Comments on: Are You Ready for the Coming Revolution?

By: B-to-B Marketers: Analytics Key To Your Internet Success «

B-to-B Marketers: Analytics Key To Your Internet Success « — Wed, 02 Dec 2009 14:23:51 +0000

[...] The “coming revolution” in analytics, are you ready? Analytics consultant Eric T. Peterson says, “From where I sit, we have seen more changes in the last 24 months than we had in the entire 12 years previous (my tenure in the sector) combined.” [...]

By: Die Zukunft der Markensteuerung – Ipsos, Doubleclick oder Radian6? Teil 2: Die Webanalyse « CEO Blog

Tue, 01 Dec 2009 08:32:12 +0000

[...] und flexibler einsetzen können. Ein absolut lesenswerter Post zu dem Thema findet sich auch auf webanalyticsdemystified, wo auch ein umfangreiches Whitepaper zur Verfügung gestellt wird. Share and Enjoy: Diese Icons [...]

By: Zukunft der Markenführung: Teil 2 - webtracking, adserver tracking, webanalytics | Minced Marketing

Fri, 27 Nov 2009 10:51:36 +0000

[...] und flexibler einsetzen koennen. Ein absolut lesenswerter Post zu dem Thema findet sich auch webanalyticsdemystified, wo auch ein umfangreiches whitepaper zur Verfuegung gestellt [...]

By: La mia presentazione al RWME • Google Analytics in 30 secondi

La mia presentazione al RWME • Google Analytics in 30 secondi — Tue, 24 Nov 2009 12:52:47 +0000

[...] alcuni aspetti coi quali dovremo fare i conti prima o poi, partendo da una citazione di Eric Peterson. La prima questione posta riguarda l’orizzontalità o la verticalità dei prossimi sistemi di [...]

By: Matt Gershoff

Matt Gershoff — Sun, 15 Nov 2009 23:46:39 +0000

This is a really good post. A couple of quick things to think about:
1)Online is real-time. Okay, so that is painfully obvious but it does mean that some of the approaches from database marketing /BI and what SAS is talking about might need to be recast a bit.
2)Just in time vs Batch Learning. One potential advantage with real-time environments is that each request provides the opportunity to provide the user with the ‘best’ response based on the most current information. In order to effectively provide this capability the organization must in some sense balance learning (trying out different options/policies) with exploiting (making use of the learning) – this trade-off is often referred to as the explore/exploit problem and is often looked at within the context of the Bandit problem (a wiki search for either ‘Bandit’ or ‘explore/exploit’ will provide a good start).
3)Most current formal approaches learn in a batch mode (e.g. AB/MVT but can include other supervised (regression) and non-supervised (segmentation) learning methods). Batch methods spend some time collecting response data, and then analyze/model the data. For hypothesis testing approaches, after a certain confidence threshold is reached the ‘best’ option is applied to the majority of user requests. Issues with Batch approaches:
a. Opportunity Costs. The longer the learning / data collection phase takes the longer the users are receiving suboptimal responses (see ‘Regret’).
b. Learnings are often perishable – one implicit assumption of batch approaches is that the environments are stationary. For the analytics situation, this just means that learned relationships tend not to change much over time. If this assumption does not hold, you can spend resources (time, opportunity costs) formally establishing a relationship that is actually unraveling over time. These unraveling relationships can be anything from the simple (e.g. what is the best banner to use) to the more complex: (e.g. a fancy Gaussian mixture segmentation model or look-alike targeting models (think regression or ANN)). Methods that continuously update these relationships will be needed.
4) Sequential Decision Making –websites are complex objects and methods that are able to account for this structure will be able to learn faster and more accurately. This is closely related to attribution. Desired behaviors may take place downstream in the decision path. Proper attribution improves the selection of decisions that take place several steps before the goal behaviors.
As we move forward there will be more pressure for real-time learning methods and to move the intelligence closer and closer to the decision points within each application.
Eric thanks for getting the discussion going and putting up the SAS paper – it is good stuff.

By: 11/13/09 E-Alert: Newspapers Readership Remains a Part of our Routine and Culture - Growing Audience

Fri, 13 Nov 2009 19:56:14 +0000

[...] Source: Web Analytics Demystified [...]

By: Web Analytics for Ad Agency New Business « FUEL LINES Fueling Ad Agency New Business Through Social Media

Thu, 12 Nov 2009 12:37:07 +0000

[...] The “coming revolution” in analytics, are you ready? Analytics consultant Eric T. Peterson says, “From where I sit we have seen more changes in the last 24 months than we had in the entire 12 years previous (my tenure in the sector) combined.” [...]

By: Adrian Palacios

Adrian Palacios — Wed, 11 Nov 2009 15:19:58 +0000

Eric, I’ve been thinking about this and realized that I really do not have a good idea of what, if anything, I would like to start modeling/predicting at the Analytics 3.0 stage.

I know for certain that I want to start quantifying each stage: X amount of dollars for X amount of leads for X amount of people to the sales office for X amount of people closing on a new home, and the costs associated with each stage of that funnel.

But what to model? Hmmm. I know the biggest thing is to map out a sort-of “efficient frontier” of advertising dollars spent vs. closings. In tandem with that, I want to know what creative and advertising sources produce the people most likely to close. Beyond these, with the limited data set we produce, I don’t know what else I could know/want to know :-/

Sidenote: I wonder how many others there are like me, who, reaching this stage, don’t even really know what they want?

By: Tom Erik Støwer

Tom Erik Støwer — Mon, 09 Nov 2009 21:35:57 +0000

Thanks for the link Eric, I still feel that the specific problems that concern editorial staff aren’t addressed, at least not directly. Or maybe I’m missing something. Interesting paper, nonetheless.

I agree that intelligent depth is the next step, and necessarily tailored for each user “persona”. At least, in large organizations, developers, designers and content producers care about different things.

By: eric

eric — Mon, 09 Nov 2009 18:42:48 +0000

Tom: I think the biggest (or at least most noticeable) shift will result from the blending of web analytics and traditional business and customer intelligence tools. Deserved or not, BI tools get a bad rap for being slow and complex, and are best known for providing depth in the results they provide.

Conversely, web analytics tools, at least the best of them, are known for providing fast (even real-time) results from relatively easy-to-learn systems, but get dinged for not being particularly deep.

My vision of “Business Intelligence 2.0″ is the convergence of these types of tools — “easy” to use applications that also provide depth in their results. You see some of this out there today in applications like Quantivo and others who bring a more traditional BI focus to the digital landscape. I suspect we’ll see A LOT more of that in the next three to five years (maybe sooner!)

Regarding journalism, while a little dated you can have a look at a paper I did for the Newspaper Association of America (NAA) awhile back. Free download here: http://www.webanalyticsdemystified.com/sample/KPIs_for_Media_Properties_-_Newspaper_Association_of_America.pdf

Adrian: I’m glad you weren’t driving, that would have been a disaster no doubt! ;-)

Regarding your point about small sample sizes and statistical analysis, I’m not a hardcore statistician but I have worked with enough of them to know that you can surely conduct a more robust analysis with small sample sizes. What will be limiting is the kinds of questions you can ask and your ability to segment the resulting data, but based on the work I see people do all the time you should be fine.

Best example? Have a look at Google Analytics new “Intelligence” feature, fully grounded in good, old-fashioned statistics. If they’re able to provide this feature to low volume sites (which many GA sites surely are) then you should be good to go, at least with some questions worth asking.

Another good example is the use of Voice of Customer solutions like Foresee Results and OpinionLab. These solutions provide great insights based on relatively small samples all the time. In fact, if you’re not already using a survey tool that might be the best way to get deeper insights from the audience you’re targeting.

Let me know if this is helpful and if not I’m happy to hop on the phone to discuss.

Thanks to both of you for your comments and positive feedback!