Net Lift Modeling For Campaign Management, Return Maximization & Incrementality
When brands select to invest precious budgetary dollars to perform a customer campaign, it’s not always desirable to target an entire marketable universe. Whether budgetary constraints stand in the way, or communicating with irrelevant customer segments wastes money, there are rationale reasons to consider efficiency. This means brands may choose to focus on the top 10/20/30% of customers with the highest likelihood to convert for a campaign effort. This is typically referred to as response modeling.
The main goal of this approach is to develop an analytical solution that selects the customers who are most likely to respond to a marketing offer. This action can be a purchase, upgrade, registration, donation, subscription or offer redemption. Several modeling techniques, typically within the category of supervised learning, can be applied such as logistic regression, decision trees, neural networks, etc. in order to provide us with insight. The key in response modeling is that the algorithm an analyst selects will predict which customers have the highest chance to take a desired action based on historical campaign data. This labeled data contains a set of customers with their individual characteristics and their associated response to the historical campaign offer. However, is this really a brand's goal?
Marketing teams aim to contact the customers most likely to respond. In addition, they desire to contact customers who respond to offers because of the campaign. This leads us to an approach which goes under many names, such as net lift/uplift modeling, differential response analysis, etc.. The net lift modeling approach assumes that there are customers:
- Who will always take an action.
- Who will never take an action regardless of the campaign.
- Who react negatively to a campaign regardless of their original feelings.
To illustrate this hypothesis, let's provide an example. A brand decides to contact a segment of customers to convince them of a subscription upgrade. A subset of these customers have the intention of making an upgrade, and thus, these are the customers a brand could save budgetary dollars on by not contacting. There is another subset of customers who will be annoyed by an outbound marketing intervention, and this could cause an unexpected negative outcome. Lastly, yet another subset of the customer group isn’t currently considering the upgrade, and analytically has a likelihood to convert. This is the group that interests the marketing team. If the marketing intervention can convince them to upgrade, this creates incremental value.
Similar to response modeling, analysts can use multiple modeling techniques to build a predictive net lift solution. The main differences are determined by the type of historic data one uses and the maximization function. Net lift modeling attempts to optimize the difference in response of the customers who have received the campaign intervention with those who didn’t. In this way, a brand still ends up with customers who are most likely to respond, but also with an audience who reacts specifically to the campaign thus avoiding the subset who would convert regardless of the campaign. This saves on the cost of the outbound campaign effort for customers who are already convinced. Moreover, it avoids the concern that a marketing team scares away customers (churn, attrition, etc.).
When one compares traditional response modeling with net lift analysis, a number of value propositions become apparent. The biggest advantage of net lift modeling is that it gives analysts a true answer to the question if marketing is making a positive impact (or not). Important considerations for performing this type of analysis include:
- Is it possible that a subset of customers can react negatively to a brand's marketing interventions? The high cost related to losing an existing customer is a well known problem.
- What is the macro-cost of a campaign? If a brand takes on a response modeling approach, it will also contact customers who would have reacted positively regardless of the offer intervention. In this situation, a brand is pushing budgetary dollars into the campaign without gaining a return on the investment.
Before proceeding, check out this introductory video extracted from a recent on-demand webinar summarizing the net lift & incrementality approach as a viable business use case.
Now, let's proceed to a more detailed example. The objective is to target a product offer to customers on a brand's website or mobile app. Impressions of the targeted intervention would occur on the brand's product-specific web pages or app screens. Whether the customer selects to immediately convert or not, the brand will monitor the customer's behavior related to this intervention for 90 days. The test design results of this campaign would look like this:
- Targeted test group: Received an offer.
- Non-targeted (or holdout) control group: Did not receive an offer.
Let's assume the overall customer 90-day conversion rate for this test resulted in 1.5%.
| Customer Cell | Conversion Rate |
| Test Group | 5.01% |
| Control Group | 5.00% |
So, why did we not see any campaign lift?
| Customer Segment | Outcome |
| Not Interested | Will never purchase the product. No point in marketing to them. |
| Very Interested | Likely to purchase the product on their own. Marketing could even have an adverse effect. The campaign targeted too many of these clients. |
| Influenced Customers | Interested in the product but need to be motivated to buy it. Target more of these clients. |
A solution to the problem observed above would be net lift modeling (as opposed to response modeling).
| Net Score Audiences |
Test |
Control | Net Lift |
| Top 20% | 6.10% | 3.90% | 2.20% |
| Bottom 80% | 4.75% | 5.28% | -0.48% |
To close out this example, let's summarize the differences between net lift vs. response modeling in the context of this use case.
Net Conversion Rate = Test Group Conversion Rate - Control Group Conversion Rate
Propensity Modeling
- Most common approach.
- Targets the clients with the highest probability of making a purchase following a marketing contact.
- Maximizes the test group conversion rate.
Net Lift Modeling
- Targets the customers that can be motivated by marketing.
- Maximizes the incremental conversion rate and profitability.
Net lift models predict which customer segments are likely to make a purchase ONLY if prompted by a marketing undertaking, as well as maximizing return on campaign investment. With that said, let’s pivot to an industry-specific demonstration on how SAS can be leveraged as an end-to-end solution for campaign management, return maximization and incrementality in the context of net lift-driven marketing.
Casino Gaming & Hospitality Demonstration: Optimizing Player Reinvestment with Machine Learning
The use case we walk through here shows how to decide which casino player in our database receives the most profitable offer within a campaign. One industry-specific term used in this demonstration is theoretical spend or “theo,” the amount of money a player is expected to spend on the casino floor. It is the theoretical revenue and profit numbers which drive many casino marketing decisions.
The first step in the net lift modeling process reviews the results of the last 90-day marketing campaign. The two main targets of this analysis are player visits (our conversion event) and theoretical spend (our revenue metric). The aim is to maximize the combination of these two metrics to determine which is the recommended offer to send to each individual player.
The demo video below demonstrates the exploratory visual analysis and assessment of this net lift modeling example for casino gaming.
Now that we have explored and summarized our visual modeling outcomes, let’s next demonstrate the model pipelining process to show how analysts and data scientists have the optionality of deeper control and transparency in authoring machine learning assets. The focus of this demonstration is on the first goal of the net lift process: predicting a player’s likelihood to return within the next year. The second goal of predicting a player’s estimated spend would be accomplished using a similar workflow.
As mentioned in the demonstration, machine learning models must be transparent and fair to be implemented into business processes. To showcase one plot provided for the demographic variables we selected to assess for bias in the model project, let’s review the performance bias chart auto-generated by SAS of a player’s gender. In Figure 1, SAS compares how accurate the champion model is for each gender group – thus we can determine if it is treating one group of people differently than another. Here, we see limited differences in bar size across our groups – thus the model performs at a similar accuracy rate across all levels of gender in our data.
Now that we have confirmed our model is fair, we need to ensure it is transparent for those decision makers dependent on the results. One way to do such is to look at an individual player to understand the key influences of his/her predicted outcome. In the HyperSHAP plot shown below, when looking at this individual, the most significant variables which positively contributed to their high likelihood of return are the number of total trips, days since last trip, and loyalty tier code, while the one which most negatively impact return probability is the player tenure.
We now have a machine learning model in which we are confident in. To implement it into a customer decisioning process, we want to integrate all available domain knowledge to ensure our model serves our bespoke business use case. Thus, we need to integrate the modeling output with a brand's business rules (or best practices) to generate a final treatment decision for our players: who is eligible for a specific marketing offer?
Realize the Marketing Vision With SAS Customer Intelligence 360: Activating Prescriptive Analytics To Improve Profit
SAS Customer Intelligence 360 enables brands to use first-party data to make better customer decisions using predictive analytics and machine learning in conjunction with business rules across a hub of channel touch points. As your brand's journey into analytical marketing use cases progresses, usage of modeling intellectual property cannot be under-exploited. It’s competitive differentiation awaiting to be deployed.
All of this culminates into a brand's guiding light framed around scalable customer decisioning. Rooted in a variety of analytical approaches that can be leveraged within a wide set of marketing use cases, it doesn't matter if one or multiple technology solutions serve as the bridge to the finish line. There is innovation being served from the software industry to appreciate, explore and experiment with. We (at SAS) simply want to help our customers through partnership and adoption, and whether it involves augmenting a 3rd party martech application to derive incremental value, or using SAS standalone to resolve challenges in orchestrating profitable customer experiences, one theme is clear.
2024 is showing a strong propensity for how marketing divisions will fall in love (again) with analytics, machine learning and AI for an array of new customer use cases. Our last demonstration video below will summarize how SAS analytically-driven martech capabilities broaden/enhance the value propositions of our casino gaming and hospitality use case.
Activating Prescriptive Analytics To Improve Profit With 3rd Party Applications
Decisioning use cases can either be executed on a schedule or in real-time, depending on the business need. For informing marketing campaigns of eligible players (or customers) with specific offers, a scheduled job may suffice to deploy those insights into a brand's 3rd party marketing hub when needed. Real-time execution receives live interactions or data points, processes them, and outputs the decision at the time and in the application necessary for the targeted consumer. Let’s go through one example summarizing when a brand would like to deploy the customer decisioning treatments into a 3rd party web app. This last demo will also provide an opportunity to go deeper and learn more about SAS Intelligent Decisioning.
Our vision at SAS is to serve as the market leader in advanced audience creation & targeting, independent of channel, for enterprise customers leveraging complex, disparate data sources and wishing to consistently deliver superior understanding into their customer journeys. In other words, we want to empower brands to practice responsible marketing.
Learn more about how SAS can be applied for customer analytics, journey personalization and integrated marketing here. For those who want to dive deeper into the net lift marketing use case highlighted in this article, check out a recently published webinar available on-demand here.
