We are going to make the prediction that 2018 is the year where machine learning really comes into the mainstream in digital marketing. The shift will be so radical that some digital marketing professionals will find themselves out of work, especially if they don’t keep up.
First, let’s step back and talk about how performance marketing, and particularly, media buying, have changed over the years. Digital media buying has gone through three phases:
- Phase 1 – Just Struggling to Get Online. In this initial phase, companies were just trying to figure out when and how to buy online ads. Should they be on Facebook, Twitter, Instagram, etc.?
- Phase 2 – Sophisticated Manual Optimizations. In the second phase, companies realized that by employing small teams of super smart analytical types and using reams of data, they could optimize their ROI by an order of magnitude or more. At TeamSnap we use several hundred dashboards and hundreds of metrics to optimize our return on ad spend (ROAS). Ultimately most decisions come down to humans reviewing the data and making decisions about what types of people are most likely to buy. This can range from reviewing the keywords you purchase, to the demographics you bid on, to the channels you use, and the creatives you design. Every time we add more data to the mix, we improve the ROAS by fine-tuning the ad buying strategy. However, there is a real limit to how far you can take this. Even with rocket scientist types on your team, and the best possible processes, it becomes very difficult for humans to keep track of more than a couple hundred metrics (and most people cannot even do that). That is where phase 3 comes in.
- Phase 3 – The Machines Do the Optimizations. In phase 3, we turn many of the ad buying decisions over to machines using machine learning. Instead of looking at just a couple hundred pieces of data, machines can look at millions or billions of pieces of data. In phase 2 we often use very blunt tools to optimize spend. We use demographics to bid up/down or exclude major groups (e.g., based on age or gender). Just because people age 18-24 have historically had a lower conversion rate doesn’t mean that the optimal strategy is to exclude all people age 18-24. Machines can sort through the data and figure out that the people most likely to buy are ones who include certain keywords in their social media posts, emails, or chats. Alternately, they like certain types of content or exhibit certain types of online behavior that is very specific to your product. In the same way that moving from phase 1 to phase 2 produced an order of magnitude jump in efficiency, moving from phase 2 to phase 3 has the same potential.
We have seen some elements of phase 3 already. Some examples include look-a-like audiences on Facebook and setting your campaigns in AdWords to deliver the ads that Google determines are most likely to convert. However, the big ad platforms have really gotten the machine learning bug and are rolling out lots of products that leverage machine learning. For example, Google has their Smart Display ads that use machine learning to find the right audience and their Universal App Campaigns that create and optimize ads for you. Facebook has their own machine learning initiatives. For example, you can upload LTV by customer, have Facebook analyze the commonalities across high-value customers, and in turn, find prospects that look like your high-value customers.
This is not to say that the machine learning tools on Google, Facebook, and the other ad platforms have reached maturity. We are still in the early stages, but we have gotten to the point where phase 3 is starting to become more efficient than phase 2, and the tools are going to evolve rapidly, making phase 3 more and more efficient.
So if you are a digital marketing professional, where does that leave you? First, if you spend your days creating lots of campaigns by hand and analyzing hundreds of metrics trying to fine tune your campaigns, your current job is going away. Period. Luckily, a new, related job is opening up. People will be needed to set up these machine learning campaigns, compare the results across machine learning tools, and analyze the results at a higher level. In particular, machine learning is only as good as the data you feed it. You just don’t have to deal with a lot of the drudgery that is required to buy and manage ad campaigns today. Said another way, the really sharp media buyers will be even more valuable in the future if they excel at deploying machine learning, particularly in feeding it the right inputs and optimizing toward the right outputs.
While this sounds simple, it actually is incredibly complex.
For example, when you pick your goals for the machines to optimize to (e.g., LTV or conversion event value), you have to think long and hard about how you calculate them. Do you care about time value of money in LTV? What costs are fixed and what are variable when you think about conversion event values? Machine learning operates so quickly that it can burn through an enormous amount of budget optimizing for the wrong thing if you are not careful.
Media buyers are not the only ones at risk – machine learning will cause additional entropy in the ranks of digital marketing execs. If you are stuck in phase 2 (or phase 1 where many companies still are), you can quickly be upstaged by competitors who have advanced to phase 3. That can mean that your company suffers and/or that you lose your job to someone who can leverage machine learning across your team.
2018 will be the year of machine learning in digital marketing. Now is the time to make sure you take advantage of it instead of it taking advantage of you.