So there you are: you have millions and billions line of raw data. Regardless of which tools are you using to analyze it (Google Analytics, Mixpanel, Python, SQL, R, or even Big Data tools like Hadoop, Spark, and so much more…), you need to decide first what do you want to get out from your data?
There are usually 3 different ways of using data:

  1. Reporting
  2. Optimizing
  3. Predicting

I’ll dig into all of these a little bit more in the next chapters, but here I give you a quick overview first.

Reporting

Classic Google Analytics Report

Classic Google Analytics Report

When you are kicking-off your new startup, e-commerce store or any online business, you most probably will start with reporting. Reporting is very essential, because you have to know what’s going on with your business. The best thing is if you have reports on a daily basis. You check those right after your morning coffee at the office and right before you leave. It takes 5+5 minutes from your day and makes you very happy (or sometimes sad), but at least makes you confident, because you have facts. A good report is generated automatically day-by-day, so you only need to set it up once. And maybe change it once a year, when you are changing your company goals (let’s say, start to focus on retention instead of registrations).

But don’t stop at reporting. If you want to have a really cool product, you don’t only need to know things, but act on things. Unfortunately reports are not really actionable by definition, because their aim is – again – to let you know, how things are going generally. But when it comes to optimization, you will look after what do you need to change – specifically. So what do you need to do, when you start to focus on optimization?

Optimization

Qualitative Research --» Data Research --» AB-test --» Repeat

Qualitative Research –» Data Research –» AB-test –» Repeat

Well, I see more and more recently, that most of the companies are doing AB-tests only and nothing more, when it comes to conversion-rate-optimization (aka. CRO). Creative guys (designers, copy-writers, etc.) and project managers sit together, come up with nice ideas and just start their split tests.

Optimization equals AB-testing? No, it’s definitely not! Running split tests and do nothing else is an extremely bad practice! Actually running a test should be the very last step of your CRO-project. The 95% of your time should go with research instead: usability tests, user interviews, 5-sec tests, heatmap-analyses, funnel analyses, segmenting, cohort analyses, exit-intent surveys and so on… You should have a solid research background about the why. Why you want to change anything on your website or in your app? What data or researches confirm your hypothesis?

When I was working with smaller startups and e-commerce companies as a consultant, time-to-time somebody came to me, that we should AB-test this or that. After a while I developed a very simple process: I asked if they can show me at least 3 research results (5-sec-test results, Google Analytics chart, a heatmap, etc.), that proves that this could be an AB-test worth to try. If not, we just passed the idea.

Don’t get me wrong, gut feeling is an important thing, but you will always have better and quicker results, when your gut feeling is backed with data.

Predicting

prediction with decision trees

prediction with decision trees – source: http://www.r2d3.us/

The third most common usage of data is predicting. In one sentence: predicting is nothing else, but telling the future – based on your historical data.

Imagine that you are in a big grocery shop, you want to pay and you need to choose a line to stand in. Which line you choose? Of course the fastest one. Your brain knows – based on previous experiences – that this line will be the shortest one (with the less fellow customers). If you are an experienced shopper, your brain also knows, that the length of the line is not the only thing, so you quickly check the number of items in the baskets of the other customers in your line. Then if you are very experienced, you also check the cashier person – if you know him/her and you have an experience with him/her, where all thing went smooth, it’s also a positive indicator. And so on and so forth.

Your brain predicts based on historical data (past experience). It optimizes to a target variable (fastest line) and analyzes different explanatory variables (length, total items, cashier person, etc.) That’s what computers do, when they predict. The only thing, that they do this via specific statistical methods and with more accurate data. Spotify does this (or at least something really similar called collaborative filtering), when it gives you the Discover Weekly. Netflix does this, when it recommends you a movie. The very famous Google Flue Trends did this, when it predicted where flue will show up on the planet.
Let’s get back to prediction later this chapter.

Conclusion

Until then remember, use your data to:

  1. Reports, but don’t stop there, and spend time with
  2. Optimization – based on researches. And when you are a pro on these 2, let’s
  3. Predict and tell the future!

Have fun and stay tuned!

And if you want to be notified first about new content on data36 blog (like articles, videos, handbooks, etc.), sign up for the Newsletter!
Tomi Mester

Share:Tweet about this on Twitter3Share on LinkedIn7Share on Facebook0