Would you prefer Big Data or Big Useful Data?
We all know online marketing is here to stay, to grow and to develop even more in the next years and decades. The good news is that, by now, marketers no longer need to prove in the boardroom why they’re spending so much on online and social media marketing, today is all about how they’re spending the budgets and how it actually affects the turnover. So it’s all about proving conversion and ROI, really. Nothing ever really changes, I guess…
Well, today is #conversionday at the Brussels’ Event Lounge with 300+ online marketers attending @bloovi ‘s yearly Conversion Day. One of the speakers this morning was Dr. Maurits Kaptein (@mauritskate). Maurits talks big about big data. And he does so quite effortlessly and naturally. Without slides. A welcome rarity at conferences if you ask me. Maurits studied Economic psychology, Computer science, has a PhD in statistics and currently works for the Tilburg University. Furthermore he runs Science Rockstars: a big-data agency specialized in personalized online algorithms computing solutions.
He kicked off his talk by stating we all know that big data works. A simple and world-known case is Amazon’s recommendation algorithm of course: derived from data of other people, it makes you buy more. Very simple, very useful. But for a lot of companies ‘big data’ still stands for ‘big problems’. For a lot of organizations ‘big data’ seems to be the solution to an unknown problem. Here is Maurits’ take on the whole big data question mark for e-businesses, summarized in 3 big lessons:
1. If you want to use big data, focus on useful data.
Collecting, testing, logging, tracking data is easy, but it has the disadvantage of endlessly adding millions and millions of new data points every second to the pile of data. This is not the way forward. Start selecting the useful data from the data-entry moment onwards. Focus on logging what make people buy things for example. Don’t worry and over-test background colors or stuff that does not have any impact on consumer behavior. Focus on logging data that actually psychologically influences people to buy.
Social proof, scarcity, price sensitivity… those are the powerful mechanisms you want to track and collect big data from.
2. Leap from analytics to actual decision taking
Doing something with the collected data is key. And that is not about creating cool graphs or dashboards showing great visualized data. It’s about the decisions that subsequently follow from actually interpreting the graphs and observations. That’s typically what marketers do. But we often make wrong decisions based on the data, misinterpretations and subjectivity tend to creep into the decisions. And it costs a lot of resources and time from a marketing team.
The new wave in big data decision taking is outsourcing this decision taking to computers. Based on the mathematics principles behind ‘reinforcement learnings‘, computers can run the show by themselves. Think about it metaphorically as a ‘multi-armed bandit‘: suppose there are two slot machines with two arms in a casino and you have $1,000 to spend. In classical A/B testing, you would spend half a dollar in either machines and see which machine will deliver the most money. However, if you let the principles of probability measurements in and leave it to the computer to decide, your overall winnings will increase. How? Let the computer randomize every single half-dollar bet by flipping a coin and betting on machine A or B. Keep randomizing, but change the probability in favor of the machine that delivers more money. Repeat this until you have spent your $1,000 and you will end up with more money than if you had done an A/B test with a small part of your $1,000 and then gone all-in for A or B. So Maurits pleas for mathematics and machines to make the decisions rather than people and simple analytics.
3. Invest in the infrastructure, don’t just scale your small data methods, but truly change
This is a bit of a technical point, but still an important one. Since a decade or so, a lot of companies have solid databases running, great CRM packages containing huge data, adequate counting and tracking software and large online marketing teams to do something with the data. And that is exactly the challenge: how can systems put in place 10 years ago by people trained 15 years ago cope with today’s data abundance?
So how to scale this shift effectively as a company?
Well, by moving from offline to online analysis seems to be Maurits’ somewhat technical and dare I say ‘nerdy‘ answer. Counting and adding lines is the offline, old-school method. Instead use a counter rather than to count. In other words, start processing the data in a data stream. Treat and aggregate data as a stream filling a sales or conversion KPI rather than to collect separate data files one by one by one. That way you win time, efforts and just as important: server space.
If there is one thing that became apparent during this talk, it is that most companies are struggling with big data and need a lot of educating before wanting and being able to actually take the next steps.