A Comprehensive Guide to Data-Driven Customer Lifecycle Management

Author(s): Saif Ali Kheraj

Originally published on Towards AI.

From the standpoint of data strategy, the customer lifecycle is a critical topic in the field of targeted and precise marketing. Let’s take a closer look at how it works.

There are three critical steps to remember:

1. Acquisition: Customers must be acquired and converted into customers as the first stage. Here, brand awareness is critical.
2. Maximization: At this stage, customers should be treated with incentives focused on value maximization.
3. Retention: We must prevent customers from churning using the right predictive modeling.

Customer Acquisition

The first important step is customer acquisition because the primary goal of any brand, whether it is e-commerce, telecom, or any subscription-based model, is to acquire new customers. As customer acquisition can be costly, it is critical that we retain and track customers throughout their lifecycle. If we don’t, we may have to spend more and start over to acquire customers. Thus, each stage of the customer lifecycle is crucial. Each process of this lifecycle management relies heavily on data analytics and predictive modeling.

Engagement and Value Maximization

The second stage entails engagement and maximizing value. We must try to extract as much value from our customers as possible. This process is very significant in preventing customers from churning during the initial stages and making them active by engagement. Inactive days or dormancy days, for example, can be divided into different bands for 1 day, 2–6 days, 7–14 days, 15–29, 30–60 days, and so on. This is a critical component of the targeting process. We must then determine the movement through the various stages of the lifecycle using predictive modelling, and if we can quantify this , we can identify the right customers as well as the right objectives to maximize gains.

Inactivity Bands Movement

The above represents a very powerful tool to engage customers from the very early stages. Here, we are tracking the transition from one inactivity band to another inactivity band. We can calculate the transition probability for each movement in the above diagram. For example, the customer can be in an active state (0 days inactivity band). Then, if the customer is inactive the next day or for 7 days, he will move to the next state, which is 1–7. If at any point he starts using service, he moves to 0 state again. We can thus create a machine learning model to predict movement from inactive state to active state. I won’t go over so much detail over this, but this is a very nice tool taught by one of my mentors.

Value Maximization Curve

We also need to find customers who are more likely to progress along the lifecycle curve and generate more revenue. We can create a propensity model to move customers from a lower package to a higher package, or to sell new items.

In short, we can use Propensity Modelling to predict the likelihood of a customer performing a specific action (for example, purchasing, clicking a link, or churning).

In addition to propensity, we must track the incremental value gained from customers after marketing actions based on the propensity model.
Profit from marketing actions can be calculated in a variety of ways. This is a critical step, and it is where “response modeling” comes into play. Response modeling’s primary goal is to predict who will respond based on probability. After determining the likelihood of a response, we can multiply by profit to get a gain and then subtract the cost of those who may not respond.

Gain Equation:

[ Probability of response x Expected Profit ] — [Probability of Not responding x Cost]

The first part represents the expected gain from a responding consumer, while the second part represents the expected loss from sending the campaign to those who won’t respond. The goal is to maximize total profit. What we can do is determine the response probability of each individual customer, as well as the expected profit, and then create a group to which we must send a campaign. However, it is not as simple as you might think.

In the real world, there would be marketing costs and budget criteria to consider. Thus, we will need to build an optimization model on top of this to select the best combination. I won’t go into too much detail here.
We can send the campaign after we have created a set of customers, but the question is how we will evaluate its effectiveness.

Experiment using A/B Testing

To conduct an experiment, we can divide customers into two groups:

Target Group: This is the group for which we will run campaigns.
Control Group: This is the group to which no campaigns will be sent.

If the control group makes the same purchases as the target group, the campaign is a waste of time.

Thus, response modeling is critical in determining whether marketing actions will make a real difference. When used in conjunction with response modeling, propensity modeling can make a significant difference. Let’s make this a little smarter by employing uplift modeling.

Uplift Modelling:

The analysis identifies four types of customers based on their likelihood to respond to a marketing promotion.

1. Lost Causes might not be ideal targets for marketing campaigns.
2. Sure Things: Customers who will likely make a purchase regardless of whether they receive a promotion.
3. Do-not-disturbs: Customers who might be negatively impacted by a promotion.
4. Persuadables: The ideal target group. These customers are more likely to make a purchase if they receive a promotion.

Thus, we must target persuadables as our primary target group.

We can thus use a combination of both the Propensity Model and Uplift modeling for precise and targeted marketing campaigns.

Other Important Targeting Models in Place

Lifetime Value Models: This calculates the long-term value of customers considering the frequency of purchase, spending power, and other variables.

Tiered Modelling: We can categorize customers based on their spending and classify them as gold, silver, or bronze. For example, we can make the top 20% gold, the next 20% silver, and the last 20% bronze. The rest are not eligible for advancement.

RFM Analysis: This is a very powerful tool that includes three key variables: recency, frequency, and monetary value. We can also use segmentation or clustering with these three variables to generate a nice spider chart to better understand customer behavior.

Conclusion:

While this article only provides a strategy and summary, we can delve deeper into each of these models and combine them to work in an automated manner. For example, for effective targeting, we can combine tiered modeling, RFM analysis, propensity modeling, and response modeling.

References:

[1] Uplift Modelling: https://www.semanticscholar.org/paper/Exploring-uplift-modelling-in-direct-marketing-Mayes-Govender/049de9093d1bb49b1f74c73fbaf9c196832f84a8

[2] Customer Value Management: https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/unlocking-the-value-of-personalization-at-scale-for-operators

[3] https://www.zfort.com/blog/propensity-model

[4] Response Modelling: https://en.wikipedia.org/wiki/Response_modeling_methodology

[5] https://blog.hubspot.com/service/customer-lifecycle-management

[6] https://ambiata.com/blog/2020-07-07-uplift-modeling/

[7] https://github.com/ikatsov/tensor-house/

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI