Uncover Actionable Insights Using XAI

Uncover Actionable Insights with XAI

Advertising is more data driven today than ever before. Digital advertising has evolved over the last ten years to define audiences at much greater levels of granularity. Thousands of data segments litter the digital ecosystem and robust data sets are available from every major media platform and ad tech provider spanning TV, online video, gaming, and mobile. But most campaigns still run on “data” selects and most customer interactions are subjected to the same basic information set.

Data, by itself, is not insight. Insight is an interesting word to think about – what is it really? Going within the data, and the underlying trends, patterns correlations and signals, is where the truth lies about your customers and what motivates their behavior; positive or negative. Insight is what we really should be seeking in managing campaigns and customer interactions. Yet when it comes to marketing, we ignore so much of what is inside of our own data – and maybe it’s because we can’t surface the information easily. “I’m drowning in data but have no insights,” is a common theme we hear across major brands.

We have also settled, as an advertising ecosystem, into a set of standard practices that aren’t based on the customer but based on performance metrics like CTR and attribution reporting that are focused on meeting “outside in” objectives. Outside in is focused on how many ads got delivered and who got credit for a sale vs. “inside out” which is focused on the customer’s expectations, needs, cumulative experiences, preferences, and the contextual situation of their interaction with you. There is a huge difference between the two.  And this is where AI for marketing becomes very important – to enable an inside out approach and make it a reality.

As a result, custom modeled audiences, using an advertisers’ first, second and thirty party data, are now the “new normal” for creating ad campaigns that reach a desired audience. Enter Dynamic Predictive Audiences based on explainable AI machine learning technology that reveals predictive audience data, in weighted order of predictive value, and clusters each individual audience prediction that shares common characteristics into dynamically generated segments for campaign execution.

In their book, Marketing to the Entitled Consumer, the authors, Nick Worth and Dave Frankland, talk about consumer expectations and the overwhelming shortcomings of many brands in meeting these expectations.

Too many emails; too many ads; too little relevance. The net effect of this marketing assault is to attenuate the value of marketing messages. By itself, that reduction in effectiveness would be challenging enough for marketers. But even as marketing clutter increases, consumers are paying less attention to the media they’re consuming. Why? Because they are multi-tasking.

In fact, the authors go on to liken some marketing to being like a “loud mouth boor”, who rudely overwhelms based on what they want vs. what the customer wants. An obvious example is continuing to get hundreds of ad impressions for an item you already purchased online months ago. The anecdote to this is knowledge, or insights, that the authors point out through many examples.

Known data facts about your customers are not insight. Data is fuel to generate insights but not insight in and of itself. Insight is based on analyzing what the data facts mean to understand current and future customer needs and expectations; in the moment, in the near term and in the long term – pretty much all of the time in an always on world fueled by digital devices.

Obviously, this is incredibly difficult for traditional analytics to keep up with. Layer on top of that anti-customer business challenges including fragmented data, siloed channels, outside in performance metrics, technology advances like 5G, and standard practices (and I haven’t even mentioned privacy yet) – and companies are way behind where they need and could be. Bottom line: the business of surfacing customer insights needs to be more automated.

Actionable insights are out there, but the ultimate objective of customer relevancy, in the moment, across any channel is far from reality on a day to day basis. Machine learning is great at speeding things up and being more accurate. Speed and efficiency are the areas where ML is impacting low hanging fruit – like programmatic advertising. But when it comes to insights – this is hard to come by in the world of open source, black box ML.

Forrester comments in their research report titled, “Turn Data into Insights with Customer Analytics,” February 1, 2019;

Missing the moment. CI pros have made significant advances over the past few years. In 2018, 89% of them used predictive analytics to inform decision making. Despite this, few CI pros can deliver contextually relevant experiences at the moment of customer interaction — only 26% are using real-time analytics. In a world where relevance is table stakes and insights are highly perishable, CI pros must get up to speed and start designing the next best experience for customers and prospects.

Not only are insights hard to come by now but they are going to be harder to come by in the future. Identifiable data about customers has become a bad word, putting constraints on available data. Think about it. The data driven, yet blunt, non-customer centric programs and interactions prevalent today are going to be further constrained in a world that requires a high degree of relevance.

For those brands that are competitive, agile, and customer centric, however, there is still a lot of shared customer information. This increasingly scarce resource not only needs to be carefully managed for compliance and trust, but also mined as a highly valuable resource continuously, to be used for relevancy and good service. And this is where machine learning that enables insights and action, founded on the principles of explainability, comes to bear.

Seeing, and therefore knowing, the key drivers of customer behavior is becoming a pretty big deal. There are several ways Explainable AI, or XAI, enables this to happen. It gives your data a voice. Like a refinery, extracting and transforming raw oil into fuel, explainable AI can refine data into insights that fuel action. It is actionable intelligence and letting the machine surface insights for each and every customer, every day and in real-time if necessary to drive customer interactions or for aggregating insights into manageable segments; that is the strength of this new technology.

In the hands of analysts, insight managers, customer experience managers and campaign managers, explainability AI does the heavy lifting so you can spend time crafting what action to take with what message and treatment – now, in the moment. And, by the way, your customers aren’t static, neither should your analytics, segments and marketing be either!

Here are a few examples of how XAI can help you;

Explainable AI reveals the Why behind individual customer interactions. Running at tremendous speed, XAI can inform decisioning systems with not just the underlying predictive drivers of consumer behavior for that interaction, but the importance of each in rank order.

Explainable predictions offer a major step up function to being “relevant” in the moment of an interaction. XAI  can be used for all kind of purposes including reviewing potential fraud so you don’t inadvertently cancel a real customer’s order when a machine flags it as a potential risk, due to the presence of a new device for example.

Dynamic Predictive Segmentation based on XAI enables the clustering of predictions with similar weighted factors. This is the new segmentation standard for generating the most relevant treatments, including creative, messages and offers, when sending out ad campaigns.

In Forrester’s report, sponsored by simMachines, “Capture the Customer Moment with Dynamic Predictive Segmentation“, the value of this XAI enabled capability is broadly explored with marketers across multiple industries. “Dynamic predictive segmentation is just as useful for agencies and programmatic ad platforms as it is for the brands they serve. Today’s most sophisticated agencies and demand-side platforms are using classic machine learning techniques to identify prospects with a high likelihood to purchase. While this technique identifies the right customer to target, it does nothing to reveal the best message for that customer. For example, one customer may be particularly enamored by a luxury car’s exterior while another may be more interested in the engine under the hood. Dynamic predictive segmentation reveals these customers’ differing preferences, so agencies can exploit them by delivering the right content to each.”

An example of this includes the analysis of recent buyers of pet supplements that showed, through predictive weighting, that family characteristics were the most dominant factor vs. the type of pet owned. This led to a totally different segmentation architecture based on family dynamics vs. pet ownership.

You can even fill in the blanks for certain customers missing information by evaluating the behavior of other customers that are similar – based on state of the art distance functions that operate across hundreds if not thousands of columns of data. XAI for marketing helps assign consumers to segments with common predictive behavioral characteristics.

Evaluating the real drivers of customer conversion behavior to a particular ad treatment is now possible with XAI. The ability to see the underlying most important factors associated with positive vs. negative ad effect can be achieved to help advertisers stop sending ads that have a negative effect, and send ads that have a positive effect to those who look like these positive responding consumers.

Further refinement of messaging and offers based on analyzing insights revealed by the machine speaks to how XAI benefits marketing. This is low hanging fruit for marketers wanting to immediately have an impact on being more relevant, or less upsetting, to consumers with their ads.

Brands can leverage XAI technology to surface insights systematically. Moving from quarterly segmentation to daily for example. Moving from broad based customer segmentation to micro-segments. Stopping negative ad effect treatments right away and focusing on the positive ones to the right audiences. Upgrading consumer experience at the point of interaction based on real-time explainable predictions.  All of these basic improvements are now made easier through advanced machine learning.

It’s easy to test and learn your way across various applications – one use case at a time. The most innovative, customer-centric businesses will start this process and evolve their analytics to automate and apply insights systematically across the business using the latest in explainable AI technology now – before it’s too late to catch up.

AI Enabled Customer Segmentation Will Transform Marketing