Personalized marketing for retail consumers and account-based marketing for B2B customers now have proven value. Online interactions with customers generate large volumes of data for granular learning about consumer behavior for customization of product recommendations, messages, and content.

The missing piece is a scalable and just-in-time way to gauge customer preferences and make product recommendations while visitors engage with websites. Deep reinforcement learning algorithms have been trained at the threshold level where they begin to achieve conversion rates to match the costs of data analysis.

Reinforcement Learning

The touchstone of reinforcement learning (RL) is that it experiments with multiple pathways to achieve the objective of acquiring customers or any other goal. It iterates for a diversity of prices, product recommendations, advertisements, and content to zero down on the most effective method that will motivate any individual to become a customer. (Read also: Reinforcement Learning Vs. Deep Reinforcement Learning: What’s the Difference?)

Deep learning (DL), by contrast, finds patterns in data with specified algorithms but not with predetermined end goals in mind. For purposes of marketing, RL maps out the spectrum of journeys of customers that conclude with their acquisition in a diversity of market segments. (Read also: What is the difference between deep learning and machine learning?)

Learning is for keeps and informs the interactions with customers for up-selling, cross-selling, and for earning higher revenues from each customer throughout the lifecycle of their relationship with vendors.

Personalized Marketing

Undoubtedly, personalized marketing considerably improves efficiencies in customer acquisition. The costs of acquisition declines by a whopping 50%, revenues rise by 5 to 15 %, and the effectiveness of marketing budgets increases by 10 to 30 %. Data from voice, visuals, and emotions create vivid views of individuals to inform the personalization that resonates with customers.

If a customer, for example, expresses frustration with musty towels, an emotion detection artificial intelligence (AI) engine is able to read it and can instantly recommend an alternative towel, launched recently in the market, that dries quickly, to clinch a sale.

Account-based marketing (ABM) in B2B sales has exponentially increased the average contract value as specific problems of customers are addressed, cross-selling is more widespread, and by cultivating deeper relationships. The average contract value is reported to have risen by 171 percent. The ROI realized from ABM was higher than other marketing initiatives by 71% of a sample of executives surveyed.

In terms of relationships, 75% reported significant improvement in their relationships with prospects of more than 10%.

Scaling Personalized Marketing

Nikolas Kairinos, co-founder of Fountech Ventures, an incubation and venture financing company specializing in AI, and founder of Prospex, a lead generation company, explained how reinforcement learning achieves scalable levels of conversion rates:

“Reinforcement learning evolves as it learns from the interactions of customers throughout their purchase journey. It monitors whether an appointment was made, the sharing of website content on social media sites, or did the lead enquire about the attributes of the product. The data of the activity in one domain, such as book sales, can be used to predict behavior in other related domains such as movie purchases for the purpose of making recommendations.”

Methods

Relationship graphs are the bedrock of the organization of customer data for analysis with RL and other forms of AI. These graphs, for example, map social networks of people.

The purchase behavior of customers is impacted by word-of-mouth in their social network. These social groups are influenced by messages communicated by mass media and conversations on social media. Their neighborhoods are telling of their cultural influences, and their socialization patterns such as interaction in professional groups reveal their interests. Several layers of data are overlaid on the relationship graph to understand causes and effects.

Inevitably, the large volumes of data, initially, have a great deal of unexplained noise. “The larger the volumes of data you have, the more likely is the ability to estimate baseline averages for categories and sub-categories of customers with increasing levels of precision,” Kairinos surmised. “Noise levels diminish further as more is learned about individual behavior and the cause-and-effect of the purchase activity of each person,” he added.

Kairinos also shared his thoughts about gaps in data and how it is still possible to derive results that can inform decision-making. Consumers, for example, choose between product offerings from competitors while vendors can only analyze data from their own websites and other publicly available data.

“In most cases, we do not get competitor information which is protected by ethical norms for privacy. AI tools like Prospex aggregate data for categories such as the pattern of behavior in certain regions, industries, and age which includes all customers. The propensities for individual customers can be inferred from their affiliations in groups,” said Kairinos.

The Human Touch

Despite the widespread use of sophisticated algorithms, telemarketing firms continue to exist to validate and augment marketing datasets, that include information on target customers, before their clients use them for business decisions.

Shahida Afzal, founder of Trifle Solutions, shared her thoughts on learning how the services of her company continue to be of value to consumers of market information despite the automation of data collection.

“Clients are looking for subsets of information that answers questions that their executives want to ask. The larger commercial databases are often not able to provide specific, often granular, information for their exact needs whose accuracy is guaranteed,” said Afzal.

”It takes human communication to understand the precise data needs of clients and to craft a strategy for querying respondents to extract the desired information,” Afzal explained.

“Datasets are augmented with contextual information, gained from asking questions about the backdrop of where, when, what, and how something happened to understand the data in perspective,” Afzal added.

People are often understandably wary of sharing internal company information, especially if it is confidential, or just plain guarded.

“We build rapport with our respondents, remain politely persistent until they let their guard down and feel comfortable talking about their business. For confidential information, we ask questions about interrelated data that helps us to infer the values of the desired data,” said Afzal.

Above all, clients consider the accuracy of the data as an overriding criterion for the selection of their data source. “We make 50 phone calls, in their presence, to demonstrate that commercial datasets are replete with errors. Eventually, they end up choosing our data eight times out of ten,” said Afzal.

Final Thoughts

DL has often been criticized as a Blackbox barely able to explain the cause-and-effect in the mass of data that it analyzes. By charting the pathways of customers on their way to purchase and the influences on their choices, RL paves the way to predict their future actions considerably improving the outcomes.

Yet, human involvement is required to fill the gaps left by machines.