By Mark Razzell as published in Marketing Mag.
Predicting and influencing the consumer behaviour of visitors is difficult because websites aren’t linear, so trying to measure your customers in a linear manner isn’t indicative of reality. Instead, think of your visitors like fish in a tank.
Many years ago, I spent 90 hours sat in front of a four-metre fish tank with a notepad and pen as part of my studies in evolutionary biology and behavioural ecology. To most people these fish would look much the same (because they were very closely related), but a fish geek like me could identify at least 30 species.
There were easily over 200 fish in the tank, all seemingly performing similar, random behaviours. However, we wanted to measure their behaviour and identify differences between species, if indeed there were any.
The long and short of this was that we identified significantly different behavioural repertoires between species, which helped explain some important evolutionary based questions. We didn’t disturb the fish. We didn’t force them into situations where they must adapt. We just watched, noted a huge amount of data, and then ran a multivariate regression model.
The tank is your website; the fish are your visitors
Regression models take variables and, effectively, analyse whether or not there is a genuine relationship between them. We can measure a mind boggling number of variables against one another and we can see how they all interact.
Let’s consider a hypothetical scenario. You have an ecommerce site and you want more conversions. The question is ‘which behaviours that users undertake on my site are related to conversion?’ To answer this question, you need to ring-fence some of your user data.
Identify a sample of, say, 400 visitors and label whether they converted or not. (400 is an arbitrary number here, but a large sample size is fundamental to statistical viability.) Then, identify a number of other data points that could be considered as ‘behaviours’. These could be time on site, number of pages visited, their point of entry, what brought them there (EDM, sponsored links, search, etc.) The possibilities are almost endless.
The way this data is analysed will then identify the relationships between variables and their impact on conversions. Let’s say, as a highly simplified example, that you identify a clear and reliable relationship between conversions and people who arrive at site via organic search, spend at least 10 minutes on the site, and view at least five product pages. What you now know is that people who arrive on your site in the future and exhibit these behaviours are likely there with intent to buy.
What we have just done is identify the dependent variable (conversions) and measured them against our independent variables (everything else). These are so called because the change in our conversions, according to the statistical analysis, is dependent on fluctuations in the rest of our data.
You can set anything you want as your independent variable, depending on the type of question you ask. As another example for those who don’t use ecommerce as a success measure, you could look in to what consumer behaviours performed on your site are conducive to making contact. To do this, you would simply set ‘contact us/me’ as your dependent variable.
Ways to predict and influence consumer behaviour
As an approach, this provides us with a great way of predicting visitors’ consumer behaviour, especially when we don’t have any prior information (like we do when returning visitors have cookies). It is far more powerful than traditional methods of measuring consumer behaviour, which often relies on validation exercises via percentage based goals. These methods make rather heavy assumptions about the journeys your customers are taking through the site and their reliability from a statistical perspective is low.
If you’re smart, you can tailor your site to be adaptive to this behaviour. So, when someone is exhibiting these traits on your site, have your site do everything in its power to increase the chance of them performing that ‘successful’ behaviour. In the case of ecommerce, this might be the inclusion of a limited offer voucher pop up (10% off or BOGOF). Or, for those who are displaying interest in making contact, enhance the visibility of the ‘contact us’ section. This is relatively simple for any developer worth their pay.
In terms of how you would go about doing this, there are two ways of getting hold of the data. The first is to make sure you know what questions you want answered when you set up your own back end data collection. That way, you get a no-fuss anytime snapshot of the data you want as and when you need it. The other way, if having a complex back end isn’t feasible, is to utilise Google Analytics’ capabilities. There are certain behaviours that are captured readily (duration on site, number of pages, etc), but others might need custom codes to get the data you want. Custom variables might require hired expertise, but they won’t require re-mortgaging of the family home. Once you’ve got all your variables, you can easily set up custom reports in Google Analytics to get the data anytime you want it.
There are also two ways to go about the analysis. You could engage a statistician who will manually crunch the numbers for you. Another way is through automated software. Some savvy statisticians have linked up with talented developers and are putting together customised software that will do all the legwork automatically; you just upload your data set. These offerings have been a staple of conglomerates like Amazon and Google for some time, but they are increasingly becoming available to the wider market.
The fact of the matter is that most websites aren’t linear, so trying to measure your customers in a linear manner isn’t indicative of reality. Instead, think of your visitors like fish in a tank.