Page 2 of 2
In Search of Doppelgängers (Your Best Customers)
How would you like to find individuals who behave like your best customers? If you knew who these doppelgängers (aka look-alikes) were, would you treat these individuals differently to achieve your organizational goals?
The application of advanced analytics can do just this, allow you to find individuals in your customer base who act and behave like your best customers. Your goal is to figure out who these individuals are as early as possible in your relationship so you can nurture that relationship to the benefit of your company.
You will follow five key steps:
- Define your objective
- Segment your audience
- Establish the attributes of your target segment
- Develop your model
- Uncover your hidden doppelgängers
Step 1: Define your objective
Your first step is to identify what success means for you. This will help you identify your target customers. If your objective is to increase revenue by 20 percent, you will want to identify customers who are increasing their purchases monthly and contributing to that revenue target.
Step 2: Segment your audience
Next, segment your existing customer base. If your definition of success is increased revenue, you will segment your customers into two groups: those who are contributing to increased revenue and those who aren't. To segment these customers, you might ask:
- How important is the longevity of your relationship with your customer? Is consistency in purchasing behaviors as important as longevity?
- Are you concerned about total sales volume or are you seeking the most profitable customers?
- Are you interested only in the sale or in the total life cycle costs of a customer, including customer service interactions and returns and exchanges?
Once you answer these questions, you will have a definition of what an ideal customer looks like for your specific business. With this definition, you can divide your existing customer base into two categories, ideal and not ideal.
Step 3: Establish the attributes of your target segment
Your next step is to define which attributes separate the customers who contribute to your definition of success from those who don't. This can often be one of the most challenging and time-consuming steps, but it's critical because these attributes become the foundation for your advanced analytics model.
To start, assess the attributes associated with the questions from Step 2 in conjunction with your definition of success. Your goal is to identify which characteristics separate those individuals in your target group from others.
An example could look like this:
- Business Objective: Increase revenue by 20 percent
- Target: Customers who increase their purchase amount by at least 20 percent in their first six months
- Attributes: time between purchases, total amount of purchases, successive months purchasing, place and time of purchase, quantity of customer service calls, amount of returns
Finally, determine if there are other attributes which are not explicitly captured about the customer profile, but can be inferred. This practice of teasing out attributes as part of the data preparation phase is often referred to as feature engineering. A simple case of this in the example above would be the percent increase of sales from one month to the next. That number is not stored explicitly, but can be easily calculated and may have bearing on your model.
Step 4: Develop your model
Once you have a set of attributes, the next step is to develop a model. A model is a mathematical formula that combines the attributes in such a way that that the result represents a prediction. In a simple example, your model would be the slope intercept formula.
y = mx + b
In this case, y is the prediction, x is a single attribute, m is the weight of the attribute and b is a constant. Although most advanced analytics models use more than one attribute and can take many forms, the basic concepts are the same: attributes are joined together with different weights to create a result.
In developing your model, you want to identify which attributes (x) are most highly correlated to your target (y). By doing this, you can eliminate attributes and decrease the complexity of your model, but still produce a great result. Simplicity will reduce the resources expended when the model is created and when it is deployed.
In addition to evaluating correlation, you can further reduce model complexity by removing attributes that are effectively redundant. Identifying the covariance of the sets of attributes will allow you to see if two individual attributes have a repeat impact on the target attribute.
For example, if you have identified the attributes of "amount of increase from last month" and "percent of increase from last month," both have the same direction and similar impact on the target variable. The only real difference is scale. Using both might or might not have a significant impact on the accuracy of the model, but will increase the complexity of the model. Having too many attributes can slow down model generation and make the model unnecessarily complex. This has often been referred to as the curse of dimensionality.
With a definition of success, customers segmented, and a list of attributes describing existing ideal customers, you can now develop your model. With a target classification that is binary in nature (is the customer ideal or not), you have many types of models you can use: logistic regression, decision trees, random forests, artificial neural nets, or support vector machines.
Many tools on the market allow you to feed your data in with a point-and-click interface and will perform the heavy lifting of the mathematics behind generating the model. If you don't have an analytics platform, languages such as Python and R have prebuilt libraries that will allow you to develop models with your data. They require more coding than a point-and-click tool, but also provide a high level of configuration for your model generation.
In the end, regardless of whether you use a platform or a coding language, your model needs to be a packaged process that can be run against all your customers to identify which ones have the highest potential of helping you achieve success.
Step 5: Uncover your hidden doppelgängers
Finally, you will employ this model and search for your potential ideal customers. To accomplish this, you will identify the same data attributes for all your customers that you used to develop your model minus the target classification. You will run this data set through your model. The output will be a prediction of whether the individuals in this data set exemplify the same characteristics as your target individuals.
With this prediction in hand, you can now arrange the necessary resources to best nurture your relationship with these customers. Early nurturing will ensure that these customers have the highest probability of helping you achieve your organizational goals.
The real value of the investment in advanced analytics is when the combination of more unusual attributes are the most impactful factors in identifying a potentially high-value customer. A model that predicts a long-term customer based only on the those who place a second purchase is straightforward and not very advanced. On the other hand, it is when your model combines the hour of day of the second purchase with the expiration month of the payment method and the number of distinct item categories purchased in the second order that you start to uncover something that was previously unknown without advanced analytics.
When you can identify customers who exemplify these characteristics within a few months or even days and can nurture them into your best future customers, this is where greatest return on investment of advanced analytics is found.
Time to Get Started
Businesses often wonder how they can find their next generation of great customers. As you apply this five-step process, you can quickly and efficiently enhance your company's bottom line. Good luck as you embark on your doppelgänger hunt!
Troy Hiltbrand is the chief digital officer at Kyäni where he is responsible for digital strategy and transformation. You can reach the author at firstname.lastname@example.org.