A recent exposé in the Wall Street Journal told the story of Engineer.ai, a company that duped investors and customers with a mobile-app engineering platform that purported to use Artificial Intelligence to automate design and programming tasks, while in reality the work was done by inexpensive human programmers. The Journal’s story shined an uncomfortable light on the problem of “AI-hype” in tech, financial and media circles, whereby companies and investors are quick to shower money and attention on anything that has “AI” in its description.
The marketing world hasn’t been immune to AI hype in recent years, and many are beginning to wonder if machine-learning and AI have real, practical applications for marketers who often find themselves awash in data with little idea of how to use it. In order to answer this question, it’s first important to define what we mean when we talk about AI, and what necessary preconditions it needs to be both dynamic and useful.
To talk about AI in a marketing context, it’s important to make a distinction between batch learning, which is what happens when a computer program is given a large batch of data to digest and then spits out an operating algorithm meant to maximize certain outcomes, and dynamic or “online” learning, whereby an operating algorithm self-improves and adjusts according to a constant, live stream of data. Batch learning, the first generation of machine learning, has been around for decades. Batch learning originally became associated with the term “AI” because the programming was sufficiently complicated to make predictive conclusions and implementations from models that were too complex for human oversight.
Dynamic or online learning models build on the foundation of early batch learning, but add in the ability for the programming to adapt to a live, always-on stream of shifting variables. A good example of this might be a modern martech platform that is constantly adjusting the content directed at sales prospects according to uninterrupted data feeds from sources as varied as individual browsing histories, market trends, and social media activity.
While modern dynamic AI is quite good at executing the granular tactics of performance marketing campaigns, it still relies on human guidance. Ambitious companies with vast resources (think Google) are still hard at work at developing AI that can effectively adapt to real-time changes in consumer behavior by instantly building and propagating new models on the fly.
Most versions of AI available to the modern marketer still aren’t capable of making high-level decisions on their own. If you suddenly get inspired and you want to start targeting ads for rain gear according to the weather forecast, you’re still going to have to manually source that data and plug it into your marketing platform. Most available marketing software platforms don’t have the programming or the autonomy to make such decisions themselves.
What’s changed is that modern AI-enabled marketing platforms can adapt to new data sets in real time. A modern platform with an AI functionality will look at how weather data stacks up against sales data in this latest example and begin to track trends, build consumer behavior models and make adjustments accordingly. An older, batch-learning approach would need to be taken offline and fed large, existing data sets (which are necessarily retrospective) to generate models, which are static and outdated the minute they launch.
The end result of this shift has been to make AI-enabled applications much more accessible to marketers. Self-improving, dynamic platforms are less expensive and easier to implement than ever before, and the efficiencies and ROI they promise are there for the taking, they just still need human guidance to point them in the right direction. A modern marketer still might need consultation from a vendor to implement an AI solution, but it’s less likely that the marketer will need to have a PhD in computer science. Modern AI has made the entire value chain cheaper and easier, and most importantly, more effective.
The efficiencies available with modern AI are simpler, cheaper and easier to use, but they can’t yet solve the whole equation by themselves. Spotify is probably pretty good and recommending music you’ll like, but it can’t yet compose music you’ll like. That promise is still a ways off in the future, but with computers self-improving every day we certainly seem to be marching in that direction.