Optimizing marketing campaigns with data

2025

Abstract

Effective marketing relies on targeting the right customers. This project analyzes a retail food company’s direct marketing campaign, which suffered a low response rate (15%) and financial loss. Using data-driven strategies, we optimized future campaigns through customer segmentation, predictive modeling, and profit analysis.

Customers were grouped into five segments using RFM analysis, revealing that high-value customers (Valuable & Elite segments) were significantly more likely to convert. A logistic regression model identified key purchase drivers, showing that Elite customers were over 140 times more likely to buy than Dormant customers.

Profit analysis determined the optimal targeting threshold (0.27), increasing the conversion rate from 15% to 32.5% and turning losses into a projected 3,722 MU gain. This study highlights how data-driven marketing can improve efficiency, reduce costs, and maximize returns, providing actionable insights for businesses looking to optimize their campaigns.

The dataset and the project notebooks, with all the code, can be seen on the following GitHub repository:

Introduction

Marketing campaigns are a critical component of business growth, but their success hinges on targeting the right customers. The case study presented here focuses on a real-world scenario where a retail food company launched a direct marketing campaign that resulted in a low response rate and financial loss. Leveraging data-driven strategies, we aim to optimize future campaigns by identifying key customer segments, predicting purchase behavior, and maximizing profitability.

The methodology involves:

• Data cleansing and preprocessing to refine the dataset for analysis.
• Exploratory data analysis to understand customer behavior and past campaign responses.
• Customer segmentation based on RFM analysis to identify distinct behavioral patterns.
• Predictive modeling to forecast customer responses.
• Profit analysis to determine the optimal targeting strategy for future campaigns.
• Recommendations for marketing teams to enhance campaign effectiveness.

The dataset and the project notebooks, with all the code, can be seen on the following GitHub repository:

The company context

The dataset consists of customer information, purchase history, and responses to past marketing campaigns of a retail food company. Let’s understand the company context first.

They sell products from 5 major categories: wines, rare meat products, exotic fruits, specially prepared fish and sweet products. These can further be divided into gold and regular products. The customers can order and acquire products through 3 sales channels: physical stores, catalogs and company’s website. Globally, the company had solid revenues and a healthy bottom line in the past 3 years, but the profit growth perspectives for the next 3 years are not promising… For this reason, several strategic initiatives are being considered to invert this situation. One is to improve the performance of marketing activities, with a special focus on marketing campaigns.

The marketing department was pressured to spend its annual budget more wisely. The CMO perceives the importance of having a more quantitative approach when taking decisions, reason why a small team of data scientists was hired with a clear objective in mind: to build a predictive model which will support direct marketing initiatives. Desirably, the success of these activities will prove the value of the approach and convince the more skeptical within the company.

The aim of the project

The objective of the team is to build a predictive model that will produce the highest profit for the next direct marketing campaign, scheduled for the next month. The new campaign, sixth, aims at selling a new gadget to the Customer Database.

To build the model, a pilot campaign involving 2.240 customers was carried out. The customers were selected at random and contacted by phone regarding the acquisition of the gadget. During the following months, customers who bought the offer were properly labeled. The success rate of the campaign was 15%.

The objective of the team is to develop a model that predicts customer behavior and to apply it to the rest of the customer base. Hopefully, the model will allow the company to cherry pick the customers who are most likely to purchase the offer while leaving out the non-respondents, making the next campaign highly profitable. Moreover, apart from maximizing the profit of the campaign, the CMO is interested in understanding the characteristic features of those customers who are willing to buy the gadget.

Understanding the Data

The data set contains socio-demographic and firmographic features about 2.240 customers who were contacted. Additionally, it contains a flag for those customers who responded the pilot campaign, by buying the product.

The following features were included in the dataset:

• Personal Data:
  • ID: Unique customer identifier
  • Year_Birth: Customer’s year of birth
  • Education: Customer’s education level
  • Marital_Status: Customer’s marital status
  • Income: Customer’s annual family income
  • Kidhome: Number of children in the customer’s household
  • Teenhome: Number of teenagers in the customer’s household
  • Dt_Customer: Date the customer enrolled with the company
  • Recency: Number of days since the customer’s last purchase
  • Complain: 1 if the customer has complained in the last 2 years, 0 otherwise
• Product Data
  • MntWines: Amount spent on wine in the last 2 years
  • MntFruits: Amount spent on fruits in the last 2 years
  • MntMeatProducts: Amount spent on meat in the last 2 years
  • MntFishProducts: Amount spent on fish in the last 2 years
  • MntSweetProducts: Amount spent on sweets in the last 2 years
  • MntGoldProds: Amount spent on gold products in the last 2 years
• Campaign Data
  • NumDealsPurchases: Number of purchases made with a discount
  • AcceptedCmp1: 1 if the customer accepted the offer in the 1st campaign, 0 otherwise
  • AcceptedCmp2: 1 if the customer accepted the offer in the 2nd campaign, 0 otherwise
  • AcceptedCmp3: 1 if the customer accepted the offer in the 3rd campaign, 0 otherwise
  • AcceptedCmp4: 1 if the customer accepted the offer in the 4th campaign, 0 otherwise
  • AcceptedCmp5: 1 if the customer accepted the offer in the 5th campaign, 0 otherwise
  • Response: 1 if the customer accepted the offer in the last (pilot) campaign, 0 otherwise (target)
• Purchase Location Data
  • NumWebPurchases: Number of purchases made through the company’s website
  • NumCatalogPurchases: Number of purchases made using a catalog
  • NumStorePurchases: Number of purchases made directly in stores
  • NumWebVisitsMonth: Number of visits to the company’s website in the last month

Data cleansing and feature engineering

The dataset was available as a CSV file, which was loaded into a Pandas DataFrame for further analysis. The initial exploration revealed a few missing values, which were dropped from the dataset. Additionally, some features were deemed irrelevant for the analysis and were removed. Outliers were identified and treated accordingly to ensure the integrity of the data.

Some features were transformed to better represent the customer’s behavior, such as the Dt_Customer column, which was converted to the number of days and months since the customer enrolled with the company. This transformation allowed for a more meaningful analysis of the recency feature. Also, the age of the customers was calculated based on their year of birth to provide a more intuitive understanding of the data.

As described earlier, the company sells products from 5 major categories, which were represented in the dataset as the amount spent on each category in the past 2 years. The values of these features were summed to create a total amount spent feature, which was used to analyze the customer’s purchasing behavior in subsequent steps.

Finally, the features were divided into numerical and categorical groups for further analysis.

Exploratory Data Analysis

After cleaning and preprocessing the data, an exploratory data analysis was conducted to gain insights into customer behavior and past campaign responses. The analysis included descriptive statistics, distribution plots, and correlation matrices to identify patterns and relationships within the data.

Concerning the numerical features, it was observed that most features have a right-skewed distribution, indicating that a few customers spend significantly more than the rest. The histograms below show the histograms of these features, using the target variable Response to differentiate between customers who accepted the offer and those who did not. A Response value of 1 indicates that the customer accepted the offer in the pilot campaign, while a value of 0 indicates that the customer did not accept the offer.

We see that customers who accepted the offer in the last campaign tend to have higher expenses in most product categories, as well as a higher number of purchases and lower recency values. This suggests that customers who are more engaged with the company are more likely to respond positively to marketing campaigns.

For the categorical features, the distribution of education levels, marital status, among others, was analyzed to understand the customer profile. The bar plots below show the distribution of these features based on the Response variable.

The analysis revealed that customers with higher education levels and income tend to respond more positively to marketing campaigns. Additionally, customers with partners and those with children in the household are less likely to accept the offer. The age does not seem to have a significant impact on the response rate.

Some of these relationships were further explored through boxplots:

We can see that customers with children tend to have lower incomes therefore, they are less likely to accept the offer. The education level seems to have a significant impact on the income of the customers.

Besides the visual analysis, statistical tests were performed to identify significant differences between the groups of customers who accepted and did not accept the offer. Mann-Whitney U tests were conducted for numerical features, while chi-square tests were used for categorical features. The results of these tests were used to identify the most relevant features to keep in the dataset for further analysis.

Customer Segmentation: Identifying Behavioral Patterns

Customer segmentation is a powerful technique that allows businesses to group customers based on their behavior and preferences. In this project, we used the RFM segmentation methodology.

RFM analysis is a customer segmentation technique that uses past purchase behavior to divide customers into groups. RFM stands for Recency, Frequency, and Monetary Value.

• Recency: How recently a customer has made a purchase. The more recent the purchase, the higher the score.
• Frequency: How often a customer makes a purchase. The more frequent the purchase, the higher the score.
• Monetary Value: How much money a customer spends on purchases. The more money spent, the higher the score.
• RFM score: A score calculated by combining the three RFM values.
• RFM segment: A segment of customers with similar RFM scores.

There are many ways to calculate the RFM score. The number of groups may vary depending on the business needs, the size of the dataset, and the distribution of the RFM values. Similarly, the way to calculate the RFM score may vary depending on the business needs. And, finally, the segment labels may vary depending on the business needs.

For small and medium datasets, it is better to calculate the RFM score as a number. This number can be calculated by summing the three RFM values. Considering a scale of 1 to 5 for each RFM value, the RFM score will be between 3 (1 + 1 + 1 = 3) and 15 (5 + 5 + 5 = 15). So, the number of possible RFM scores is 15 - 3 + 1 = 13. This is the approach we are going to use in this project since our dataset is relatively small.

Irrespective of the way to calculate the RFM score, the RFM segment is calculated by dividing the RFM score into groups. Each group can be labeled with a name, such as “Champions”, “Loyal Customers”, “Potential Loyalists”, “New Customers”, “At Risk”, “Can’t Lose Them”, “Hibernating”, “Lost Customers”, etc. The number of groups and the labels may vary depending on the business needs and the distribution of the RFM values. The labels try to describe the behavior of the customers in each group.

For instance, in this project, we are going to consider the following RFM score calculation:

• Each RFM value is divided into 5 equal groups, with a score of 1 to 5.
• The RFM score is calculated by the sum of the three RFM values.
• The RFM score is divided into 5 segments, that may have different marketing strategies, as follows:

1. Dormant (Score: 3 - 5)

• Behavior: These customers have the lowest engagement. They haven’t purchased recently, buy infrequently, and spend little.

2. Occasional (Score: 6 - 7)

• Behavior: Customers in this group make occasional purchases but are not highly engaged. They may have made a few transactions but do not shop consistently.

3. Engaged (Score: 8 - 10)

• Behavior: These customers have moderate engagement. They purchase somewhat frequently, and their spending is stable.

4. Valuable (Score: 11 - 12)

• Behavior: These are high-value customers who purchase often and spend significantly. They have shown consistent engagement.

5. Elite (Score: 13 - 15)

• Behavior: The most valuable customers. They buy frequently, spend a lot, and have remained engaged over time.

This segmentation allows for targeted marketing strategies to maximize customer retention and revenue growth.

This segmentation was applied to the dataset, and the customers were assigned to one of the five segments based on their RFM scores. The distribution of scores and the number of customers in each score is shown below:

Customers were then grouped into the five segments based on their RFM scores. The distribution of customers across the segments is shown below:

Considering the median values of the normalized RFM scores, we can plot the distribution of the RFM scores for each segment:

We can see that the Elite and Valuable segments are the most important in terms of recency, frequency, and monetary value. These segments represent the ones which are most likely to respond positively to marketing campaigns and generate higher revenue for the company.

Profiling the Customer Segments

It is always important to understand the characteristics of each customer segment to tailor marketing strategies effectively. One way to do this is by analyzing the key features of each segment and how they differ in terms of engagement and spending. We can create profiles, or personas, for each segment to better understand their behavior and preferences.

In order to do this, let’s analyze how the numerical and categorical features vary across the five segments. First, we will look at the numerical features:

As we can see, the Elite and Valuable segments have higher spending, more frequent purchases, and more recent transactions compared to the other segments. Also, they have a higher income. This indicates that these segments are more engaged with the company and are likely to respond positively to marketing campaigns.

For the categorical features, we can use each category as a hue to see how the segments differ in terms of education, marital status, and other characteristics:

It is clear that the higher the RFM score the lower the tendency to have children. Also, the most valuable segments have older customers (higher percentages for 46-60 and 61+ age groups) and higher percentages of customers with higher education levels (graduation or higher).

The marital status is more evenly distributed among the segments, with no clear pattern.

We can as well use the segments as a hue:

The findings are consistent with the previous analysis. There is a clear pattern in the distribution of the segments for the number of children, age group, and education level. The marital status is more evenly distributed among the segments.

We can now describe our personas or customer profiles for each segment:

• Dormant
  • Description: These customers have the lowest engagement. They haven’t purchased recently, buy infrequently, and spend little.
  • Marketing Strategy: Reactivation campaigns such as discounts or personalized offers may help re-engage them.
  • Customer Profile:
    • Demographics: Customers with children, younger age group, lower education level, and lower income.
    • Behavior: Low recency, low frequency, low monetary value.
    • Tendency to convert: Lowest.
• Occasional
  • Description: Customers in this group make occasional purchases but are not highly engaged. They may have made a few transactions but do not shop consistently.
  • Marketing Strategy: Encourage more frequent purchases through targeted promotions or loyalty incentives.
  • Customer Profile:
    • Demographics: Customers with children, younger age group, lower education level, and lower income.
    • Behavior: Mid-range recency, low frequency, low monetary value.
    • Tendency to convert: Mid to low.
• Engaged
  • Description: These customers have moderate engagement. They purchase somewhat frequently, and their spending is stable.
  • Marketing Strategy: Strengthen the relationship by providing personalized recommendations and exclusive offers.
  • Customer Profile:
    • Demographics: Customers with children, mid-age group, at least graduation as education level, and mid-range income.
    • Behavior: Mid-range recency, mid-range frequency, low monetary value.
    • Tendency to convert: Mid-range.
• Valuable
  • Description: These are high-value customers who purchase often and spend significantly. They have shown consistent engagement.
  • Marketing Strategy: Maintain their loyalty through rewards programs, premium offers, and early access to promotions.
  • Customer Profile:
    • Demographics: Customers may have children or not, older age group, higher education level, and higher income.
    • Behavior: Mid-range recency, high frequency, high monetary value.
    • Tendency to convert: High.
• Elite
  • Description: The most valuable customers. They buy frequently, spend a lot, and have remained engaged over time.
  • Marketing Strategy: Focus on VIP treatment, premium experiences, and strong retention strategies to ensure long-term loyalty.
  • Customer Profile:
    • Demographics: Customers probably without children, older age group, higher education level, and higher income.
    • Behavior: Low recency, high frequency, high monetary value.
    • Tendency to convert: Highest.

Predicting Customer Responses with Machine Learning

The segmentation analysis provided valuable insights into customer behavior, but to maximize the effectiveness of future marketing campaigns, we need to predict customer responses. This is where machine learning comes into play.

The segments identified through RFM analysis can be used as input features, along with the existing customer data, to build a predictive model that forecasts the likelihood of a customer accepting the offer in the next campaign.

The Sci-Kit Learn library in Python offers a wide range of machine learning algorithms for classification tasks. For this project, an initial evaluation of several models was conducted with the following models:

• Dummy Classifier (Baseline)
• Logistic Regression
• Decision Tree
• LightGBM
• Support Vector Machine
• K-Nearest Neighbors

Since the target variable is imbalanced, with only 15% of customers accepting the offer in the pilot campaign, the parameter class_weight='balanced' was used to account for this imbalance in the models that support it.

The preprocessing steps included one-hot encoding and ordinal encoding for the categorical features, and standard and power transformations for the numerical features (only for the non-tree-based models).

A stratified 5-fold cross-validation was used to evaluate the models’ performance. The evaluation metrics considered were accuracy, precision, recall, F1-score, AUROC, and AUPRC. The time to train and predict was also considered. The results are summarized in the performance comparison chart below:

The choice of the best model should consider some factors:

• The metrics: which model has the best performance in the metrics that are most important for the business;
• The time: the model that has the best performance in the metrics and the lowest time should be chosen;
• The interpretability: some models are more interpretable than others. If interpretability is important, the model that is easier to interpret should be chosen.

It happens that the best model in terms of metrics is not always the best model in terms of time or interpretability. So, the choice of the best model should consider all these factors.

Concerning the metrics, some literature suggests that the area under the precision-recall curve (AUPRC, represented as average_precision on Scikit-Learn) is a better metric for imbalanced datasets than the area under the ROC curve (AUROC, or roc_auc). This is because the ROC curve is not sensitive to changes in the false positive rate when the true positive rate is high. The precision-recall curve, on the other hand, is sensitive to changes in the false positive rate when the true positive rate is high. I’ve written an article about this topic, which can be seen here.

We can see that the logistic regression model has the best performance in terms of the chosen metric and is also one of the fastest models. Besides, it is a very interpretable model. So, we chose the logistic regression model for this project.

The chosen model was then fine-tuned using grid search to optimize the hyperparameters.

The final model had the following metrics:

Metric	Mean	Std Dev
Accuracy	0.819	0.006
Balanced Accuracy	0.817	0.020
F1	0.576	0.019
Precision	0.445	0.009
Recall	0.814	0.042
ROC AUC	0.898	0.011
Average Precision	0.669	0.028

We see that our model highly favors recall in detriment of precision. This is because the model is giving more importance to the conversion class, which is the minority class. It seems fine, as we are more interested in capturing the conversion cases.

The coefficients of the logistic regression model can provide insights into which factors influence the likelihood of a customer accepting the offer. The coefficients can be transformed into odds ratios to understand the impact of each feature on the target variable. Below is a chart showing the odds ratios of the features:

Using MntMeatProducts as an example for a numerical feature, we see that the odds ratio is ~4.47. This means that a one-unit increase in the MntMeatProducts feature multiplies the odds of the target variable by 4.47. This corresponds to a 347% increase in the odds (i.e., the odds become 4.47 times larger).

For HasAcceptedCmp_1, a binary feature, we have to consider that we have set the OneHotEncoder to drop the first category when dealing with binary features. So, the reference category is the category that was dropped. In this case, the reference category is 0, that is why there is not HasAcceptedCmp_0 in the coefficients list. The odds ratio is ~1.42. This means that customers who have accepted at least one previous campaign have 1.42 times the odds of converting compared to customers who have not accepted any campaign.

Finally, considering now a multi-class feature, Segment. We have set the OneHotEncoder to drop the first category only for binary features. Since Segment has 5 categories, we have 5 coefficients. It is easier to interpret the odds ratio of a multi-class feature by comparing the odds ratio of each category to a reference category. This can be done dividing the odds ratio of each category by the odds ratio of the reference category. Let’s consider the Dormant category as the reference category. The odds ratio of the Dormant category is 1.

\[\text{Relative Odds Ratio} = \frac{\text{Odds Ratio of Segment}}{\text{Odds Ratio of Lowest Segment (Dormant)}}\]

Applying this to your data:

Segment	Absolute Odds Ratio	Adjusted (Relative to Dormant)
Dormant	0.0619	1.00 (Baseline)
Occasional	0.3161	\(\frac{0.3161}{0.0619} = 5.11\)
Engaged	1.2011	\(\frac{1.2011}{0.0619} = 19.40\)
Valuable	3.9805	\(\frac{3.9805}{0.0619} = 64.28\)
Elite	8.8405	\(\frac{8.8405}{0.0619} = 142.76\)

• Occasional customers are 5.1 times more likely to convert compared to Dormant customers.
• Engaged customers are 19.4 times more likely than Dormant customers.
• Valuable customers are 64.3 times more likely than Dormant customers.
• Elite customers are 142.8 times more likely than Dormant customers.

These are important insights that can help the marketing team to focus on the most valuable customers and improve the conversion rate. The figures are quite impressive, showing that the segmentation of customers is a powerful tool to predict the probability of conversion.

Maximizing Campaign Profitability

We have already built a model that predicts the probability of a customer buying the gadget. Also, we have already segmented our customers. Now, we will use this model to predict the profit of the campaign, aiming to maximize it.

The profit of the campaign is given by the formula:

\[\text{Profit} = \text{Number of customers} \times \text{Probability of buying} \times \text{Revenue} - \text{Number of customers} \times \text{Cost}\]

Where:

• Number of customers: the number of customers that we target in the campaign
• Probability of buying: the probability of a customer buying the gadget
• Revenue: the revenue that we get from selling the gadget
• Cost: the cost of the campaign
• Profit: the profit of the campaign

From the problem description, we know that:

• The revenue from selling the gadget is 11
• The cost of the campaign is 3

We can use our model to predict the probability of each customer buying the gadget. Then, we will search for a threshold that maximizes the profit of the campaign. We will try different thresholds and calculate the profit for each one. We will choose the threshold that gives us the highest profit.

What is the threshold? The threshold is the probability above which we will consider a customer as a buyer. For example, if the threshold is 0.5, we will consider that a customer is a buyer if the probability of buying is greater than 0.5. If the threshold is 0.7, we will consider that a customer is a buyer if the probability of buying is greater than 0.7.

Why do we need to choose a threshold? Because the model gives us the probability of a customer buying the gadget, not a binary answer. We need to choose a threshold to convert the probability into a binary answer (buyer or not buyer).

If our threshold is too low, we will consider many customers as buyers, but many of them will not buy the gadget. Our cost will be high (many customers targeted), but our revenue will be low (few customers buying the gadget).

If our threshold is too high, we will consider few customers as buyers (low cost), and even considering that all of them will buy the gadget, our revenue will also be low (few customers buying the gadget).

We need to find a balance between these two situations.

For each threshold, we are going to calculate:

• the number of customers that we target in the campaign
• the expected conversion (probability of buying, we will consider as the sum of the probabilities of the customers that we target)
• the conversion rate (expected conversion divided by the number of customers that we target)
• the expected revenue (expected conversion times the revenue)
• the expected cost (number of customers that we target times the cost)
• the profit (expected revenue minus expected cost)

We can then plot the profit as a function of the threshold. The threshold that gives us the highest profit is the optimal threshold. The chart below shows the profit as a function of the threshold:

The optimal threshold is 0.27. This threshold gives us the highest profit for the campaign, with a profit of 3722.18.

Using this threshold, we can have a look at the selected customers (those with a probability of buying equal or greater than 0.27) and see the segments they belong to. The table below shows the conversion rates per segment at the optimal threshold:

Segment	Conv. Rate	Selected Cust.	Cum. % of Total	Converted Cust.	Cum. % Converted
Dormant	0.231	26	1.2	6	0.6
Occasional	0.218	119	6.6	25	3.2
Engaged	0.275	309	20.6	84	11.9
Valuable	0.336	271	32.9	91	21.3
Elite	0.452	239	43.7	108	32.5

We see that this threshold takes 43.7% of the original customers of our dataset. This is good, as we are not targeting too few customers (which would decrease our revenue) or too many customers (which would increase our cost).

From the selected customers, we converted 32.5% of them into buyers. This is a good conversion rate, as it is higher than the conversion rate of pilot campaigns (15 %). So the segmentation and the model are working well. Also, the conversion rate per segment is higher than the conversion rate of the pilot campaign.

Most of the selected customers are in segments “Valuable” and “Elite”. These are the segments that we are most interested in targeting, as they are the most likely to buy the gadget (higher conversion rate).

We can plot the conversion rates per segment at the optimal threshold:

Another interesting result is that every segment has a higher conversion rate than the pilot campaign. The pilot campaign had a conversion rate of 15%, and all segments have a conversion rate higher than 15%.

We can think of this as the following: without the model, if we targeted all customers, randomly, we would have a conversion rate of 15%. With the model, we can target only 43.7% of the customers and have a conversion rate of 32.5%. This is a significant improvement. More, each segment has a conversion rate higher than 15%. So, even segments with lower conversion rates are better than the pilot campaign.

A formal way of saying the above statement is that the model lifts the conversion rate of the campaign. The lift is the ratio between the conversion rate of the campaign with the model and the conversion rate of the campaign without the model. We can calculate the lift for each segment based on the conversion rate of the pilot campaign and the overall lift:

We see that all segments have a lift higher than 1, which means that the model is improving the conversion rate of the campaign. The overall lift is 2.15, which means that the model is improving the conversion rate of the campaign by 115%. The lift is even higher for the segments “Valuable” and “Elite”, which are the segments that we are most interested in targeting.

Key Takeaways and Business Impact

This project demonstrated how data-driven strategies can significantly enhance the effectiveness of marketing campaigns. By leveraging customer segmentation and predictive modeling, we identified high-value customers, optimized targeting strategies, and improved campaign profitability.

Main Insights from the Analysis

1. Customer behavior varies significantly by segment

   • Customers were successfully segmented using RFM analysis into Dormant, Occasional, Engaged, Valuable, and Elite groups.
   • Higher-value segments (Valuable & Elite) exhibit more frequent purchases, higher spending, and greater engagement with previous campaigns.

2. Predictive modeling improves campaign efficiency

   • A logistic regression model was trained and fine-tuned to predict customer responses.
   • Key factors influencing purchase behavior included past campaign engagement, spending habits, and recency of purchases.
   • The model was able to identify customers up to 142 times more likely to convert compared to the least engaged segment.

3. Data-driven targeting leads to higher profitability

   • The optimal decision threshold (0.27) increased the conversion rate from 15% to 32.5% while reducing the number of customers contacted.
   • The campaign profit, which was negative in the pilot test, was optimized to a projected 3,722 MU gain.
   • Every customer segment performed better than the original pilot campaign when targeted using model predictions.

How These Findings Improve Future Campaigns

• More precise targeting: By focusing on high-potential customers, marketing teams can reduce costs while increasing conversion rates.
• Better budget allocation: Resources can be directed towards segments with the highest expected return, maximizing overall profitability.
• Personalized marketing: Understanding customer segments allows for tailored promotions that align with specific behaviors and preferences.
• Data-driven decision-making: Predictive analytics can help forecast campaign performance and refine strategies over time.

Actionable Recommendations for Marketing Teams

📌 Leverage customer segmentation

• Develop targeted messaging and promotions for Valuable and Elite segments, as they have the highest likelihood of conversion.
• Design reactivation campaigns for Dormant and Occasional customers to increase engagement.

📌 Use predictive modeling to refine campaign targeting

• Apply machine learning models to predict customer responses and focus marketing efforts on the most promising prospects.
• Optimize decision thresholds based on profitability rather than simple conversion rates.

📌 Incorporate profitability analysis into marketing strategy

• Regularly analyze profit vs. targeting threshold to balance reach and expected return.
• Continuously monitor campaign performance and adjust strategies accordingly.

📌 Experiment with personalized offers

• Tailor promotions based on customer purchase history and engagement level to increase response rates.
• Test different pricing strategies for higher- and lower-engagement customer groups.

Final Thoughts

This case study highlights the transformative power of data in marketing decision-making. By combining segmentation, predictive modeling, and profit analysis, businesses can enhance efficiency, improve customer engagement, and maximize returns. Companies that adopt these strategies will gain a competitive advantage in customer retention and revenue growth.

See the full project in the GitHub repository.