Marketing Analytics: HowTo Translate Numbers Into Insight
You know how you wake up early everyday to go to work because traffic is a bitch in the morning, except that one day you oversleep and you think you’re going to be late but then you find that the roads are empty and you end up getting to work on time?
You’re so happy, you think you’ve got it all figured out.
So the next day you intentionally wake up later, counting on your insight that the roads are empty at this later hour, but this time traffic is a bitch and you get to work half an hour late.
Either god hates you or your assumption was wrong [or both].
Data interpretation, and specifically – marketing analytics, is what I want to talk about today. It’s crucial to the success of a business and yet we tend to get it wrong. Why?
Topics covered in this post:
- Analytics in a nutshell
- The 2-step process to insightful marketing analytics
2.1 Data collection
2.2 Data interpretation
- Why do we interpret data wrong?
3.1 Intuition about causality
3.2 Small sample bias
- How to translate numbers into insight the right way
4.1 Start with a goal
4.2 Choose your hypotheses
4.4 Listen to what the numbers have to say
Analytics in a nutshell
For businesses, the main goal is [usually] to get more active customers and ensure the business is ROI positive (that the return on the investment you made to bring these people in is above zero).
The point of marketing analytics is to understand user behavior in order to influence it.
Once you know what people do on your site, you can start hypothesizing why the act the way they do, test your hypotheses and optimize your conversion funnels accordingly in order to improve your bottom line.
It sounds simple enough right? Here’s a breakdown of what you should do, what will make you get it wrong and how to do it right:
The 2-step process to insightful marketing analytics:
1. Data collection
It probably goes without saying but I’ll say it anyway – if you don’t have data then you have nothing to work with.
Thanks to the major advancements in online analytics this is an incredibly easy problem to solve – make sure every user interaction you have, marketing or otherwise, is comprehensively tracked (everything from the pages a user visits on your site, and the order of pages he sees, to what he purchases and when).
That means implementing Google Analytics, MixPanel or even something you built yourself – you need to be able to know exactly what actions every person who got to your website did [and didn’t do] by tracking his entire journey.
Define every single activity as an event (for instance – “clicked a CTA button”) so that you have all the data you need in order to interpret the behavior of your users.
Not tracking everything (and again, I mean absolutely EVERYTHING, including but not limited to the color of a button, the location of the CTA he clicked on the page, the amount of time he spent reading the terms and conditions, everything!) means you’ll have blind spots and when you have blind spots you start making assumptions based on your intuition and that’s when you start making big mistakes.
2. Data interpretation
Even if you have all the data in the world, you could still get it wrong because data interpretation is hard.
Marketing analytics is about understanding the bigger picture and how the smaller moving parts fit in to it.
For example, after implementing Google Analytics tracking you might have discovered a 14% drop in conversion from visitors to registrations this month vs last month.
But 14% is just a number, it doesn’t mean anything good or bad, you need so much more information to understand it in context.
For instance, maybe you’ll find that the composition of visitors to your site this month is different compared to last month – this month you have more returning visitors (people who have already been to your site before, some of which are already your customers) and less new visitors than the month before.
If you have more existing customers this month, it makes sense you’ll have less registrations since they’ve already registered.
This could be because a strategic decision was made to focus on retention not acquisition (getting existing users back rather than getting new users).
If this is the case, a 14% drop in conversion might not be bad after all, it might even suggest that your retention efforts are paying off.
On the other hand, maybe no such decision was made and the 14% drop is a result of a marketing campaign that was supposed to bring in new users but didn’t deliver. That would be bad.
To sum up, data interpretation is about so much more than data, it’s about context.
The changes found by analyzing the data will point you in the direction you need to further examine, but it’s the other stakeholders in the company that’ll give you the context you need to decide what the data means and what actionable insight can be derived from it.
David Packard was quoted as saying: “Marketing is too important to be left to the marketing department.”
While I disagree with the natural interpretation of this quote as saying that a marketing department is not capable of doing the sole job it was hired to do, I do agree that no department, definitely not marketing, can do its job effectively in isolation.
If your company includes more than 1 person there is bound to be some informational gaps between people involved and in order to build a smart marketing strategy based on the real meaning behind the numbers in an analytics report, that gap has to be bridged and all departments have to work together.
Why do we interpret data wrong?
Let’s assume that you have a high functioning company that shares information and works together in order to grow the business.
That still might not be enough to overcome all interpretation errors. Here are 2 marketing analytics pitfalls you need to be careful of every time you look at numbers:
Intuition about causality
There really dumb thing about us humans is that we can never tell the difference between causality and correlation.
It’s super simple: correlation occurs when two things happen (or don’t happen) at the same time. Causality on the other hand is when one thing happens (or doesn’t happen) causing another thing to happen (or not happen).
Intuitively, when we observe 2 things happening, we immediately jump to the conclusion that one of them influenced another. While it’s fine to take such intuition and test it, it’s quite different to take it at face value and just assume it’s true and then base your behavior on that “fact”.
For instance, let’s assume you advertise you online business 24/7 and find that you get more registration at night time than at day time.
You intuitively think – wow, that means people prefer to buy at night, I should stop wasting my money during the day and just advertise at night.
So you stop advertising during the day and focus solely on the night. Your registration numbers drastically drop. Why?
Maybe those people who completed registration at night are actually people who saw your ad during the day, chose to learn more or think about purchasing and then completed registration at night?
What this example means is that while you can let your intuition formulate hypotheses for you, never act upon intuition without first performing small-scale tests first.
And that leads me to the second pitfall:
Small sample bias
Another thing most of us fail to recognize is that we can’t predict future actions on a small sample size.
In the beginning of the post I gave the example of finding the roads completely empty one day and deciding that this meant that at this hour, the roads are always empty. Clearly 1 day was too small a sample of days.
This is true in marketing analytics as well. If you don’t have enough data points (enough people performing an action and then another action), your predictions will have too much “noise”, they will not be accurate and in many cases, will simple be wrong.
So before you start analyzing the data you have, make sure you have enough of it.
What’s a good sample size? Try this SurveyMonkey sample size calculator
How to translate numbers into insight the right way
Here’s a summary of the 4 steps to do marketing analytics the right way:
1. Start with a goal
Every business activity you do must have a goal – something you want to achieve.
Since there will always be more data available than time to analyze it, the goal will give you the context in which to analyze your results. It will tell you what to focus on.
2. Choose your hypotheses
Once you have a goal, you need to make some hypotheses on how to reach that goal.
You hypothesize that X will lead to Y with Y being your goal.
You should test out a few hypotheses.
If you think there is more than one way to achieve Y, test a few of them out to find the optimal one in terms of time and money.
4. Listen to what the numbers have to say
One your tests have been completed (you’ve collected enough data per test group for the sample size to be significant), it’s time to interpret the results.
Gather all the information you have, both the marketing analytics data as well as any and all information about the test conditions you did not control (do this by asking all other company stakeholders what activity they ran during your test period that could have influenced results).
Once you have all the information, let the numbers speak to you – did your hypotheses hold up? Did you achieve your goal? If you did, scale up the efforts that worked. If you didn’t, test out new hypotheses until you do.
Have something to add to this post? Have any question or comments? Contact me here
Don’t forget to check out the homepage for some more How-To guides and insight into online marketing dilemmas