How to use Artificial Intelligence to boost ROI from advertising
For marketers, the tide of artificial intelligence has finally crashed ashore.
It has begun making waves at industry events like WSDM: The 9th ACM International Conference on Web Search and Data Mining Web Search and Data Mining conference in San Francisco. In his keynote address, Google research fellow Jeff Dean told attendees it’s time to embrace techniques like deep learning. Once a land of buzzwords, rapid improvements in machine learning have made AI a pivotal member of marketer’s toolkits. As best practice guidelines on Machine Learning at Google shared by Martin Zinkevich (technical) show, Google are actively using Machine Learning in many projects.
Dean and others are excited about AI because it will finally allow marketers to take full advantage of the hordes of data they’ve begun collecting. While we’ve been able to generate mind-boggling mounds of data for a while — a single connected jet engine, for instance, can produce 844 terabytes of data in just 12 hours — we’ve struggled to parse it in a meaningful, time-effective manner.
And yet, relatively few marketers are using the data dissection tools available to them. Today, a whopping 61 percent say they’re struggling to understand their own customer touchpoints. Think about that. Billions of dollars are wasted each year on invalid advertising, annoying and alienating people who don’t find it relevant.
Do marketers want to waste their money and customers’ time? Of course not. It’s just that, without thousands of hours of painstaking analysis, spotting ROI across dozens of channels and customer segments is incredibly difficult. Most marketers aren’t trained data scientists, but they’re being expected to wrangle data from social, programmatic, content marketing, traditional marketing, SEO, and much more. Increasingly, they’re realizing they need someone — or something — to light the way.
A Beautiful (Artificial) Mind
That’s why experts like Dean see so much promise in AI. It can craft incredibly precise profiles, free marketers up for creative work, and even give them a second sight. The following areas of marketing are ripe for AI:
1. Personality profiling
Marketers live in the land of customer personas. They collect demographic and psychographic details to better understand exactly whom they’re trying to reach.
AI puts yesterday’s personality profiling on steroids. It groups customers into audience pools based on touchpoints and sentiment analysis, which helps marketers understand how various customer segments might react to a social post, billboard, or blog. By taking into account the way customers talk to one another, it can suggest phrases and moods that resonate best with each audience segment.
If that sounds futuristic, think again. Adobe’s Target is an early-but-excellent option for marketers. The software aggregates customer data from online and offline sources ranging from social media to CRM databases. It delivers insights about specific audience pools, enabling marketers to create hypertargeted offers, optimize webpages and other communications, and even find new customer pools using lookalike modeling. In short, applying the right algorithm to the right data set opens doors to new customers while widening existing ones.
2. Augmented intelligence
We’re well past the point where computers can perform certain tasks faster than humans. Machines are now capable of doing jobs that require serious critical thinking skills and, in the past, years of education. Software can perform professional-level work, from teaching college-level computer science to predicting winning stocks to weighing moral questions of right and wrong.
That isn’t to say, however, that machines will ever completely replace marketers. Instead, machine learning is augmenting human capabilities, helping marketers accomplish more with fewer resources. In the past, it might have taken 20 strategists several weeks to analyze a data set and develop a marketing strategy around it. Now, with the help of a machine learning algorithm, two strategists can analyze the same data in less time and gain deeper insights from it — provided they ask the right questions.
That last part is key. Machines are more capable than ever of providing answers, but those answers are useless (or worse) if marketers fail to ask the right questions. That takes the sort of experience, intuition, and knowledge that only human beings can supply.
3. The great unknown
So how can two strategists with an algorithm do work that once required 20? It’s because AI can spot hidden connections in data that human marketers would likely have missed. But while AI can see trends that would’ve otherwise gone undetected, it’s up to humans to determine whether those findings are interesting or important.
A perfect example comes from the field of medicine. Last August, doctors solved a life-and-death medical mystery using neural networking. In just 10 minutes, the program cross-referenced the patient’s medical history against 20 million oncological records. It discovered the patient had a form of leukemia different than the one diagnosed by her doctors.
Again, it’s up to people, however, to interpret that data. In medicine, a genetic anomaly may be the cause of cancer — or it may be an entirely benign quirk. The world contains all sorts of spurious correlations: It takes a marketer’s critical mind to know which truly matter.
Putting the ‘Brain’ to Work
Even in its early stages, AI offers incredible possibilities to marketers. So if you’re ready to use machine learning to guide your media spend, here are a few suggestions.
First, if you’re getting started, there’s a good chance your learning algorithms aren’t very robust. You can use ensemble models — which rely on one or more learning algorithms to create a training loop — to gradually develop a more powerful decision maker.
Brieman and Cutler’s Random Forest machine-learning algorithm is a good example. It improves upon bootstrap aggregation models — better known as “bagging” — because rather than looking through all variables and values to arrive at an answer, Random Forest searches through a random sample and combines the results. The result is a solution with a smaller variance, reducing the likelihood of an erroneous answer.
Ensemble models really shine when you need multiple approaches to a training data set to achieve a desired output. For instance, the results of Random Forest, k-nearest neighbors (KNN), and Naive Bayes models may be merged to create an ensemble more powerful than the sum of its parts.
If this sounds complex, that’s because it is. Consider taking a machine-learning course online to understand the fundamentals. Stanford and Coursera offer an excellent one with free lectures and a paid certificate option.
Alternatively, check out one of the many great AI podcasts online, such as TWiML & AI, which covers the week in machine learning. Our own, AI Review, takes a decidedly human look at real-world business applications of AI. If podcasts and online courses aren’t enough, consider hiring a partner who can help incorporate machine learning into your media plan.
Now is the time to move. Machine learning and AI aren’t technologies of the future; they’re here now, and they’re here to stay.