The $64,000 question (actually,$64,000 is a very low price to put on the value of this question) in online marketing response analysis is whether data from your ad server or your own site is better for understanding the effectiveness of your online (and possibly off-line) marketing.

My colleague, Brendan Riordan-Butterworth, recently sent me an e-mail with his thoughts on the pros and cons of the two approaches, and I’ve added some of my own.

This approach works by implementing a redirector between the ad (a banner, or a paid search ad, or even a link in an e-mail) and the ultimate destination (the advertiser’s site). When the user clicks the ad, the redirection is logged by a server on the internet, and this data is compiled to show how many times the ad was clicked.

An enhancement to this approach is to use cookies to track which ads a user has seen and clicked on.

A further enhancement is to add some ‘client-side’ analysis in the form of tags or web-bugs on certain pages (e.g. conversion pages) on the advertiser site. When the user reaches these pages, the tags fire and send information back to the ad server, which correlates it back to the advertising that user responded to.

Pros of this approach

• The numbers reported relate to what you’re actually charged for the advertising, making ROI calculations easy
• Most advertising networks automatically detect and exclude invalid clicks, and these are excluded from the reports
• No client instrumentation cost for simple click-through tracking; minimal client instrumentation cost for conversion tracking
• Impressions (for display advertising) as well as clicks can be measured and related to subsequent conversion behavior (through the use of 3rd-party cookies)

Cons of this approach

• Click rates are typically overstated (compared to arrival rates) by ~10%
• Reliance on 3rd-party cookies reduces accuracy (especially over longer periods)
• Difficult/impossible to compare and de-duplicate multi-channel response (e.g. e-mail vs Paid Search) without using cumbersome (and sometimes costly) redirects (or ‘non-served placements’) through the ad server
• Completely unsuitable for certain marketing channels (especially offline) since click-thru URLs are so complex

Site- (‘Client-‘)side analysis (web analytics)
This approach uses a web analytics system to analyse all the behavior on the advertiser’s site. Correlation with marketing response is achieved by directing all marketing click-throughs to customised landing URLs (usually just standard URLs with some dummy parameters added). Typically tracking tag code is added to every page on the site, and configured to extract conversion value information (i.e. shopping cart values). A 1st-party cookie is used to correlate conversion behaviour with marketing response.

Pros of this approach

• The users measured by this method have definitely reached the site (rather than just clicked an ad)
• Externded page view activity can be reported (e.g. number of pages viewed per keyword clicked)
• Can de-duplicate multiple clicks from the same user within a single visit, and use ratio of clicks (or touchdowns, technically) to visits to identify possible click fraud
• Reporting is based on web clients that support JavaScript (when using Page Tagging analytics), so that basic scripts and bots will not trigger it
• No automatic exclusion of invalid clicks – all the data is presented, and it is up to the viewer to analyze
• Correlation with other behaviours (e.g. product interest) provides more value than just analyzing expected conversion events
• Correlation with user demographic/CRM data (held by the advertiser) enables better targeting of messages & ad placement
• Able to measure multiple marketing channels side by side (e.g. e-mail, display, search, affiliate, even offline) & de-duplicate their effects

Cons of this approach

• No post-impression analysis
• Can be very complex to set up for true multi-channel analysis (depending on the tool)
• Difficult to tally up with numbers from server-side analysis (and vice versa, of course)

The holy grail here is to combine server- and client-side analysis into the same tool. Google Analytics achieves this in a fairly clunky way just by delivering different reports in the same interface based on different data sets. As soon as you look at the reports side by side the differences become obvious, and confusing for users (#1 reason for a user to stop using their web analytics tool, IMHO, is that they don’t feel they can trust the numbers). I’ll blog about techniques for this in a future post.