Is AI for Marketing All Hype? 3 Experts Weigh In
Source – martechadvisor.com
If you search “Marketing and AI”, Google returns almost four million results. Recent headlines range from the strategic – AI Will Make Marketing Less Manual (well, duh) – to the tactical, like 4 ways AI can improve email marketing. The hype machine is in full swing. But here’s the rub: the vast majority of the conversations about marketing and artificial intelligence aren’t about AI at all. They’re actually talking about machine learning, and that’s not the same thing.
This is a hot topic at my company, Conversion Logic, where we use machine learning algorithms for marketing attribution. It’s tempting, from our own marketing perspective, to jump on the bandwagon and “AI-wash” our offering. But our leadership and the experts behind our data science aren’t having it. I spoke to the team to get a better understanding of what AI for marketing is and isn’t, the difference between AI and machine learning, and how these concepts will apply to marketing today and in the future.
Brian Baumgart, CEO at Conversion Logic
Artificial intelligence (AI) is seemingly everywhere in the media today, with sentiments ranging from fears of a jobless future to unbridled claims of capturing once-impossible opportunities. In martech, and especially in marketing performance measurement, AI is headlining claims of superior analytics precision, automation, and unrivaled predictive capabilities. As is usually the case, the realities of what is possible and the resulting benefits are quite different outside the hype machine’s distortion field.
The first issue revolves around what AI actually is, particularly in relation to machine learning. “AI” is sometimes broken down into general and specific, with the former classifying the AI we all know and love from science fiction – a human-like intelligence capable of learning nearly anything at an inhuman scale and pace – while the latter typically refers to the application of specific model, or algorithm, sets that perform highly specialized work.
Machine learning (ML) is indiscernible from specific AI in most contexts. At its core, ML consists of an algorithm, or sets of algorithms, that solves very specific problems, such as creating a highly accurate predictive model at the touchpoint level from billions of advertising exposures.
Appreciating this difference between AI classifications and ML allows us to see through the fog of PR and brand positioning to a clear view of what the final, achievable business results can be. Want higher overall media efficiency? Campaign, brand, and audience goals met more effectively? Deep, actionable insights into what is working and what is not? Recommendations across and within media?
All of these complex, highly specialized questions are answerable at previously unachievable levels of accuracy, but only as a result of applying the right algorithms to the right problem spaces and then having a data science team expertly tune these systems using deep subject matter expertise.
Dan Cox, Head of Product at Conversion Logic
What’s really in a name? Quite a lot, it turns out, if you’re interested in achievable end results. While it’s true that AI will undeniably be the intelligence backbone of likely all technology systems at some future time, our current machine learning systems will still be critical for context-specific, specialized applications aimed at answering the same fundamental questions we have today.
What’s the difference? AI as people talk about it – a system that learns and improves itself on its own – does not exist right now. Machine learning, on the other hand, is very real. The method we use most often in marketing is called reinforcement learning. In reinforcement learning, the machine learning system interacts with its environment. Based on the feedback (penalty or reward), it updates itself to maximize or minimize a defined cost function.
This concept is applied by researchers to develop systems that learn to play a video or strategy game (like Go or chess) by playing it with itself and others. Google’s deepmind alphaGo beat the world’s top human Go player using this method.
To use this framework in marketing, for optimization in particular, the marketing measurement and media buying systems need to be tightly integrated. The full system will then start buying ads with different attributes (various publishers, placements, creatives, strategy, etc.) Based on the instant feedback, like clicks and conversions, and its internal attribution system, it will then update itself to buy more efficient ads to maximize/minimize a desired KPI(s). The marketing practitioner’s job will be to define the cost function that will define the reward or penalty for the system.
We are not there yet, but I think this scenario is quite possible within next 10 years, once adtech and martech more tightly integrate with each other.
Erkut Aykutlug, Senior Data Scientist at Conversion Logic
Artificial intelligence (AI) is a system that “perceives its environment and takes actions that maximize its chance of success at some goal.” It includes sub-systems for reasoning, knowledge representation, planning, and learning.
Machine learning is one of the approaches to AI and focuses on learning, specifically how to perform a task without explicit rules or steps provided. Ian Goodfellow at OpenAI illustrated the relationship between AI, machine learning, and deep learning, which is another catchy term, well in his book, Deep Learning:
Despite many companies’ claim to AI-powered solutions, I do not see any solution that currently provides AI for marketers. There is no solution that can replace human marketers with its capabilities around reasoning, knowledge representation, planning, and learning.
On the other hand, machine learning has been widely adopted to solve challenges across various marketing tasks, including personalization/recommendation, customer segmentation/clustering, conversion prediction, churn prediction, budget optimization, and more. These achievements are meaningful and will be critical to building a true AI system.
It’s helpful to use the analogy of DeepBlue, IBM’s AI system that mastered chess, to understand AI and machine learning in marketing. We have machine learning-powered solutions to make some moves within a limited scope, for example, when we have only 4 pieces left or when we need to sacrifice between Pawn and Knight, but we haven’t figured out the AI system that can play the game of marketing from beginning to end instead of a human marketer.
The achievements by machine learning-powered solutions are great. We shouldn’t make those achievements controversial or discounted by making incorrect claims of AI.