AI: Putting the Habit Back Into Customer Marketing
The renowned behavioral psychologist B.F. Skinner once said, “The major difference between rats and humans is that rats learn from experience.” His words are certainly comical and do contain an element of truth, but human beings are, first and foremost, creatures of habit. Their behaviors come from experience, and these behaviors are highly quantifiable. This is even true for their shopping behaviors, as the proliferation of retail analytics and apps has shown.
If marketers can understand the habits of their customers, they can predict potential customer purchases, which can help businesses in a multitude of ways. Smart businesses can utilize these behavior patterns in a predictive way and optimize their manufacturing, supply chain logistics, labor management, marketing, and sales processes. With its powerful predictive capability, AI can help with this optimization.
As Vinicius Andrade Brei explains in his article Machine learning in marketing: overview, learning strategies, applications, and future developments, “ML offers several advantages and new perspectives for knowledge generation in marketing. It can be applied for solving regression and classification problems, clustering, visualization, dimensionality reduction, creating association rules, and developing reinforcement-learning agents, among other utilizations.” For marketers, in particular, ML models can help with the following:
- Choice modeling
- Understanding consumer behavior
- Recommender systems
- Relationship marketing
Built on sound economic principles, choice modeling assumes people make decisions of choice based on an evaluation of the utility available from each alternative available, and a person will choose the one he or she believes has the highest utility.
Choice modeling can help companies understand what aspects of a product customers value. It can also improve a company’s product offerings, as well as show if there is even a demand for a particular product and/or service. For company marketers, it could reveal how to improve their marketing messages.
Choice modeling is useful for:
- Analyzing price sensitivity
- Bundling product and service features
- Optimizing brand strategy
- Improving product-line planning
- Maximizing media advertising effectiveness
- Improving promotional offers
- Optimizing advertising messages
In their paper A Nonparametric Joint Assortment and Price Choice Model, Jagabathula and Rusmevichientong describe a choice modeling experiment that would have made Skinner proud — they added an element of rationality into their choice modeling application. This is because consumers’ preferences aren’t always rational. They developed a method to quantify the limit of rationality (LoR) in choice modeling applications. Their methods allow businesses to compute LoR efficiently on grocery sales data to identify product categories that require going beyond rational choice models to attain acceptable performance.
Wang, Aribarg, and Atchade also explored how an individual’s product choices are shaped by the product choices of those they’re connected with, specifically for fashion and technology items. The results, although unsurprising, were useful: experts exerted an asymmetrically-greater influence on technology-related purchases, while popular individuals exerted a greater influence on fashion-related products. Their model also gave influence predictions for a technology-related product, and they found choices made by early decision-makers were “more influential than choices made later for the technology-related product.” The lesson here: with a new technology product, get the tech influencers on board and pound the marketing as soon as the item hits the market.
Understanding Consumer Behavior
Utilizing ML to affect consumer behavior is common in areas like search, knowledge acquisition, idea generation, preference assessment, shopping patterns, consumer responses, and post-consumption evaluation. Capturing customer behavior is extremely important, as companies can use this to decide what products to sell, what services to offer, how to layout a store, how to market an item, and how to sell to a customer.
The explosion of offerings in today’s market makes understanding customer preference imperative for companies that have huge product catalogs. In their paper Capturing Heterogeneity Among Consumers with Multi-taste Preferences, Liu and Dzyabura point out that a “single consumer may have multiple tastes within a given product category.” Multi-taste preferences are likely present in categories that are characterized by large product attribute spaces, and in many diverse products, such as music, videos, restaurants, and books. Capturing similarities among multi-taste consumers requires new methods, say Liu and Dzyabura, because two consumers may share some tastes with one another, but not with others. This is a different type of heterogeneity than captured by existing models, such as mixed logit or latent class models, which estimate only one taste per individual.
In their paper, Liu and Dzyabura propose a model that allows for heterogeneity among consumers with multiple tastes, and an estimation procedure that scales to potentially very high dimensional attribute spaces. In a numerical study, Liu and Dzyabura simulated consumers with multiple preferences, and demonstrated that “the model uncovers rich patterns of underlying heterogeneity, such as what the different tastes are, how many consumers have each taste, and which tastes tend to be more or less likely to occur in the same individual.”
Recommendation systems based on ML models are nothing new in marketing. They have been used to help with product recommendations, to understand the power of online social networks, and to analyze customer purchasing preferences.
Moon and Russel developed a product recommendation model created by looking at customer preference similarities based on prior purchase behavior. As Brei explains in his paper Machine Learning in Marketing: Overview, Learning Strategies, Applications, and Future Developments, Moon and Russel’s model “was based on joint space mapping (placing customers and products on the same psychological map), and spatial choice modeling (allowing observed choices to be correlated across customers). It achieved superior forecast performance of purchase behavior compared to benchmark models.”
Van Roy and Xiang also showed that nearest neighbor algorithms, which are widely used commercially, are highly susceptible to manipulation. The authors proposed new collaborative filtering ML algorithms that were much more robust.
Brei also points out that Chung and Rao’s general consumer preference model for experience products, like movies, overcomes the limitations of standard consumer choice models. “By using Bayesian estimation methods and Markov chain Monte Carlo simulation inference, and testing the model in online consumer ratings and offline consumer viewership for movies, their approach outperformed several alternative collaborative filtering and attribute-based preference models,” contends Brei.
ML is commonly used in recommendation systems, and even more frequently used in customer relationship marketing, including for customer acquisition, customer engagement, customer experience, patron valuation, customer relationship management, word-of-mouth marketing, online reputation management, and customer churn.
In their paper Customer Acquisition via Display Advertising Using Multi-Armed Bandit Experiments, Schwartz, Bradlow, and Fader explain that many online advertisers are using A/B testing in a testing and learning market environment, but they are missing out on a more profitable option — earning while learning. Schwartz, Bradlow, and Fader state, with online advertisers, organizations “typically handle this earning vs. learning (or explore-exploit) trade-off in two phases — test then rollout. They equally allocate impressions to each ad version (explore phase), and after stopping the test, they shift all future impressions to the best performing ad (exploit phase).” However, as Brei contends, the optimal test phase length can’t be known in advance. “Instead of a discrete switch from exploration (learning) to exploitation (earning), firms should [simultaneously] mix the two and change the mix with a smooth transition from one to the other (earning while learning),” recommends Brei.
Customer retention is a core dimension of customer relationship management (CRM), and it is gaining strong attention from companies, as there is a strong association between customer retention and company profitability. The longer a customer remains with a company, the more loyal and valuable that customer becomes. In their article In pursuit of enhanced customer retention management, Ascarza et al. present an integrated framework for managing retention that utilizes new methodologies like machine learning. Ascarza et al. argue that too much time is being devoted to predicting customer churn when it’s more important to distinguish between “which customers are at risk and which should be targeted — as they aren’t necessarily the same customers.”
Ascarza et al. believe the central idea behind customer retention should be continuity, i.e., the customer continuing to interact with the firm. Secondary to this is the idea that “customer retention is a form of customer behavior — a behavior that firms intend to manage.” Accordingly, Ascarza et al. propose that “Customer retention is the customer continuing to transact with the firm.” Dimensionality reduction and variable selection techniques “may be useful for retention research and practice,” say Ascarza et al. Churn modelers might also want to consider the Cox proportional hazard models as it has been proven particularly effective, and it is available in the R package ‘glmnet’, contend Ascarza et al.
The future of marketing will be automation, and not the type of automation one normally thinks of in advertising. As Niraj Dawar contends in his article How Marketing Changes When Shopping Is Automated, retailers will soon know their customers’ buying habits so intimately that they will supply the 200 or so wanted products on a regularly-scheduled basis according to a retailer’s replenishment algorithm. “Products will flow to the household like a utility,” claims Dawar. This means, if you’re not top of the consumer’s mind, you’re not making a sale. However, if you are, the sales will be flowing automatically, at least until the consumer makes a change in the ordering process.
“Challenging incumbents, increasing rates of consumption, and influencing algorithm designers and owners” should be the aim for marketers, says Dawar. Influencing algorithms, in particular, should take precedence, with the goal of becoming a native or default brand to the consumer. AI, ML, and deep learning can help with this.
We probably aren’t too far away from a time when many purchasing decisions will be managed by a bot connected to the IoT. Consumers might find that consuming is all they have to do. This is all the more reason to ensure their behaviors are understood and their habits locked in. The goal for personalization marketing has always been to surgically strike offers to the interested one rather than blast a shotgun of offerings to the indifferent many. There’s a reason why today’s email marketers often have a 99% failure rate. People are overwhelmed by the 5,000 ads they see each and every day and getting noticed amidst the clutter is getting more and more difficult. However, old habits die hard, and for the marketers who understand and utilize their customers’ behaviors in their customer relationship management and advertising, that could be music to their ears.