From algorithms to advertising: 7 steps to introducing AI to marketing
Contributor Tomer Naveh provides a simplified look at how artificial intelligence (AI) is applied to automate digital marketing programs from start to finish.
Artificial intelligence — once the rarefied domain of big-name, ambitious projects like Google’s self-driving car or IBM’s Watson — is now finding its way into everyday business. In advertising and marketing specifically, brands might not be completely overhauling their existing ad tech and martech stacks to make room for AI just yet, but many are getting a feel for it by experimenting with single-touch AI solutions that focus on isolated tasks, like recommendations, ad buying and optimization.
The coming wave of AI in marketing will be defined by the automation of complex, multi-step processes — not just one-off aspects of a larger campaign. For brands, this will mean relinquishing control, trusting the technology to come in and quickly understand processes comprised of numerous tasks, channels, people and procedures, without messing things up.
Before handing over the reins, it’s helpful to understand how AI works — and how entire human thought processes are converted into algorithms. For all its complexity, here’s a simplified look at seven steps to introducing an AI that can automate holistic digital marketing programs from start to finish.
1. Obsessively observe and unravel every step in the process
Creating artificial intelligence for “self-driving” marketing technology is not so different from creating AI for a self-driving car. In the case of the car, it must know how close it is to other cars. It must know how to make a turn and ensure that it’s in the right position at the end of the turn, when to hit the gas pedal to go faster, what the road conditions are like, and so on — all without the driver telling it what to do.
Like driving a car, many of the thought processes that go into the day-to-day execution of marketing programs also happen automatically and largely on the subconscious level. Transforming these subtle processes into a tangible series of algorithms means isolating logic and reasoning that humans often aren’t even aware they’re engaging in.
This begins with the acute observation of marketers and account managers as they execute each step of a process, over and over again. Often, things that seem trivial — like determining which image and headline combo work best for Facebook, how much budget to spend where, or picking keywords for a search campaign — are critical parts of a larger process.
2. Understand why human marketers make the decisions they do
The AI doesn’t just need to know what steps to take; it ultimately needs to understand why each decision was made, whether it was based on experience, logic and reasoning or simply knee-jerk instincts.
This requires asking marketers and account managers to describe their decision-making process, which can be difficult considering that, as we’ve already discovered, they have no idea what motivated them half the time: Why did you keep these words and ditch those ones? How did you decide your bid size? Say you see a keyword doing well and you increase it by 20 percent — how did you choose 20 percent? What is the best time of day to send stuff to that person? Okay, what about that other person?
3. Teach technology how to understand abstract information, such as creative
Data in the form of words and numbers are unquestionably the domain of AI. So, what happens when technology is asked to process and make decisions that are more creative in nature?
For a human, understanding why certain images and text make more sense as a first interaction with a consumer rather than as a secondary or final interaction is almost second nature. A machine, on the other hand, needs to be told (or programmed) with this knowledge in order to be able to judge images and text, determine where they should appear along the journey, and not have to rely on humans to make these decisions for it.
4. Program the AI to do things humans can’t
AI can process millions of data points in a minute. And every minute, there are several variables — or combinations of variables — that will influence an exponential number of outcomes in any process. “If I do this, that will happen. But if I do that, this will happen.”
In a marketing campaign, maybe this is choosing which of the available headline options is best to use on a specific channel in this exact moment considering its specific audience, their known level of familiarity with the product and so on. The technology must be able to predict the performance of a single headline in relation to all others and specifically under these conditions.
5. Make individual building blocks work together as a holistic system
One major problem with multi-step processes in business right now is that more often than not, different parts of the process are handled by different people. Take a digital marketing campaign where one person is responsible for Facebook ads, another for search, another for Instagram, another for Twitter, another for display, email, SMS and so on. These people each receive different insights, which they use to calibrate their respective efforts.
But it’s very rare that insights from one channel are used to inform decisions on other channels. Even if the intent were there, it would be difficult, if not impossible, for humans to keep up with and apply those insights across channels. AI, on the other hand, is a pro at using any data it can get its “hands” on and applying it holistically, understanding the interplay of those discrete functions to properly relate and sequence the many moving pieces.
6. Introduce checks and balances so the AI doesn’t go rogue
Perhaps one of the most important things for AI users to understand is that their technology isn’t going to go rogue on them. That’s why introducing checks and balances is so important. These built-in rules ensure that the AI doesn’t make decisions that are so out of sync with what a human would do in a given situation that the technology is at cross-purposes with the people or organization it’s serving. This is the imagined scenario that allows evil robots to take over the earth.
This is especially true when it comes to any decisions related to budget. It could be, for instance, that algorithms predict that you should triple your spend on a given ad campaign. In this case, checks and balances would kick in to reflect the fact that this is unusual behavior and a choice that a person would approach with caution even if the rational logic is there from the machine.
A human in this position would want to understand the market conditions surrounding the suggestion, as well as the potential outcomes, so this layer of human precaution must be reflected in the machine, too.
7. Make sure the AI works autonomously and at an impossible scale
Finally, all of this must happen autonomously. Artificial intelligence is not interested in doing the same things as us flawed humans at the same pace and scale. Good AI understands everything that goes into a process and then does it way better and on a far greater — impossible-for-humans — scale. If a human can realistically manage a search campaign with 500 keywords, artificial intelligence wants to manage 200,000 keywords.
Opinions expressed in this article are those of the guest author and not necessarily MarTech Today. Staff authors are listed here.