Web Analytics Applied

Flat CTR’s and Google Enterprise Analytics

This week, various news outlets reported the results of a ComScore analysis showing that CTR’s on Google ads have essentially flattened out (http://biz.yahoo.com/ap/080328/google_paid_clicks.html).  YOY January data showed no change, while February was up only 3% from last year.  This compares unfavorably with the historical double-digit YOY growth.

As a web analyst, there are so many reasons this doesn’t surprise me: AdSense ads have now been around for years, and web-users may have become immune to their novelty; saturation of Google ads on top keywords probably reduces user interest or confidence.  But above all, it boils down to the inconvenient truth that as a PPC campaign expands, using new keywords, Content Match, Broad Matching, and the like, the quality of leads goes down, with the result that (eventually) advertisers take notice and scale back.  It’s a quality-over-quantity calculation that many times leads to the axing of Pay-Per-Click altogether from a company’s media spend, and there’s very little Google can do about it.  When PPC visits see bounce rates of 80% or more, or when brand-keywords are used primarily as a substitute for the browser’s address bar, or when most content match referrals come from AdSense gamesters’ auto-search websites or trashy foreign blogs, no wonder advertisers scale back ad placement on Google. 

But advertisers don’t know any of this unless they have a web analytics solution in place.  Google’s own spin in response to this week’s articles has been that they are going after quality of clicks, not quantity, in order to make each click more “valuable” to the advertiser.  That’s good rhetoric, but how is anyone going to validate this, and what will Google do to act on this initiative?  Google’s algorithms are still, after all, CTR-based, because, for the most part, they don’t have insight into website engagement or conversion, for which there are few standards anyway.

Enter Google Analytics.  Many have puzzled why Google launched this free application to begin with, much like other free toys available from Labs.Google.com (?Google Mars?).  But Google may use this flattening of CTR’s to re-think how they market – and use – web analytics.  Google may now discover that it might be worth investing in – and supporting – an Enterprise Level Web Analytics Solution that would be able to measure and analyze all these pieces of PPC engagement.  By providing this application (for a price), Google might be able to both control and exploit the data it generates to optimize its ranking and bidding algorithms, while putting some action behind the rhetoric of “more valuable clicks”.  It could even do this without the advertiser or client ever knowing, just as Google exploits its page-content databases to syntactically model the English language for Search Algorithm refinement.  The dozen or so SEM Analyses possible with a robust, enterprise-level web analytics solution could be automated and streamlined by a Google version – much like Omniture’s Search Center but without the third-partly hassles.  And because Google has a vested interest in tying this to its PPC, it can probably afford to price such as solution much lower than those currently on the market.

We’ve often pondered whether Google will enter this space, and one answer has always been, “Why would they?”  Flat CTR’s and rhetoric about click-value might be a good excuse.

Web Analytics Overkill

When people ask me “what should I track?” my usual philosophy is that tracking should be implemented in such a way as to record in detail anything that is analytically valuable, but also in a way as to make reporting as easy and digestible as possible.  In a forthcoming article, I’ll be discussing the last piece of this statement; but I’d like to say a few words about the first piece – the key phrase is “analytically valuable”.

There is such a thing as “overkill” in Web Analytics.  It often occurs when there is poor communication between the business requirements group or product owner, and the team responsible for designing and implementing the tags, who decide on a policy of “cover our bases” and tag everything they can think of.  This means scores of success events on the site, meaningless variable roll-ups, over-complex campaign tracking codes, and custom link-tracking server requests which slow down performance and result in huge amounts of data within the web analytics tool, the analysis of which would not be worth the effort (literally).

What does “analytically valuable” mean, exactly?  It does not refer to data for reporting or KPI’s (that’s what variable structure is for); rather, it refers to a data-set whose comprehensive analysis would be worth the time and resources, producing valuable, actionable recommendations about website design, user behavior, or marketing effectiveness.  A deep-dive analysis on usage of the “close” button from different popup windows is probably not worth the money.  A team implementing web analytics on a site has to not only ask, “can we capture this behavior?”, but more importantly “can I foresee someone analyzing this data to potentially produce actionable results that would be worth the resources expended?”

This last element is why the overall budget allocated to a website – now and in the future -- should be taken into account when designing a WA implementation.  As a web analyst, this may sound like heresy, but in the real world, it’s why mom-and-pop shops choose Google Analytics (if anything) instead of Omniture.  Alexander’s Pizza Shop on Main Street might have a website with a menu, directions, and pictures, but obviously would waste its money by implementing NetInsight.  It would be dishonest to recommend a state-of-the-art, “measure everything” implementation when it is clear that devoting resources to executing it and analyzing it subsequently would be a waste of money because the website is small-scale and not a significant part of business success.  That’s overkill.  If it becomes a bigger piece of the business, then a more robust implementation might be warranted.

Even within robust, enterprise-level implementations, measurement overkill is possible.  Here are some examples:

·         extensive tagging of the footer on a site.  “Terms of Use”, “Privacy Policy”, “Corporate Info” – these pages on a website usually exist for legal or compliance purposes, and exhaustive measurement of their usage would make no difference.

·         an extensive analysis of a Site Map would probably make no difference either, because they often exist more for SEO than for anything else (though if it’s used more than your navigation, you have a problem!).

·         Over-use of campaign classifications: channel, creative, ad type, adgroup, keyword, date-stamp, banner size, link within an email, landing page à all these can be legitimate and useful pieces of a campaign tracking code.   But requirements and resources will dictate what is more important; thousands of permutations are possible with all these being tracked in combination, and before implementation, questions should be asked as to whether all of these are useful.  A walk-then-run approach might be more appropriate.

·         Over-Use of success events: while websites typically can have multiple success events, some implementers take advantage of the availability of dozens of success events to tag almost any action as a separate success event.  This leads to opaque reporting because it becomes unclear which of these success events gets included in an overall total effectiveness calculation. 

·         Over-redundancy in page-naming: some sites record pages as the URL, then pass a user-friendly name into one or two variables, then pass a hierarchically-defined page name into another variable, then pass another variable in an onclick handler recording another version of the page-name of the link clicked on, while also populating site section and hierarchy variables.  When a manager wants to see a pages report, they don’t know which one to use: “www.mysite.com”, “homepage”, “hp_”, “mycompany|mainsite|homepage” – you get the picture.  And chances are, the numbers for these won’t match up.

There are many other examples, and I’m probably guilty of a few. 

I’m not saying that a “measure everything” approach is bad – on the contrary, with analytical resources available it can be a vast asset for value-optimization of the online channel.  Rather, I’m saying that a “measure everything” approach has to take into consideration the analytical value of the data, and the resources that would be required to implement and take advantage of it.