In our increasingly technological world, digital advertising has solidified itself as the norm for a majority of marketers. It has experienced double digit increases year-to-year in the past decade, with a 17% year-to-year growth from 2018 to 2019. Alongside this boom comes more pervasive use of artificial intelligence (AI) in digital marketing — experts like Gerry Murray, research director for IDC’s Marketing and Sales Technology service, cite that “AI is proliferating in every major marketing cloud in nearly every marketing point solution, already proving to be a powerful tool when leveraged by marketers”. This emerging trend makes sense: In a recent BlueShift and Kelton Global study, 58% of marketers stated that AI was able to help them achieve higher revenue. Yet, Murray also states that “most organizations are still experimenting with [AI]”. The question for digital marketers in the coming years, then, is not “Should I use AI?” but “How do I use AI effectively?”. AI is undoubtedly powerful, but it doesn’t come without limitations and concerns. It is key to understand these limitations and work to evolve the modern marketing team with the technology, instead of becoming overly reliant on what — despite its power — is still a tool.
In navigating the expanding web of software offerings that integrate advanced technology, it is easy to get lost in the jargon used by computer scientists. Buzzwords like machine learning (ML), deep learning, and neural networks all have distinct meanings, but they are often used interchangeably to refer to this suite of digital tools. Simply put, AI is any sort of machine intelligence that can solve problems that typically require human intelligence. ML is a subfield of AI that focuses on learning a behavior from data — for example how to best promote ads to consumers given data on the success rate of past strategies. This breaks into further subcategories that represent different approaches to learning, as well as various techniques. Deep learning, an especially popular technique, uses a multi-layered neural network (think interconnected layers that pass information like neurons in our own brains) to learn complex, unlabeled relationships within a dataset and then utilise this understanding to leverage new data.
Marketing presents a prime use case for AI: Time and financial constraints make it impossible to promote a tailored advertisement to every customer. However, AI theoretically provides an efficient and effective way to optimally distribute resources, offering speed and insight based on past data that humans wouldn’t be able to achieve. Commonly this is achieved using deep learning, which requires large sets of data to work, but the millions of daily users on e-commerce platforms provide this data. Deep learning is able to find patterns from this vast amount of user traffic, which the marketing team can then leverage for growth.
Programmatic ad display is one area of major use; AI optimizes which users and websites ads should be pushed to, dynamically learning what choices result in clicks and purchases. For example, using the AI tool AdTheorent, Norwegian Airlines experienced a cost-per-booking CPA 170% lower than the target. AI is also used to develop more robust user profile segmentation: Nuanced relationships between the traits of different users are picked up by the program, powering ad distribution that human marketers might not be able to achieve. In another case study, The Humane Society drove higher pet adoption rates when the AI learned that since pets are a family decision, targeting ads to non-purchasing family members would drive eventual adoption. Good marketers would have learned this eventually, but in this case the speedy insight was made possible by the creation of user profiles that identified a connection to the original purchaser.
However, AI is not a panacea, and requires both management and shaping to attain its strongest effects. AI has an incredible ability to ascertain unexpected relationships between data, but those relationships are complex. Focusing solely on favorable ROI or CPA statistics instead of why certain tactics are working can end up distracting a team from the underlying relationship, which will be especially true as AI grows within the sector.
Users and creators of AI tools also have to remain cautious of the decisions AI makes. While detailed relationships can be found in data, this relies on the validity and neutrality of the data. If a dataset is biased, the relationships learned will be biased. Without careful monitoring by the team using AI, misuse can occur — and already has. Predictive tools used in the justice system have been found to disproportionately target African-Americans and other people of color. COMPAS, a tool used by the Wisconsin Department of Corrections, uses AI to calculate a recidivism rate that is then factored into the length of criminal sentencing. While the Department of Corrections argued that the tool was fair as race wasn’t programmed in as a factor, studies proved that the AI learned the human-based racial bias against African Americans from past cases, incorrectly predicting a higher recidivism rate.
Teams that use AI need to be incredibly vigilant because of situations like these — AI learns existing relationships, not targeted ones. While marketing cannot improperly jail someone, AI-based advertising can reinforce stereotypes and even reduce opportunities. For example, women being underrepresented in managerial positions in the present can lead to ads targeting jobs predominantly towards men, as in the figurative eyes of the AI, this is who would be more likely to take the position.
It is hard to argue the power of AI, or that it has already become a larger part of the marketing world and will inevitably continue to grow. The marketing leaders of the present and future need to then understand how to best leverage the tool, but also how to train their team to work alongside AI instead of becoming too overly reliant on its decisions. This will provide a long-term monetary benefit, but also an ethical one. AI doesn’t learn to be moral or have what we would consider “common sense,” and needs humans to keep it in check. There is still a long way to go until artificial general intelligence, and until then it is imperative that AI is managed responsibly.