(First published in the WAO/FACTOR newsletter in August 2013)
Quite a lot of Digital Analytics, and plain Analytics for that matter, is about optimization. I mean by optimization the whole set of activities, with analytics and testing at their foundation, done in order to improve the ratio of X doing Y. The most typical manifestation of optimization in Digital Analytics is increasing conversion rates. This is very powerful stuff. This is why Digital Analytics is so important; because conversion rate optimization works and pays off. A lot.
Conversion can of course be many things, and even its calculation can vary depending on solutions and people. I, for example, will calculate the overall site conversion rate in order to give me an idea of the site general output, but I will only try to optimize the process conversion rate, thus getting rid of all the visits that are on the site for other reason than the conversion process goal (and which can skew the rate a lot).
Funnels and Tunnels
The first level of optimization for conversion, or at least the one that is better served by traditional Digital Analytics solutions, is certainly process optimization, of which the e-commerce checkout process attracts the most attention. Most conversions happen on same-session, multiple-step processes, often call funnels, with people abandoning at each stage. This is a very classical Digital Analytics model (when we called it Web Analytics), and it looks at all visits to the process start as equal opportunities to go all the way to the very final conversion stage. Obviously, in many cases, the purchase cycle can be over more than a single visit, but suffice to say for now that the original optimization model is content with improving the process on a same-visit basis. And why not? It works fine, and there are tons of low hanging fruits to be picked.
The logic behind that type of optimization is, all things being equal, a small improvement at any step will have great positive repercussions at the end of the process, since improving a step will send more people to the next one, and so forth. This is what I call “mechanical” analysis, because the process is seen as an ensemble of cogs, bolts, and belts that just need more tweaking. It doesn’t require to be really good at marketing, really. It doesn’t even require to understand why changing this tiny element brings a 0.56% improvement. It is enough that, through tests, one has statistically validated the predictability of that change yield. And why not? It works very well, and there are still tons of companies that don’t even do it.
Sources and Origins
Once content (copy, navigation, processes, etc.) has been optimized to the bones, some analysts will look at where people came from, and look for significant differences. Yes, segmentation. Sure, one can very well segment on people’s behaviors on the site, but looking at points of origin can inform the marketing dollar a lot more, and we know that there many of those dollars spent on bringing people. This being said, such optimization, i.e. increasing the yield of the marketing dollar, is most of the time done in relation with conversion, which is, let us not forget, the last moment of the purchase cycle. But again, why not? This type of optimization sure offers tons of value creation opportunities.
Here you will certainly see attribution questions starting to show. One can optimize sources by investing substantially more in the “winning” one, but wait! What if it was rather a combination of those sources (often called touch points) that would bring even more results, thus demanding that the said combination be optimized? Sure, it makes sense. However, this is an incredibly complex endeavor, and I am far from being certain that concepts, models, or even technology are reliable enough to allow for fine-tuned optimization. Again, a more “mechanical” approach here, i.e. finding to the best source, will reap a lot of short-term benefits, and keep you from being distracted by the buzz machine.
Qualities and Characteristics
All that optimization is behavioral. What if one could extract even more value by examining what characteristics are shared by people who converted, i.e. socio-demographics, psychographics, in short, qualities rather than behavior. We know that this has been done for ages in Marketing, but it is still just starting in Digital Analytics, because the more mechanical types of optimization have been so much easier, and plain darn attractive!
Here too, the optimization stays focused on conversion, and it can certainly get even more incredibly complex when one starts examining buyer characteristics in conjunction with sources, and even more so within an omnichannel perspective, making technology vendors dizzy with expectations. Thousands of fine points can be optimized at this stage, with endless variations, applied to more and more massive data sets, so much so that one can doubt if it will really be possible to be perform by humans (see an interesting discussion on this newsletter LinkedIn Group about real-time).
At this point, one is full gear in sophisticated Digital Analytics, heck, in Marketing’s top class! But wait! There is more.
Marriage and Loyalty
Once the acquisition is done, there are still many optimization opportunities. Obviously, customer profitability, often called LTV (Life Time Value), is a major model analysis grid. A customer can be very profitable with only one transaction. However it is believed that she can be even more profitable on the long run with the accumulation of profitable transactions.
Sorting customers by profitability, and loyalty (and they don’t necessarily go together!) is here too the realm of more traditional Marketing. However, the Digital Analyst can inform those analyses a lot, if s/he is not the one performing them. Actually, with almost all customer data being digital now, this brings the question of what Digital Analysts actually do that other types of analysts don’t… and vice versa. But this is another topic.
Certainty and Creativity
Digital Analytics optimization is thus extremely powerful, especially since it integrates more and more mathematics (i.e. statistical methodologies) to make it much more able to produce predictive, even prescriptive recommendations. The whole past can be a very powerful predictor of the near future.
But is business life only a matter of “last year plus two percent”? Well, in many circumstances that 2% can be extremely valuable, and you should chase it as much as you can. At least, start! However, is Analytics only about pecking tiny bits of gain here and there? Even when it makes strong predictions, isn’t Analytics rather backward looking? What is its role in the innovation process? How can it inform and support a vision that tries to see what it not there yet? Is Analytics a ceiling or a springboard?
These are questions that have obsessed me in the last year, and I will share some thoughts about them in the coming one.