In the customer experience realm, data, insight and action are the pillars for what it takes to deliver personalized interactions at scale across an omnichannel customer journey. The “insight” leg, however, is often misunderstood. One reason for the misunderstanding is that companies believe that once they’ve collected and perfected all the customer data they need, they can activate the data on behalf of the customer. In a sense, they’re not wrong in that insight may not technically be a requirement. But it does make all the difference in terms of delivering the depth of relevance that customers expect. Consider, for instance, the latest Harris Poll survey commissioned by Redpoint that explored the customer experience gap. More than three-quarters (82 percent) of customers surveyed said they are loyal to brands that demonstrate a thorough understanding of them as a unique customer. Further, 39 percent of customers said they will not do business with any brand that fails to offer a personalized experience.
Think of a simple real-world consumer experience of walking into a showroom to buy a car. A salesperson, without knowing anything about the customer beyond maybe a name and address, may steer the customer toward one vehicle over another based solely on the customer’s approximate age, who they’re with, how they’re dressed or whether they’re wearing a wedding ring. This is, in a word, insight. The salesperson is augmenting some basic data with knowledge driven by experience to inform a next-best action: here is the perfect vehicle for you.
Deliver Consistent Relevance
In retail, healthcare, travel and hospitality or any industry with a customer-facing dynamic, gathering in-person insights, while valuable, is clearly not scalable for a host of reasons. Beyond volume, customers increasingly engage with brands across digital and physical touchpoints, often with no obviously discernible pattern. The research and evaluation that goes into buying a car, for example, likely consists of visits to a dozen or so websites on multiple devices.
Within the construct of millions of customers moving freely through unpredictable customer journeys, skipping out on insight – going right from data collection to activation/orchestration – misses a key opportunity to layer data with context that is meaningful to an individual customer. In other words, using only demographic or diagnostic data – a customer’s name, address, ZIP, email address, etc. – will generally not reveal what makes a customer unique or provide any details about a customer’s intent, preferences or behaviors, all of which get at the heart of what customer experience (CX) really means. Because while on the one hand, CX refers to the sum of all the interactions with a brand over the life of the relationship the company, on the other it also entails the feelings, emotions and perceptions a customer has about those interactions. And to strike the right chord with a customer to generate a lasting, positive perception requires consistent relevance.
Dynamic Audience Segmentation
Consistent relevance is a byproduct of advanced analytics. The aim of machine learning, which is a subset of artificial intelligence (AI) is to augment human intelligence by using algorithms to enhance CX; humans determine a business objective and let predictive, clustering or next-best action models, such as those that recommend products, to find patterns in the data.
Machine learning analyzes data at scale to provide the needed context that ultimately produces a next-best action. In the Redpoint rg1 platform, machine learning is a key component of the actionable insights layer that bridges the gap between collecting/perfecting data and orchestrating a consistently relevant omnichannel experience.
For a closer look at what we mean by actionable insights, we’ll start with segmentation. One important feature of rg1 is that segments are rules-based and dictated by the data. Instead of cutting a list of customers based on demographic data, rules-based segmentation is dynamic. Machine learning algorithms access updated data from an operational data store and train on the new information, with models creating new segments as needed to optimize the business objective. The two main types of models are supervised and unsupervised; the former uses historical data to train and predict future results, while the latter uses data to find underlying patterns and presents them to users. Another guarantee against introducing stale data is that the models themselves are automatically retrained and updated; a model will run against an algorithm only when it’s been called for batch or real-time use.
rg1’s clustered audiences capability allows marketers to provide a large group of customers and let machine learning create a number of segments from it, effectively removing the work marketers have to do to create those segments manually. The clustering algorithm finds correlations in the data – deriving context beyond the ability of humans sorting through data. But because marketers ultimately run campaigns, it makes sense they should know why – or why not – a machine learning model has made a correlation. Another rg1 feature is an audience insights dashboard for users to view the segments that were created, which include a description taken from the decision tree used to generate the clusters showing users the decision tree model for why a set of customers is unique.
Machine Learning + Distributed Data
Audience insights provide another layer of trust for marketers to have confidence that the data will ultimately provide a hyper-relevant customer experience in the context of an individual customer journey. The need to have complete trust in data underscores the importance of establishing clear goals for machine learning models at the outset, and ensuring that a model has the right data – and enough data – to support a model.
Because a model is only as good as the data it’s been given to learn on, it is sometimes the case that without the correct distribution of data that a predictive algorithm will produce incorrect results. The introduction of synthetic data helps ameliorate this problem. A model that is optimized to detect fraud, for example, will be biased against detecting fraud if the vast majority of transactions it analyzes are without fraud. By introducing fraud in the data as part of the training set, a model will eventually learn how to detect fraud. Alternatively, humans would either have to manually pick datasets correctly or assess what the algorithm is doing and bias the algorithm after the fact.
Machine Learning + Next-Best Action
Model outputs are dependent on the type of data being used, as well as how the data needs to be leveraged. A propensity score, for example, might be represented in an audience insights dashboard by a range of “least likely” to “most likely” in terms of a customer’s propensity to purchase, again showing a marketer or business user why the model arrived at a conclusion.
Whatever the resulting propensity, the output highlights the purpose of machine learning: find insights that a human could not based on an objective set by a human, i.e., “find me X.”.
These types of calculations bring us into how a next-best action is determined, recognizing that there are many decisions made by a customer before a purchase – what links did the customer visit, what did the customer click on, what did they view, how long was an online session, did they contact the call center, visit a store, post on social media, etc.?
In essence, figuring out how each facet relates to another is what machine learning algorithms are tasked with. As an example, for machine learning to provide the next-best action at any point in the customer journey, it would need to be aware of everything that happened for a potential sale to occur, for instance, and then depending on the next inbound or outbound interaction will offer the next-best action that will lead to the sale.
Advanced analytics driven by machine learning ultimately provide a customer with the type of digital-first personalized experience that they’ve come to expect, all without marketers showing their hand. Meaning the customer likely is unaware that machine learning is behind the perfectly relevant offer or message, only that it’s been delivered at precisely the right time, on the right channel and in the cadence of a unique customer journey.
Note: A follow-up blog on actionable insights will explore pre-built machine learning models, their role in next-best action calculations, and how machine learning prepares data for activation through intelligent orchestration in rg1.
Related Redpoint Orchard Blogs
Machine Learning Misconceptions, Dispelled
Use Predictive Marketing to Demystify AI and Machine Learning
