Modeling Human Behavior Through AI-Powered Marketing

Early successes in bringing AI to marketing may reveal patterns in human thought and behavior that can translate to other applications. Tech companies that have dominated and transformed the marketing industry have been leaders in AI development; consider popular human-like assistants, including Siri from Apple, Alexa from Amazon, and Google Home. The marketing industry has proven itself a worthy testing ground for innovation in AI and, particularly, in modeling human behavior and understanding AI-human interaction.

Decision-Making Algorithms

As AI has started trickling into aspects of daily life, there has been an increased need for machines to “think” the way humans do. With increased awareness of the ethical and moral implications of AI, experts have focused on decision-making algorithms of self-driving cars, predictive policing, and autonomous weapons, to name a few. These are particularly difficult models to train, as historical data may be strongly biased. To avoid potential catastrophe, these algorithms must sufficiently understand human cognition to make the same choice a morally conscious human would tend to make under similar circumstances.

Both humans and machines rely on patterns for cognition, but critics of AI are worried that machines are unable to truly understand human thought. Current limitations of AI hinder a machine’s ability to use abductive reasoning or creative thinking. However, machine learning models have demonstrated immense progress in recreating human behavior patterns, with much of the leading research stemming from digital marketing.

Machine Learning’s Initial Success in Marketing

AI has inspired a complete transformation of the marketing industry and continues to drive new developments. Over 60% of US companies employ AI in their current marketing schemes; most commonly, machine learning is used to predict customer demand, provide product recommendations, optimize advertising campaigns, and automate customer service.

Supervised Machine Learning

As the most common machine learning method supervised learning has a wide range of applications, from training chatbots to sorting images. These models are extremely effective at filtering through cross-cutting variables to determine an outcome and are successfully able to mimic patterns in human behavior. However, this method relies heavily on high-quality datasets with established correct answers to use as training material.

Unsupervised Machine Learning

While supervised learning expects a particular outcome from data, unsupervised learning focuses on an examination of the data itself. This eliminates the need for pre-labeled training data. Algorithms like clustering, which search for underlying correlations in datasets, can be used for segmenting customers and markets, classification, and detecting outliers. Unsupervised machine learning is analogous to bringing in a fresh set of eyes or an independent consultant to discover entirely new or unexpected connections.

Reinforcement Machine Learning

When historical datasets are unavailable, reinforcement learning can be used to evaluate incoming data on the fly. This algorithm actively learns and adjusts in nearly real-time. Much like human trial and error, this method relies on quick actions with immediate feedback. Reinforcement machine learning is ideal for tailoring recommendation systems with new releases or adjusting new ad campaigns as they propagate through social media.

Leading Innovations in AI-Based Business

McKinsey estimates that U.S. customers will manage 85% of their brand relationships without human interaction by December 2020. While many companies are investing heavily in machine learning-based marketing, only 6% of companies are leveraging more advanced AI, such as personalized campaigns, collaborative filtering, and predictive models.

Management and strategy have traditionally been viewed as a role that goes beyond AI’s reach — managers are expected to balance both efficiency and fairness, which has proven challenging to quantify. However, given machine learning’s undeniable success in other human-centric aspects of business, like marketing, researchers are beginning to explore how to bring AI into management and marketing strategy. In the short term, a promising approach is using fuzzy logic to develop a shortlist of proposed solutions and outcomes, where a human operator is responsible for making the final decision. As computational capacity increases and data sets are developed, eventually these models could become fully autonomous. Pilot projects for autonomous decision making have begun in recommender systems, digital advertising, and dynamic pricing algorithms. The handful of companies that have made progress in AI-based strategic marketing are keeping their work highly protected, demonstrating clear value even for early adopters.

Future Applications

The $40 billion of expected revenue that AI-powered marketing will bring by 2025 promises a hefty stream of funding for continued AI research across industries. New algorithms and methodologies for predictive analytics, human-AI interaction, autonomy, and cognition are emerging at an ever-increasing rate, both within digital marketing and beyond. The insight on human behavior gained through marketing-based research will guide tech leaders and innovators toward autonomous systems that can make choices the way humans do. Or, perhaps these systems will go beyond mirroring human decision-making and will help businesses make better decisions than humans could.

Caitlyn Caggia is a content writer for PDF Electric & Supply. She is an experienced systems integrator and solutions architect, and she holds an MS in Electrical and Computer Engineering from Georgia Tech.

Topics:

artificial intelligence (ai),
machine learning,
digital marketing,
human behavior,
decision management,
human-computer interaction,
ai ethics,
ai bias,
autonomy

Read More

Caitlyn Caggia