How machine learning frees up creativity and strategy for marketers

Artificial intelligence (AI) and machine learning (ML) have been massively hyped over the years. These days it seems every company is an AI/ML company — and reality is, as American researcher, scientist, and futurist Roy Amara, stated, “We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.”

When a new technology is developed or deployed, people often talk about it suddenly transforming everything in the next couple of years. However, we also tend to underestimate the effect of it entirely, especially if it is the kind of technology that could fundamentally change the way we solve marketers’ problems and interact with customers. If we’re going to leverage the full benefits of AI and ML, it’s important to first understand the technology and discern between the facts and fiction of how it works today. Only then can we understand what is real, how this technology can be transformative, and how machine learning and AI can free up creativity and strategic thinking for marketers.

Machine learning starts with data

Without the ability to analyze data, identify patterns, and put it to use, data is effectively useless. Machines are ruthless optimizers that can organize data on a level that is impossible for humans to replicate. However, this also works in reverse, as machines today cannot replicate the creative thinking and strategies that humans can generate and act on. The data optimized through the machine with machine learning provides marketers with a supercharged ability to make the most informed decisions and then enact a creative strategy to achieve their desired outcome.

Machine learning for marketers: Asking the right questions

The things that matter to companies and to individuals are decisions and actions. Back when I used to consult large companies spending millions or tens of millions on “data strategy” or equally poorly defined areas, I would often advise that before they start to worry about the data they need to collect, they need to start with what decisions and actions they need to take as a business. Starting from that perspective, businesses can ask themselves: What decisions do you wish you could make smarter and faster? Are you structurally set up as an organization to make those decisions? Once those are defined, you can then ask questions like, what information do I need to make these decisions faster and smarter? And which of these decisions can be automated? 

So, where does machine learning come in? Which category of problems can it help us with? In order to answer these questions, it is first useful to understand the limitations of this technology. ML does not replicate the amazing generality and adaptability of human intelligence — instead (and consistently with other technologies) it augments human intelligence and solves a more specific set of problems with superhuman capability. To work out if ML can be applied to a problem, the following set of questions are useful: 

• Can a human solve the specific task required in less than 2 seconds? (This is a rough estimate; we have not yet reached the point of solving problems more complex than this.)
• Is it valuable to solve this problem repeatedly at scale (e.g., billions of times incredibly fast)? 
• Is it valuable to do this task repeatedly, robustly, and consistently? 
• Can we measure “success” numerically? 

If you can answer “yes” to these questions, then you have a problem that is a great fit for applying machine learning. (Interestingly, these are also the kind of tasks that humans are terrible at because we get bored, distracted and tired!) This might appear very limiting, but many problems fit into the “yes” bucket, such as identifying spam emails, detecting fraud, optimizing pricing, and making sense of language.

Solving marketers problems with machine learning

When it comes to marketing and advertising, there is a whole category of problems that also fit squarely into that “yes” bucket. Detecting audience composition and behavior changes over time, predicting if an ad will lead to a potential customer visiting my site based on the contents of the article they are reading, and tuning thousands of parameters to ensure budgets are spent efficiently and effectively are all such marketing problems. 

There are also problems that do not fit into this categorization, such as: how do I convey my complex message in a way that cuts through the noise? How do I connect effectively with an audience with whom I am not currently resonating? How do I balance long and short-term objectives?

Machine learning is not magic: it can give marketers superhuman capabilities to find patterns in data to deepen our understanding, optimize delivery against well-defined goals, react to changes rapidly and rationally, and execute our ideas predictably, with less friction and more feedback

Interacting with customers in real time

For marketing, much of the information and patterns that are useful relate to customer behavior. Digital campaigns are markedly less effective when they are unable to respond to changing conditions at the moment. To illustrate, if you are selling gourmet coffee makers, you want to reach the people that are still interested in purchasing one, not those that had been searching online for the past week and purchased one yesterday. Everyone has experienced shopping online for a product, having it arrive, and then having every device and platform they use spam them with the same product repeatedly for the next week. While this may be useful for products that customers generally continue to buy (detergent, toiletries, etc.), most people only need one gourmet coffee maker. 

Not only does real-time data ensure that campaigns are reaching the right people, but it also allows marketers to respond to changing market conditions. By combining machine learning with real-time data, marketers can see results live, instead of waiting for results at the end of a campaign. This means brands can detect and capitalize on things like a popular, recently released Netflix show or what’s trending on Twitter, or even address the quickly changing dynamics within the supply chain. If there is anything brands have learned over the past couple of years, it’s that world events can impact shopping behaviors and patterns in an instant. 

While machines can take care of analyzing data around demographics, web browsing behaviors, and past purchases, having the right creative marketer — who can connect current trends to campaign goals and ensure the right questions are being asked of the machines — is what distinguishes a good campaign from a great one. To borrow another great quote, this time from Alan Kay, “Simple things should be simple, complex things should be possible”. In addition to helping us get deeper insight and understanding of audience behavior, great technology should also make it simple for marketers to react to this information by getting new creative ideas live in minutes, not months.  

Can ML predict the future?

Predicting the future is not possible. But machine learning technology combined with real-time data can enable marketers to understand emerging trends and behavioral shifts as they happen and make it easy to react to these changes by getting automatically optimized campaigns live in minutes and seeing if they are working within hours and days. True progress is about learning, and about testing strategies and ideas. 

The underestimated impact that ML will have on the ad tech industry over the next decade will not be due to AI-generated ideas or reduced dollars spent on operations that will materialize;  the big impact will come from shortening the gaps between marketing strategy, insight, idea and execution and from allowing us to understand more deeply and quickly, be more creative, and test ideas more confidently and easily, and measure impact more effectively. This technology — like all other technologies — is not to replace humans, but free us from the repetitive and tedious and empower us to be superhuman.

Peter Day is CTO of Quantcast