Ah, Martech. I love the way you evolve by borrowing from so many other disciplines to help us better understand ourselves. AI and data are all the sizzle; however, the realization still eludes most marketing organizations due to a historic lack of data and operators to process. With that changing, and as many of you begin to leverage data in your marketing efforts, I wanted to outline some of the most common algorithms and analytic tactics to gain insights for better decision making.
There have been a tremendous number of articles talking about data science for marketing, but let’s start connecting the dots for those of you ready to go!
Goal: Customer Segmentation
Clustering gives marketers a new ability to discover customer segments. Most advanced analysis requires a very clear understanding of the relationship of data. For most novice data-enabled marketers, this is understandably a challenge without an infusion of renewed insights. Clustering techniques make that journey easier to arrive at, reaching the right segments to plan and deploy relevant offerings.
An advanced use of clustering is incorporating the time. Clustering analysis offers a snapshot in time. So, depending on your audience and variables used, the segment definition and allocation can rapidly age. While this can’t be helped for the entire spectrum of your customer data, there are interesting insights on journeys within your customer segments worth paying attention to. First, tag an individual or cohort of people you wish to follow. Next, run clustering over different ranges of time. As you observe the movement between or within clusters, consider understanding how, why and the context of the movement (or lack thereof).
This type of journey analysis helped one of our clients, a division of a global conglomerate, determine how their most prime customers operated in cyclical business conditions. It led to a proactive deployment of support account executive reps within a month, to retain over USD $23Mn that were originally marked to be lost.
Take care of emotional responses of being clustered into a group that someone may not desire or mistakenly belong. Let clustering be the starting point to discover and refine, not automate.
Goal: Customer Retention
Model: Logistic Regression
Considering the rise of customer success, it’s invariable that algorithms or models for retention would have a great deal of investment. In this case, logistic regression takes independent variables and determines the likelihood of a prospect/customer to churn. Customer churn is a challenge because there are more unpredictable variables including customer reps engagement & community response factors.
This type of model has proven to be most effective on consumer subscription businesses. One such example is a telecom client we were working with to leverage data to refine their customer retention models. In this case, we tagged voluntary churn and non-voluntary churn before applying models to aid executives to apply the best marketing or customer retention campaign responses. This improved both the reduction in churn (cost) and increase in campaign effectiveness (additional revenue generation) resulting in USD $118M in ROI and a rise of customer satisfaction by 14%.
Algorithms do not have enough sensitivity to context. The ‘art’ in the science is of this model is how self-aware of how your business is perceived by your customers. Then you can make better determinations of what variables are at play here.
Goal: Demand Forecasting
Model: Predictive Analytics
Predictive Analytics was all the rage a few years ago. Now considered table stakes, behind much of the increased awareness with ML. The significant use of martech, sales and customer success and CX, basically helps analysts determine a correlation to apply toward future scenarios.
One example is demand forecasting which is an area of predictive analytics that looks to provide future estimates of products or services to be consumed or used. It goes beyond educated guesses and looks at historical sales data or current data from test markets. Analytic techniques such as time series analysis can be very useful here.
It is absolutely critical that you invest in highly cleansed and accurate data management. The alternative will steer you wildly on the wrong path.
Goal: Look-a-Like Customer, Prospect
Model: Machine Learning
Sales has always needed a way to understand the persona to prioritize, while marketing is looking to new buckets of segmentation that deliver additional revenue opportunities. ML has become the subset of AI which has had the greatest adoption so far. It’s simplicity and relative accuracy based on the good and correct use of historical data.
Another incredible use has been to help high transaction sales teams in a startup to improve their close rates by levering algorithms to better communicate and narrate better position of the company in the call. This growth, in turn, helped them raise over $40M from venture funding.
Feedback is still necessary and on occasion, critical. In one case, a client in the financial space was unable to understand why they were suddenly losing just over $30M in investment decisions per quarter. New to its use, the same ML was used from the previous banner year. We discovered that the brokers were too complacent with their clients on the algorithm. Coupled with little feedback provided to the ML, and training for the teams were causing not only a loss but of a magnitude higher than 10x customer lifetime value.
Machine learning alone will never replace but augment your best people.
Conclusion & Caution
Use these models and algorithms carefully. The best models have people who have the best self-awareness of the business. The best algorithms are ones which are regularly nurtured. And while true that general AI progress has yet to be made, we can leverage some proven algorithms to make great strides in developing better relationships with our customers. What other marketing algorithms or models (B2B or B2C) do you feel have the greatest impact?
Originally Posted at MTA: https://www.martechadvisor.com/articles/machine-learning-amp-ai/marketings-fav-four-data-science-applications/