequently, become important. And little issues like bidding gaps become hugely significant – because you pay 0.01 more than the advert below you. So if you can find a spot in which the two adverts above you are paying just 0.01 more, but you have a 0.05 spread to the advert below you, you pay much less than your higher competitors, which may justify the lower click volume on a lower position. Lots of calculations to be performed quickly. Automation, absolutely useful.

GoTo’s model for search revenue was also quite different from that for Google. Initially, GoTo had to partner with higher volume third party sites. This lead to the acquisition by Yahoo. Google’s first advertising was on their own site, and they could then sell the higher value adverts to search partners who were already prepared, by GoTo, to handle externally supplied advert inventory. This results in an apparent difference in results between Google and Yahoo. Where Google impressions may soar markedly with position, GoTo’s already broad spread across third party properties meant that the impression volume was not so accelerated by position. That was also subtly influenced by Google’s auction mechanism – covered elsewhere on this blog – which offers absolutely the highest returned value for every advert space used.

As an early advertising system, GoTo offered a single advert. It costs the publisher cold hard cash to vet adverts. So running a single advert for each advertiser for each keyword, reduces costs. Important for a startup. However, ask anyone in direct marketing what their most powerful technique is; it is the parallel A/B test. The ability to test copy for effectiveness can transform a failing campaign into a successful campaign, sometimes with the most subtle of changes. Google’s AdWords allows alternate adverts – dozens of them for massively parallel A/B/C/…. testing. This changes the way that the bidding system works – because you need to separate the behaviour for each creative that is used.

Budgets and Bids

The way in which budgets are set and satisfied also results in some intriguing differences between the bidding systems. In GoTo’s model, it appeared that adverts participated in all auctions – even if they did not have the budget to do so. Google’s model appears to run the auction only between participants with enough budget to be able to offer an advert (there’s evidence for this that I may explain in another article – it’s moderately complex reasoning and somewhat off-topic for this thread).

Something really important to carry away is that CPM and CPC (cost per thousand impressions and cost per click) models are interchangeable – if you can make some assumptions, or some measurements. For example, if you can assume a 2.5% CTR. If you do, then for every thousand impressions you’ll see 25 clicks. If you know the CPM, you can then infer the CPC. Google makes assumptions like this, all over the place – you can even infer what the assumed values are with some experiments – I may write those up. It was pretty cute working out what Google’s hidden numbers are :)

Why CPC?

Psychologically, advertisers are happier with CPC presentations of cost. For example, small advertisers may not want to buy 1,000 impressions. When presenting to board level decision makers, you can tell them that you are buying clicks, not impressions which may or may not turn into business. Clicks might turn into business and definitely turn up on the web site – and if they don’t, you can challenge the publishing system for fraud. But impressions – a lot softer… harder to explain why you don’t get clicks or conversions. So Google’s system has gone for CPC presentation – but I’m pretty certain that the auction itself runs as a CPM model.

Broad Match – Broader Than You Might Expect

Finally, Google emphasises broad match (it is the default, it is touted in training, and gives Google control over advertisement appearances for arbitrary searches – which Google can use to its’ advantage, smoothes the landscape for bidding, etc). The broad match auction allows Google to spread the value over whatever it deems relevant. Bid on your company name in broad match, and if you’ve bid enough, your advert will appear on searches for competitor names. No one who has done any courses in Marketing Communications should be happy at this point – MarCom tells you that you need to focus an advert on a competitors specific weakness, relative to your business, if you want to capture someone searching for a competitor.

These differences mean that the nature of the auction on Google is less about fast paced changes – for most advertisers, who do not have stock that ages, and in turn leads to significant reasons for most advertisers to avoid automation or to reduce emphasis on manipulating bids as a technique. The focus for success on Google shifts from bidding strategy and into ad copy, keywords, match types and landing pages. Arguably, this is the right emphasis anyway. The nature of GoTo’s auction system drove attention to bidding, and their inability to offer parallel A/B testing of adverts, demoted interest in improving CTR.

Aggregated reports also make a huge difference, too. GoTo’s model was pretty simple. You could infer, for a given time period, what the position was, and the paid price, and the impressions and the clicks. For Google, we have a variable delay time (between around 15 minutes and 3 hours, with a minimum of 45 minutes for reliable reporting), and aggregation – you can’t get to results for the last time period, for periods of less than a day. This means that to infer what has happened late in the day, you need to know what happened earlier in the day – because you’ll subtract all of those from the aggregated result. However, this leads to a blurring of values late in the day. By 10pm, you’ve had 95% or more of the clicks that you’ll get, and so changes made to bidding have a relatively tiny effect on the numbers that are reported – things like the averaged position over the day are more heavily influenced by the earlier days’ higher activity. This means that the effective measurement times to get insightful data are reduced – you can’t make sensible decisions about changes when 90% of your results have been collected.

Seasonality

What really cripples automation for many small advertisers, and can cause some interesting problems for large advertisers, is seasonality. To effectively select the right price every Christmas, you need to have a model of the change in impressions for the previous Christmas. Building up that model is expensive. Retailers carry this in their heads, but it needs to be expressed to the model before it reaches the date… otherwise you find that the model is no longer predictive, but reactive, and always behind the curve.

Take, for example, the classic New Year resolution to stop smoking. Keywords relating to stopping smoking have a low volume of search over the year. Starting just before the New Year, it ramps, and declines by the end of January, reaching a peak that is vastly higher than the normal monthly run – just for this period. If you use a predictive model, it’ll keep projecting that tomorrow will be like last month… it’s had 11 months of being right. It will typically project that tomorrow, the volume will be down. And of course, it rises again and again through January. So the algorithm spends all its’ time chasing the rising curve. How it reacts to the declining interest is also problematic – with no history and a short time planning horizon. Most algorithms will be dominated by “tomorrow will be like today, with minor changes” – so a declining curve has them retaining bids for too long as the market declines.

Because the auction is really for CPM, and broad match means that bids are compared between wildly different searches, the whole effect is to smooth the bidding landscape and slow the variation of bidding changes. Bidding on Google is pretty stable, after the first frantic period of working out th messages and building a CTR history.

Noise

Complicating all of this, is that the Internet is a “noisy” signalling environment. Bursts of searches may reflect real changes of user interest, or just statistical variation that means nothing. Designing algorithms that take care of high noise environments. with unknown seasonality, is pretty tricky. Getting retailers to describe their expected annual search impressions is also sometimes and exercise in futility – because over- and under-statements cause problems that can be worse than having a near-Bayesian model.

None of that is the defect

The missing point is that each keyword has several built-in characteristics. Different searches have a different implication, that predisposes the search user to specific types of advert. For example, a search for a brand name, especially a business’ own name, usually indicates a high intent to hear from that business, not a competitor. The CTR for a competitor will be tiny, and the CTR for the business will be high. This leads to an inverted paid price. That is, the bidding landscape will tend to have the lowest average Cost Per Click at Position 1. At other positions, the AvCPC tends to be higher, as the advert is under competitive pressure.

Even worse, if the product being sold is a niche, the Average CPC becomes somewhat less significant – because it is important that the higher adverts weed out the segments that you can’t afford to handle. The optimum position is then dictated by competition messages – inferred by increasing conversion rates in lower positions.

Getting the bid right can’t rely on a simple Newton-Raphson optimisation, but has to look for non-local minima. Programming systems to find non-local minima is tougher than most businesses can handle. It is possible that Sayu.co.uk has handled this – but a number of the largest automated bidding systems, especially those popular with large agencies, seem to lack any serious bidding model.

When the non-local optimimum happens to coincide with the traffic maximum, and automated bidding systems fail to identify that, it is a pretty poor failing.

Anyway – thanks to Gerda Arts and Sayu.co.uk – it’s good to have someone explaining what they do in more detail than the handwaving at bigger competitors.

Ranking By CPM, Not By Bid

If Advertiser-1 has a CTR of 10% and pays (that is “pays”, not “bids”) $0.10 per click, then they have an effective CPM of 1000 impressions times 10% CTR times $0.10 Average Cost Per Click = $10.00.

If Advertiser-2 has a CTR of 1% and pays $0.99 per click, they have an effective CPM of 1000 times 1% times $0.99 = $9.90 effective CPM.

If you were running a search engine advertising company and wanted to make the most money, would you make Advertiser-1 your highest position advert, or Advertiser-2?

You’d pick, I hope, Advertiser-1.

This simple calculation demonstrates why a simple auction based on ranking bids alone fails to deliver the best revenue for the search engine. It also deeply satisfies a need for search engines… clicks. A “good page” for a search engine, in 1997, would have been one that got the most clicks – the most results were interesting to users. These days the value is a little more complex to calculate – a search engine should be looking for satisfaction with a search results page, not just clicks; if the number one result satisfies most users, and the rest of the search results satisfy all other users with a single click on just one of the other results, it has done a good job… but working out the balance between lots of clicks and inferring satisfaction is much harder than looking at percentages of clicks. Hence the AdWords Quality Score… more on that later.

Example Strategies

I’m going to describe three advertising systems, and I’ll call them Asearch, Bsearch and Csearch. I genuinely don’t have any intention of mapping these to existing paid search businesses – I’m using them to explore how apparently small changes in the mechanics of running a nominally similar auction can result in revenue changes – or, from the advertisers perspective, ROI changes.

All the businesses use the same basic model:

• Adverts are submitted with keywords and bids
• For each impression, each keyword can show up in no more than ten advert slots per page
• Adverts are ranked by revenue – that is, they are ranked by their effective CPM, not their bid

Budgets

I’m going to throw in another real world limitation – budgets. Advertisers like budgets. Budgets allow planned spending and avoid unpleasant month end surprises, or running out of cash on day 15 of the advertising month.

Hurrah for budgets.

However, budgets are interesting for paid search companies – they affect the auction in some odd ways.

Let me also assume a few advertisers. 1advertiser has a high budget. If there was a way to get someone extra to their site through search, they’d take that chance and so they bid 120% of the daily actual spend. 2advertiser has a low budget – they can only afford to spend 10% of the budget that the search engines recommend to them, and their strategy is to appear as Number 2 when they do appear. 3advertiser is B2B and sees a wide variation in search volume and conversions – weekends are pretty much dead for sales, there’s a lot of research activity on a Monday, and Thursday tends to have peak sales volumes. They bid to Position 3 with a 100% budget. These are simplified bidding strategies, to illustrate what happens. I’m not claiming that these are optimal bidding strategies or even realistic companies.

Asearch, Bsearch and Csearch look at budgets and all of them apportion 2advertiser (the 10% of estimated budget guy) a rate of 10% for showing their advert. That is, the advert only appears on one search in 10, on each of the three search engines. However, the SEs have a different way of running the auction:

• Asearch includes all advertisers in all auctions, whether they have the budget to actually appear, or not. The positional calculation is done at the beginning of the day.
• Bsearch includes all advertisers in all auctions regardless of the budget. The positional calculation is done for each impression.
• Csearch only includes advertisers with the budget to show an advert. The positional calculation is done for each impression.

Assuming that all systems have the same number of advertisers (same competition levels) and the same type of audience, and uses the same system to match search query and keyword, which of these small changes to the auction will have an effect? I believe that Bsearch will make the most money from advertisers on a pure economics basis. Why?

Who makes the most money?

Csearch apparently leaves money on the table. Each non-participating advertiser means that the values at the top of the auction are lower. If you only have 10 advertisers, and one of them drops out, then the remaining 9 advertisers will see reduced competition and on average will pay less per click than if all ten advertisers are jockeying for position. Although the auction mechanism is the same, especially with large numbers of advertisers using budgets of less than 100%, each auction is a new activity and as a result the cost per click varies greatly, depending on the exact participants in each auction. It’s actually slightly more complex than that. I might cover the effect of varying volumes on search on paid price granularity in yet another article.

Which of Asearch and Bsearch makes the most? In pure economics terms, I believe that it is Bsearch – because the revenue is optimised over shorter time periods. That is, if there are transitory reasons to prefer one advert over another, then Bsearch will respond faster to changing market place needs. However, Asearch still does better than Csearch – because the larger number of advertisers in each auction should maintain the price higher. For Asearch and Bsearch, just because 2advertiser doesn’t have the funds to pay for the advert, doesn’t mean that they aren’t in virtual position 2 and that 1advertiser should be paying a penny more than 2advertiser. Effectively the auction is run for more than 10 places, and then any advertiser that doesn’t have budget to participate, simply doesn’t have their advert shown.

Easiest to build?

In implementation terms, Csearch is the simplest. At the beginning of the day, Csearch calculates the ranking for all advertisers, and only recalculates (just once) if a participant in the auction for that keyword has changed their bid. This means that advertisers can be given a reliable price and a reliable position – so long as no other advertiser changes their response. It makes for a very predictable system. However, advertisers pay a penny more than the costs of non-participating advertisers. If this system is used, a competitive strategy would be to bid on competitor keywords, but with a miniscule budget and a high bid; this would help drive up the auction value at low expense to the advertiser, costing competitors quite a chunk of change. 1advertiser would see a consistent paid price, no matter whether 2advertiser’s adverts show or not.

Maximal Revenue

Bsearch conducts an auction for each impression, but also assumes that all advertisers are present. This allows Bsearch to optimise revenue in each cycle, which should lead to maximimum revenue. Again, the “absent” advertisers drive up the Average Cost Per Click for those with the budget to actually appear. 1advertiser would see a consistent paid price, whether 2advertiser appears or not. The most obvious way to diagnose this rather than the Csearch strategy is what happens when the system moves from using an assumed CTR to using a measured CTR – Csearch will revise the position day by day, but Bsearch will change position during the day, especially when the advert is fairly new. Um – that really needs another lengthy article. Just take it from me for now that there are some diagnostics for inspecting which models may be being used… I might give away some inference tricks in a future article.

Fairest Auction

Asearch conducts a “fair” auction. That is, the cost of a click is only calculated by reference to participating advertisers. If an advertiser doesn’t have the budget, then that advertisers bid is not part of the auction. Asearch will get the lowest revenue, but offers an auction that most advertisers will consider fair. When 2advertiser is not showing, 1advertiser will see lower competitive prices – so the average cost per click will decrease.

We can make some predictions about the behaviour of the advertising system, as seen by advertisers. However, like anything involving reality, it gets more complex as you look deeper…

CTR Assumptions And Budgets

I rather glossed over the effect of budgets with glib statements about “2advertiser only has 10% of the budget” and went on to say that that meant they’d only see 1 in 10 impressions. It is, of course, more complex.

The Click Through Rate estimate affects the spend, especially before the advertising has developed an account history. Create an advert with a CTR that beats 2.5%, and you can go over-budget. This happens because in order to estimate the 10% threshold, the SE must assume a CTR. If you beat the CTR, then the system keeps delivering impressions at the 10% rate – but with a higher CTR, you get more clicks per impression. The result is that your spend is higher than the predicted value. You go over budget.

When you go over budget like this, it isn’t your fault – the system delivered impressions based on a poor estimate of the click through rate. Reputable search engines will reimburse you for their failures to correctly estimate your spend. For example, Google will refund overspend built up this way. That’s good news for the advertiser, as they get some free clicks and perhaps even some conversion on the first day or so of overspend.

It is even possible (using tricks I won’t go into here) to manipulate Google into fairly consistently underestimating, so you get lots of days of partially free clicks. Google, of course, doesn’t like this. You’ll be dicing with a permanent ban if you use techniques that might result in heavy overspending, consistently.

Assumptions

It is possible to examine what assumptions the SE’s make. Experiments can be done to probe what an SE considers to be a normal CTR, when it makes assumptions. This appears to be increasingly harder to do… perhaps because the SE’s are now getting smarter. A few years ago, the assumed CTR was the same for all keywords, but it shouldn’t be. Different keywords in different market segment, reaching people at different stages in the buying process, will have different basal CTR’s. And that means that the CTR assumption should be different for different keywords. I see some signs that that the SE’s may be starting to do this per-keyword assignment of assumed CTR…

Going Over Budget Costs the SE

If the assumption is of an average CTR, then, on average, you haven’t lost money… Um. Wrong. When the SE overestimates the CTR, then the advertiser sees their position gets worse when the CTR history kicks in. If the CTR is better than the estimate, you get more clicks on that day, exceeding your budget, meaning that you may get free clicks and the search engine has lost revenue by feeding you inventory that it could have used for revenue generation on someone who had remaining budget. So it is vital that they decide how to adjust the advert flow rate to get the best revenue – that is, so that advertisers budgets are exhausted, when and only when the impression inventory is exhausted.

So how do search engines defend themselves from giving away clicks? Especially, how to calculate the rate of impression serving for the whole day, at the beginning of the day, and not adjusting the rate, throughout the day? That is, calculating for each advertiser, dynamically, whether to show the next impression, or not, is expensive – more expensive than deciding once, at the start of the day, what the flow rate should be.

Note that this is crucially dependent on the costs of that calculation. Deploying more hardware, especially when you consider Moore’s law, allows the decision to be made dynamic, rather than once per day.

One of the first exercises is to excessively throttle those with low budgets. If you can use up all the inventory with advertisers who offer a higher budget than needed, then you avoid giving away free clicks to advertisers who have a limited spend.

For example, if 2advertiser (the 10% of estimated spend guys) are advertising, then a search engine should allow for some variability in search user response. If the SE allows, for example, a variation of 25%, then they should only allow the advertiser 80% of the impressions that would normally be expected to exhaust the budget. Otherwise there is a risk of selling the inventory of impressions for free. If the budget is, OTOH, much larger than the expected spend, then offer every single possible impression. See Google’s policies on spending over the budget.

If advertisers can detect that the flow is throttled by the search engine, then they should change the tactical response to setting the budget… So if you given an estimate of spend of $100 per day, and you consistently spend at the $100/day mark, and you suspect that the SE might throttle based on budgets. offer $120, and see if you get more impressions per day, than when you offer $100 as a budget.

Summary

I hope that I’ve convinced you that the search engines can make different choices for how to operate an auction. Even if the auction appears to use the same rules, the details of those rules can affect the revenue returned to the search engine, and consequently the ROI perceived by the advertiser.

It should also be clear that larger auctions generally act to raise the price. This has an implication on what seems like a purely technical issue – how search queries are matched. As this discussion *should* have made clear, the apparently technical choice of which search queries map to which keywords, affects the revenue from the auction. Search Query Matching is as much an exercise in economics satisficing, as it is in technical excellence.

Budgets are not simple – they affect ad flow rate and may, depending on hidden decisions by the search engines, reduce the rate of adverts even for an advertiser intending to achieve full flow advertising. This is because search engines want to strike a balance between offering inventory for free and offering advertisements throughout the day to exhaust an advertisers’ budget. Leaving unspent advertiser budgets is as bad as delivering free inventory.

Updates

2008-09-24 Typo corrections.
2008-10-05 Typo corrections; changes for clarity.
2008-10-23 More typo corrections. Added link to Google policy on budget overruns.

Clients that use IIS do seem to suffer from the strangest problems. I kept finding indexed pages, some ranking absurdly highly, with an path infix like (X(kjhgkjfkuyfku)). In other words, if I had a page like “http://merjis.com/login”, I’d find exactly the same content as “http://merjis.com/(X(stuff))/login, making it look as if I had lots of duplicate pages. Appended parameters as a session ID, I can understand. There’s even ways to cope with those.

The format of the infix was quite rigid – 23 characters starting and ending with a parenthesis, and embedding at least one more parenthetical group. I think the regex “\/\([A-Z]\([A-Za-z0-9]*\)\)\/” will match every example that I’ve seen. Oddly, you could navigate to them using a web browser and they worked, but even spidering the site and grepping the resulting mirror failed to show these strange paths in the HTML… so how did they get there?

Cookieless Session Tracking

This looks like an effort by Microsoft to allow tracking of people that don’t want to be tracked, or who might have, for example, an office based transparent proxy that blocks cookies. AFAICS, when IIS detects that a user doesn’t permit cookies, it starts sending unique paths. The use of the “referer_info” (sic) field allows tracking a single user across the site, looking at where they were and where they’ve gone to. The positive benefit is that users can be given access to stateful services (like logging in) without needing a cookie. It is, I think, effectively a session cookie, rather than a permanent cookie, as it can’t recognise you from previous sessions – your starting point will be the same as any other visitor.

This session tracking seems pretty barking to me. If a user prevents you from serving and saving a session cookie, and you then create a trackable session by using session keys in the URL, you have just violated the privacy that the user requested. If users try to use a use a resource that needs stateful dialogue – something that remembers whether you were logged in, for example – then I think that if the user disables cookies it is perfectly reasonable to tell them that this resource won’t work for any areas requiring an authenticated session, though the public areas may be freely roamed. If they want privacy in the session, then they can’t have access to anything that requires remembering from page view to page view, who they are.

Cookies and Search Engine Spiders

Robots don’t hold cookies. If faced with an IIS server configured to handle cookieless users, spiders end up being delivered with a false directory structure. Every new visit generates a new directory structure. You get excessive crawling and pages appear in the index as unique, when they are really duplicates caused by server behaviour.

Now, this too seems like a petty piece of madness. Given that some users want to maintain privacy to the extent that they do not even want session cookies, and this number is small on web servers that offer services involving a log in or other identification service, why would you make it more difficult for search engines to spider the site? I believe that you can get more users to a properly indexed site, than the number you’d lose from failing to handle uncookied users (unless you offer a specific service for the uncookied, of course).

Robots.txt to the rescue

Fortunately the workround for this problem seems small. In Robots.txt, add a line:

Disallow: *(


For all the spiders that I care about, that seems to prevent crawling the special tracking URL. The consequences of that… well, I’m not convinced it is entirely good. But it does stop silly URLs from being indexed after a single crawl.

Technically, this line says “disallow spidering for a URL starting with anything that has a ‘(‘ in it somewhere”. Although this client seems mostly to suffer from the infix immediately after the domain name, reading around the web suggests that the infix could be put at any directory level slash. Otherwise “Disallow: /(” would work and avoids the failure possibilities of the wildcard.

Spidering Improvements?

I can’t see why spiders should behave like this.

Having grabbed a server identity from the headers and behaviour, it should be possible to then strip out the session tracking from the path. I can’t currently think of a reason to *not* do this – unless you were really trying to sneakily discourage IIS administrators from using this tracking method.

I can think of anti-competitive ways to use the cookieless IIS behaviour. For example, find the user sitemap, point to it and let spiders follow the unique links to every path – then you can explode the crawl on each site by doubling it. And it accumulates – because every session appears to be valid for a long time, allowing repeat crawls and generating new unique paths with every reference.

You’d have thought that this explosion was worth trapping and stopping in the “fast crawl” discovery bots, so that the slower inspection spiders wouldn’t add these redundant pages to the indexes. The presence of these idiot links in link reports, and the multiple crawling of them by various bots, suggests that spiders are still stupid.

Webmasters?

Well, if you can’t avoid running IIS, take a good hard look at your web servers log files. Do you really have a useful volume of search from real users who are cookieless or have you just ramped up bandwidth so that bots can crawl redundant pages without adding any revenue? Is the majority usage of this feature an escalating collection of spiders? Might you get more users if link love wasn’t being directed to duplicate pages on redundant paths?

Put in the crawl reduction “Disallow: *(” and possibly a “Disallow: /(” lines to your Robots.txt and verify them with the Search Engine webmaster tools Robots.txt checkers. At least you’ll be focusing crawl on pages that should be indexed.

Ideally, turn off the cookieless mode. AFAICS, it is a breach of the privacy rights that users were trying to assert. If you offer a service that needs state, then when you detect that you can’t cookie, offer an apology that parts of the site are unusable. IMO, that’s perfectly acceptable for both users and designers and won’t lose any business that you couldn’t have gained.

This article takes a quick scoot over the major similarities and differences in interfaces. I expect to dig into specifics in more detailed individual articles. I intend that the next detailed posting will be about how the two systems handle integration with web analytics for tracking and conversion handling, and how easy or difficult the systems make it to track search behaviour.

Identifying the audience differences is somewhat difficult to do, as this may expose too much about our clients. I’ve been wondering about how to discuss the differences in audience behavior, and if I come up with a solution that preserves confidentiality, I’ll find a way to discuss it.

Similarities

AdWords and Y!SM share a lot of features, where previously there was a radical difference of approach. Rather than spend a long time dwelling on what Overture used to do, let’s look at how the new Y!SM compares:

• Grouping Keywords in AdGroups
• Multiple adverts per keyword – supporting parallel A/B testing
• Basic match and extended match (AdWords’ Phrase Match is closest to Y!SM’s Standard Match – both get wild with Broad Match/Advanced Match)
• Dynamic Keyword Insertion
• Conversion Tracking
• Generalised Second Price Auction – Your bid and your advert effectiveness combine for placement.
• An editorial review process – assurance to publishers of quality adverts (revenue and “decency”)
• Separate Display and Destination URLs
• Default subscription to the contextual match service (AdSense or Yahoo!Publisher network)
• Text Adverts for paid search results – graphical adverts are featured on contextual match
• Bulk Upload mechanism (AdWords Editor vs Y!SMs direct import of spreadsheets)
• API – both offer a way to write software that helps or manages the account, but Google makes it easier to access
• Built in click-fraud detection – both deliver billed clicks but neither account for unbilled clicks

Display URL Improvements

Y!SM used to only allow a Destination URL and used an uncertain method to identify the proper Display URL. This was most often a problem when using a third party redirector for advert management – stuff like the Atlas DMT package, or Nedstat’s Web Analytics would sometimes show that the advert was for “Atdmt.com” or “www.sitestat.de”. The new platform removes that problem.

Policies on the use of the Display URL appear to be broadly similar. For example, neither like Duplicate Serving, having two adverts for the same advertised site.

Advert Formats

Y!SM has leaped ahead of Google in certain fronts. The Advert Generator and the wider range of shown formats (short and long form description) allow easier creation and testing of copy, and give opportunities for more publishers to present something that makes sense in their display context. Although Google doesn’t make it clear when composing adverts, they do represent adverts in a variety of formats – Y!SM’s more explicit handling of adverts better supports understanding what the advert will look like.

For example, Google’s Gmail shows AdWords adverts. When using webclips, it will show them in the format “Headline, Display URL, Desc Line 1, Desc Line 2″, as shown above. Advertisers who rely on the headline being followed by the first description line, will get a less readable advert. Additionally some AdSense publishers cut adverts back to just the Headline, or strip the Display URL. Possibly they fear that viewers might key the URL in directly? Possibly it is for space saving – more ads per inch? I don’t really know the reasons to do this. Y!SM’s advert composition and generator is arguably better for making more effective adverts in all the formats that will be published.

Google’s Faster Rate Of Evolution

Google still scores heavily on certain fronts, though:

• Google Campaigns offer country and smaller geotargeting – Y!SM still appears to need multiple accounts for multiple countries
• Google Campaigns offer language targeting – I can’t see that in Y!SM, but I haven’t really used Yahoo outside English speaking countries
• Google very recently added more control over the publishing networks, via the UI (e.g. drop Domain Parks, Error Pages, photo sharing sites, etc)
• Google spend is set at the level of budgets per campaign – Y!SM’s still seems modelled on a budget per account – I use this Google feature for all sorts of purposes.

Summary

Yahoo!Search Marketing, the original pioneer in paid search marketing, is now clearly trailing Google in both global market share and the sophistication of control that can be exercised by the expert marketeer. The newer interface has done a lot to catch up, and in some respects, offers better presentation – though I can’t currently identify better control.

Basic functions are pretty much a mirror of each other. Keywords are assigned to AdGroups, AdGroups trigger a pool of Adverts. Advert selection may be optimised by the search engine. Bidding seems to be based on the Generalised Second Price Auction. This will get some in-depth attention in a later article, as I think there’s a fundamental difference in approach by Google and Yahoo!Search Marketing for the mechanism of the auction, which affects the value the auction returns. I think this may be one place where Google might want to copy Yahoo!Search Marketing’s lead, but would probably involve a significant re-work of the entire auction mechanism.

“Part Dieux”? This is part deux (“Two”, in French) of the series, and is a deliberate pun on “Dieu” (“God” in French). Why this pun is amusing is left as an exercise for the reader.

So, let’s suppose that Google takes a cold hard look at the competitive landscape, much as they did with the recent 700MHz spectrum auction. From what I’ve been reading, Google bid $3.7B and would have been disappointed if they’d won – but it drove up the price and forced a competitor to do something that Google wanted. Might Google be equally subtle with MS and Yahoo!?

How about this? If Google annoys a substantial fraction, especially of small and niche advertisers (lots of them, relatively small revenues in Google’s terms, no profound feelings of mutuality on either side), what is their rational response? Defection? To where? Yahoo! and MSN AdCenter, and possibly to some of the third tier players.

If Google loses, say, 10% of the revenue expected for this Quarter, it makes a massive increase in the revenue potential for Yahoo!, and possibly for MSN. Google’s loss of 10%, makes, if I calculate this correctly, about a 20%+ change for Yahoo! (assuming that online advertising budgets are not reduced, just redirected). I’m not a stock market expert – but wouldn’t this expectation of return for Yahoo! start to drive up Yahoo!’s price? If Google has, according to analysts, stalled or even lost share, then Yahoo!’s performance looks even more remarkable, and the share price cranks up?

And, for Google, this is easily repairable. After the market play, Google turns to advertisers and says “Oops, mea culpa, we goofed – we’ve made AdWords work properly again.” In a quarter or two, the revenue stream is back at Google, or even enhanced (everyone loves a Big Friendly Giant).

Did I make my coffee too strong this morning, or wake up light headed and incapable of rational thought? What did I miss? Is this *subtle enough* to explain why Google would trash so many small advertisers and risk their reputation?

Should the rational response be to drop spending or just redirect it to Yahoo! and MSN, or is there a smarter strategy for advertisers?

I’m still thinking through the implications. Your thoughts are welcome :)

Oddly, in the UK, two adverts consistently show up on a search for “No Country For Old Men DVD”. So these must be super-relevant? Something about the advertiser or the landing page or the AdCopy must have sneaked these past Google’s filters and reviews. I decided I needed to see what was special about these adverts. I clicked on each advert in the results below:

Absolutely fascinating… Here’s a clip from the Ice Gadgets Landing Page:

It has the keywords on the LP. It has a photo, and a YouTube clip, information about the movie and a low price. All good stuff that should excite a searcher to buy. What about the DVDCrave web site?

Eh? Follow a link from a search for one film, an advert for that film and you get shown a completely different film?

But other relevant advertisers are being suppressed?

What’s the point of denying relevant adverts, and allowing irrelevant responses to adverts, to be shown? I think it is a symptom of a deeper illness at Google.

Other Evidence

I have clients with highly relevant, high CTR adverts, and conversions (proof that search users found the adverts to be relevant and useful) who have had their businesses crushed in the latest round of changes. They aren’t getting more than a tiny fraction of the impressions that they used to have. The adverts that do run are getting CTR’s in the range of 11-60%. But the *volume* of searchers is now so low that the businesses are no longer economically viable.

What Is Google Doing?

Is this is economically criminal? The US mismanagement of poor quality loans is already endangering the world economy. Now, the world’s leading resource for directing online spending, appears to have started destroying businesses.

When your advertising system controls the flow of a substantial fraction of the online economy, you can’t just play with it the way that you want. It has a direct global economic consequence. Perverting the system the way that Google appears to have done, helps to reduce confidence – that intangible on which economic value is founded.

Google is directly contributing to a loss of confidence in small and larger advertisers, but not because the economy is tanking – it is because has deployed, unannounced, a new set of advertising rules that have destroyed the old AdWords model of relevance.

Idiots.

What possible reason would Google have for wanting to make paid search into a pointless activity?

Search History Permutation is a strange move, at least for niche advertisers.

Suppressing irrelevant adverts is a plausibly smart move. But the selection of what is irrelevant is clearly flawed, as conversions are being reduced.

If Google is going to trim down the number of advertisers, then it *MUST* make sure that those remaining are actually the best adverts. Otherwise the confidence of searchers will be destroyed, as well as the confidence of sellers.

Summary

I’m pointing all my clients to Yahoo, and MSN. I’m recommending that Google budgets are being *zeroed* until this monstrosity of madness has been repaired.

When Google can deliver an advertising system that is fair and rational, I’ll start using it again.

Look forward to more posting about Yahoo and MSN advertising behaviour – until now, they’ve been a minor part of our activity, so I rarely write up what I see there – they’ve been too small a part of the UK online advertising scene to be important, and while some of our US clients have Yahoo! accounts, it used to be the case that the core business was driven on AdWords results, with anything else being a bonus.

Now that AdWords is a statistically irrelevant advertising medium, Yahoo & MSN are *the* most important players in online advertising.

Updates

2008-04-07 Impression rates in clients are now trending up – but still sustaining higher AvCPCs. While I’d like to claim that impression volume increase is our smart management of accounts, I’m pretty sure that it is really more to do with changes made at Google’s end. There are still effects in damaged conversion – accounts with multi-year histories of increasing conversion volume and decreasing average conversion cost, saw damage in early March, which is now beginning to show signs of recovery – not caused by the 30 day accumulation of conversion. Looks like Google are correcting some of the excesses of the new algorithmic regime.

Moderated some of the language in the “Idiots” section as a consequence. I still think that the unannounced spanner that Google apparently dropped into AdWords is irresponsible.

You know, I thought I’d written in below, about how a service company (Yahoo) and a product company (MS) would have integration problems – an issue I’ve seen before in startups where competent managers in one type of operation failed to build a company in the other sphere. This article spells that out and supplies other great detail.

I’ll have to re-read the Roughly Drafted article and see if I can pick some holes or offer some amplifications, but the general thrust is compelling. Right now, I still do see an opportunity for advertisers. But for the companies involved, a mutual cooperation agreement for an approach to advertisers looks healthier than an acquisition. For search agencies, the fewer the number of outlets, the lower the costs of managing them all and the larger the addressed audience… The cost savings of managing the combined share of market through only one interface, MSN or Yahoo, could improve the attractiveness, especially for smaller accounts in Europe. As the Rough Draft article spells out, that’s probably not enough to make the business case.

In the UK, Google has over 80% of search share. But, as I’ve said before, the Content Network is a problem and it is often ignored by larger advertisers. Even worse, for Google, non-textual (display ads) are a small fraction of the Content Network. Google’s recent advice to advertisers will improve performance, helping increase advertising, but at the expense of higher costs to advertisers through building additional AdGroups with specialist techniques that may be uneconomic if recognised in accounting as a cost of sale – coupled with a lower click CTR and conversion rate, it’s usually a lethal combination for a small advertiser.

One of my biggest problems with advertisers and the Content Network, has been proving that the Content Network makes a difference. I have a feeling that Display Advert impressions make a difference to awareness. This may translate later into an increased CTR for paid search, or organic adverts, or affiliates, or even direct access to the site – but it is an invisible story unrevealed by attributable and measurable lead generation. This is a major problem for Google – they don’t like passing on information about users and have shown no indication of any willingness to pass on information about impressions (post impression tracking). Statistical treatments to prove that impression volumes have an effect, also need to take into account which sites are being used, and need large volumes of data to demonstrate statistical correlations. And it all gets much more complex when you consider latency (the time delay between beginning to think about buying something to actually getting around to it – can be measured in months or even years in parts of the travel industry).

Microsoft and Yahoo could steal advertiser hearts and establish a new standard in internet advertising by offering cookie tracking on impressions for display advertising. If I can show my clients that Display Adverts result in increased awareness, I can sell more advertising. I don’t think that MS and Yahoo would have the qualms that Google would have about this. As usual, I don’t care about identifying individuals – the goal of post impression tracking is to look for statistical correlations between adverts and response and offer ways to tune the delivery to users that are more likely to want to see the advert (win-win for user and advertiser). If I can deliver post impression tracking results and analysis, whew, I’ll be a much happier bunny.

The increased targeting options for Microsoft also play to segmentation concerns. I have huge doubts that Google’s content matching alone is enough to provide adequate segmentation. I’ve seen too much of “(named software) download” being matched to “(ripped music) download” and “(ripped video programme) download”, or other equivalent mismatching of content. Google’s pretty much unexplained “search behaviour” targeting, where they claim to improve performance through optimising based on recent searches… well, if I don’t understand it and can’t control it, I don’t trust it. I certainly don’t trust Google to do more than optimise their own revenue.

Microsofts targeting models also leave me a little concerned – but they are a step on the right path and frankly, without real information from Google about search history and targeting, I’d rather have the controls that MS offers, than the lack of control offered by Google. I can explain to clients what my errors are – but if Google is changing the game behind my back, I can’t justify changing response to clients. So a Microsoft model is much more interesting.

Much as I loathe Microsoft as a company (I’ve detested using Windows, for years), I do see real opportunities for Microsoft to work with Yahoo to improve targeting and measurement, and to steal advertising share from Google. Whether the world is better served with a Google near-monopoly of search advertising and a Microsoft near-monopoly of display advertising is a different question. But in the short term at least, this may require that Google treats advertisers with a little more respect, and that can’t be a bad thing from where I sit.