Web Analytics For It's Own Sake Misses The Point
Posted 23 January 2012 - 02:56 PM
* necessity of a solid defined business goal;
* an articulated specific strategy for achieving the goal;
* stated measurable tactics to accomplish the strategy;
* a foundation of benchmarks of said tactics against which to compare;
* determine the relevance of available metrics;
* which analytics frameworks best display required results;
* what might be missing and what might be unnecessary and making appropriate adjustments;
* budgeting resources, i.e. time, tool cost.
Most do an adequate to exemplary job at telling you how to do whatever - few tell you why you should bother (and frequently you probably shouldn't). Analytics is NOT a one size fits all activity. It needs to be at least custom fitted to meet each new set of requirements - and if your requirements rarely change you might consider which business goals you are ignoring.
Don't just do whatever a particular tool allows. That is totally backward. Each goal you set should determine which capabilities of which tools are required. Let the goals set the requirements set the tools, not the reverse.
There is no science without fancy and no art without fact.
I use the above quotation to illustrate that the reason for web analytics is not simply to chart one's progress for good or ill but to look forward and make predictions. To determine outcomes not simply record them.
Posted 23 January 2012 - 05:20 PM
We haven't gotten to predictive analytics yet. And, trust me, you're not ready for that.
Posted 23 January 2012 - 05:25 PM
Posted 23 January 2012 - 07:26 PM
Current web analytics are, indeed, backward looking. Past performance doesn't guarantee much of anything. We haven't gotten to predictive analytics yet. And, trust me, you're not ready for that.
The essence as I understand predictive analysis is to use existing data (both present and past: yes one should store - data warehouse - one's data for reuse) to statistically predict future trends.
Perhaps a basic example can be found in the testing of alternatives, either A/B or multivariate: one can, of course simply use intuition, which is an innate human predictive mechanism, but a more efficient way (for me) is to extrapolate test choices from existing though perhaps reordered data.
Of course such reordering requires (1) quality, i.e. quantity and accuracy, of data and (2) sufficient breadth/depth of known entities to recognise/extrapolate patterns in their relationships.
Now, granted, a webdev or small company is not going to wargame (what if) scenarios quite like larger corporations but they can certainly perform well out of their weight class.
I use predictive analysis for setting budgets, setting ad rates and conditions especially thresholds, estimating traffic/conversion/revenue both by referer type and by major referer, determining which social platforms to utilise (and how), determining where and how to split test, etc.
Once you leave the comfort of 'visitor' metrics and enter the world of 'user' metrics where you look beyond the immediate to what changes will most likely/efficiently increase revenue, decrease costs, increase market segment to overall stickiness, increase repeats and recommendations, etc. you must, in some measure, engage in predictive analysis.
Perhaps because I come from a navy and business background I don't find predictive analysis any more unusual than a rolling business plan (although I know of only one other webdev who also does both ).
There are a couple of companies that have advertised predictive analytics services but I have never worked with them. I think the topic is an interesting one. I am sure it would have to entail some sort of interferometric analysis.
Please expand your thoughts Michael. You confused me with interferometric, which I understand, vaguely, as the superimposition of different EM waveslengths to extract information not available from each alone. Are you speaking metaphorically?
Posted 23 January 2012 - 07:40 PM
Basically, you're comparing two data curves to look for conjunctions, congruences, agreement, etc.
So let's say you monitor query trends and you also monitor social media activity. You should be able to find some correlations between their spikes and dips. Narrowing the field down to those correlations, you try to work out what is actually connected (as opposed to what is just coincidental).
If you find candidates for connections then you might be able to make some predictions on social activities that drive queries in search.
And vice versa.
Of course, the more disparate data sources you bring to the interferometric analysis, the more reliable your inferences tend to become. That is, you tend to filter out more noise as you add more data sources to your analysis.
Really big events like the Christmas Tsunami, the Japan Tsunami, the Chile and Haiti earthquakes, the death of Bin Laden have huge impacts in all sorts of trend lines. You can look for smaller impacts in trend lines that correlate to types of seasonal and/or annual activities, but you can also track popular movements, superband concert tours, political campaigns, etc. in similar fashion.
Posted 23 January 2012 - 07:52 PM
Darn my memory...on re-reading your article I remember reading it but had forgotten that I had...memory can be something or other but I forget what for the moment wish I could blame it on age but I've always had this problem...
Posted 24 January 2012 - 03:09 AM
This topic seems pretty interesting, but, yet again, it seems like requires quite a bit of a learning curve, initial setup efforts and maintenance for someone without a high traffic website. Though it is, indeed, curious and I should research the topic sometime
As for the starting post, Avinash Kaushik tends to write articles about web analytics that do focus on:
- defining goals
- specifying how to achieve them
- demonstrating how to do that.
Also, the book "A(lways)/B(e) Testing" by Bryan Eisenberg first declares KPIs and then demonstrates how to measure/test/improve them, if I remember right.
Edited by A.N.Onym, 24 January 2012 - 03:09 AM.
Posted 24 January 2012 - 04:59 AM
That's just my take on it.
Posted 24 January 2012 - 08:30 AM
It does list typical business KPIs and ways to find and measure them in GA easily.
Glyn, did you mean something else?
Glyn and iamlost, could you please clarify, whether the article I linked to is one-sided or more or less balanced from the overall understanding point of view?
Posted 28 January 2012 - 06:23 AM
Cost Per Acquisition.
Obsess about this metric.
At first blush, this seems reasonable. Almost smart. It can be ruinous as well.
Cost isn't value for your money. And, pray tell, acqusition???? One does hope it is short for aquistion of a customer. You can go chasing cost until your cost per acquisition skyrockets. Chasing cheaper and cheaper leads, because while naive and mistaken, the focus is on cost, not acquisition.
Companies with little money to spend have much bigger problems than little money to spend. ....What's causing them to have little money to spend, for instance.
A lot of successful companies have outlandish cost per acquisition, because cost is offset by customer lifetime value. And a lot of people focus on new customers, to the detriment of anything to keep the existing ones. Finally, it's often less costly to acquire a consumer (a.k.a. a bottom feeder) than a customer. Retailers have problems with "customers" who buy a dress, wear it once, then return it. These people will flock to you every time you announce a sale. Good customer acquisition cost -- bad everything else.
Bottom feeders have marvelous cost of acquisition. But the expense is prohibitive. Of course, some people can read a box of cereal ingredients panel and come away with five new profitable ways to interact with customers. But they are not reading this type of article.
Next, checkout abandonment. My general rule of thought on this is focus on what you want to increase, say, Checkout Recoveries. If, for instance, you look at the steps in your order system where you lose customers, you can plug the leak -- sure. What that's unlikely to do is get people back to an abandoned cart. What you're unlikely to do is add a step where you email the person and get them to complete the purchase.
Try posting a five grand job on Fiverr. Cost per acquisition? In my book the cost was $4,995 -- but few figure things the way I do.
Edited by DCrx, 28 January 2012 - 06:34 AM.
0 user(s) are reading this topic
0 members, 0 guests, 0 anonymous users