For example – while on the Light Rail (in Sydney) I saw a poster for I hate Sarah Marshall and a URL but I forgot the URL so I googled using those keywords.

This obviously has advantages but the difficulty arises when it we need to analyse the results, consolidating disparate web sites analytics (especially using Google Analytics).

Micheal: Good point on the viral aspect of incentivising a campaign. The campaign would have to be carefully planned to minimise this if any kind of accuracy of origin is required. Perhaps vanity URLs such as: mysite.com/xmas or mysite.com/radio would work better than offer_codeX? I would love to hear any ancedotal experience of that – see Chris’s comment.

Steve: Sounds like you would like a GA API? I can’t comment on product development directly, but yes it would make sense for Google to do that. Its a common feature request and it certainly fits with Google’s open source philosophy of making data accessible.

Chris: You hit the nail perfectly on the head when it comes to tracking offline sales that result from online efforts. Staff coaching and training, in order to build a culture of marketing measurement, are key to getting that to work. Sara Andersson has experience of this, so I am going to ask her for a follow up comment/post.

With so many factors affecting site traffic’s ups and downs (up to and possibly including phase of moon) it feels futile to try to infer a relationship between those ups and downs with known events like offline ads. That’s one of my problems with the idea of integrating data from all sources – once you have POS info next to site traffic info next to TV share info for a particular day, do you really want to rely on just an apparent co-variation? Perhaps over a long period of time a history of covariance will become statistically significant – but you’d have to have lots of data, enough to work out the daunting role of lag time. AND you’d have to have a stats program that is NOT the simple crosstabs that are the output of all the analytics vendors. Clementine by SPSS comes to mind. Or at least a regression analysis program.

Anyway, one thing I like for measuring online effects on offline is a short-term initiative where store clerks have to ask customers if they’ve been to the web site recently. You’d have to use some kind of odds ratio calcs or other epidemiological analysis to figure out if online made a difference, but it can be done. However, it’s really really hard to get the store people (or call center people etc) to cooperate reliably.

The best solution I’ve come up with is to apply the Goal conversion rate(s) on the coded URLs to the overall increase in direct and brand search traffic. This gives a better indication to the true value generated by the media source.

Good post which leads me to the question which was posed to me at a Web Analytics Wednesday I was speaking at in Helsinki.

I was asked “what is the hardest thing about web analytics in terms of measuring traffic sources?” which I thought was a good question.

My answer was “integrating the data”.

If you have numerous sources of data as most businesses have it becomes a difficult and often manual job to integrate that data into one place in order to be able to compare it. No tool today does this well unless you’re prepared to pay for something like Visual Site, which is a six figure yearly fee at least. You usually have to use something like Excel to do the integration and then see what and how you can compare it intelligently.

Now most systems have some sort of Excel functionality and GA is no exception having the export function to CSV.

My question is are there plans to extend this to make it easier in Google Analytics?

Currently it’s a manual job to integrate the data unlike in Omniture SC, or HBX where you can stream the data on demand into predefined excel spreadsheets. This for instance allows you to easily set-up KPis which include data from other sources. The idea is you press a button from inside excel and the data updates automatically with the latest weekly, monthly, quarterly (depending on your requirement) data streamed into excel.

This is a huge time saver when it comes to data integration.

Thanks,
Steve.