Category: Technology

Many will tell you that we are already in the age of Big Data, and maybe we are. But the truth is that whatever pile of data we have today will look puny in just a few short years. Big Data keeps getting bigger, so whether you think we have already arrived or Big Data is around the corner, how can CMOs be ready for what comes next?

It’s not that complicated, actually. The most important step any CMO can take is to move to a data-driven decision-making process. What that means is no more “golden gut.” What that means is that we don’t ask the CMO what to do–we ask the data what to do. It doesn’t matter whether the data we have now is big or medium-sized. The truth is that most companies do a crappy job with small data.

How would your job be different if every time someone asked you to decide something, you asked what the data shows? And if there is no data, you asked how we could get some? And if no one knows how to get the data, you ask what kind of experiment can we run to get the data?

For some CMOs, that’s very threatening. They believe that their job, like any good executive, is to make the tough decisions. And that’s true. But what if the toughest decision you will ever have to make is to give up your own personal control and seek the data to make the decisions?

Data-driven decision-making is a culture change. It says that instead of believing in our own opinions, that we will focus on our customers’ opinions. When we focus on data that tells us what our customers want, it’s amazing what geniuses we become.

It feels like data-driven marketing is a loss of control. But what we need to accept is that depending on our own opinions is the most out of control we can be. Each decision that we make personal might feel under control, but competing against companies that decide based on data will quickly make us extremely uncomfortable–because they are making better decisions than we are. And in just a few years, we won’t be making decisions based on data–we will be providing rules for computers to make decisions based on data. We’ll do that because people can’t decide fast enough.

Have you reconsidered your reliance on data for decisions? Moving to data-driven processes prepares you for the onslaught of data that is coming for all of us.

Machine learning seems to be the flavor of the week, so I’ve been getting a lot of questions from clients who want to know what they should know. And some requests for some kind of magic that sounds like, “Can you ladle out a dollop of that machine learning on this?”

So, first off, machine learning is a kind of artificial intelligence–that sounds lofty, but what it really means is that it provides a way for computers to solve problems that they weren’t explicitly programmed to solve. Standard software can solve a problem because a programmer researched the problem, talked to experts, and designed a solution to that particular problem. Machine learning allows computers to tackle a range of problems that the programmers and the experts wouldn’t necessarily know how to solve on their own.

Let’s take an example. I’ve designed a machine learning system for a company that performs sentiment analysis for social media conversations–such as whether a tweet is positive or negative. Now, human beings can identify sentiment accurately around 89% of the time (yes, that low), but standard software is much worse–maybe 60%–because it is hard to code a set of rules that cover all the cases. This machine learning solution is actually more accurate than human beings at times.

How does it do that? It collects training data from what the human beings decide and it looks for patterns that help it identify which tweets are more likely to be positive or negative. With enough training data, it can do a better job than a person. And it is certainly much faster and cheaper.

So, of course you’d like such a technology applied to all sorts of marketing problems that you have. But you might not be ready. Recently, a client approached us, asking that we replace their human content approval system with a machine learning system. They explained that their current process takes each new web page and subjects it to the judgement of several human experts (legal, brand, product, and more) before it is approved to be promoted to the production website. The current process is frustrating and time consuming–often two weeks go by before the approvals are in hand. They desperately want to get the approval process to a day–or even minutes.

But they aren’t ready yet. Machine learning requires that you have very tight processes with well-defined tasks and a history of data that shows that human experts generally agree with each other (such as in the 89% sentiment agreement). This client had none of that–no written standards, experts who disagree, no records of prior decisions–so they weren’t ready yet.

That’s OK. We walked through how they can fix all of those problems over the next few months. Just doing those things will make the process more consistent, less frustrating, and a little faster. At that point, we can start thinking about machine learning to gain even more. It’s always important to provide business value at every step–even what seems like the preparation step.

So what are your thoughts on machine learning? Anytime something new comes along, it can be a personality test. Some people say “we don’t need that yet because it isn’t proven” and others say “we need to do that now and get a jump on the competition.” For machine learning, it’s best to analyze the situation and start preparing your processes to take advantage–because the day is coming where no one will say it isn’t proven yet.