Optimize your results & drive more impact with Actable AI

Explore tens of videos, examples, and articles to help you learn Actable AI, Data Science, AI, and much more.
Optimizing a marketing campaign

A company can optimize its marketing campaign by leveraging advanced analytics to maximize the return on investment (ROI). For example, a company can start with a small scale pilot campaign with a subset of potential customers and use advanced analytics to predict how customers respond. Then, the company can use the insights from the pilot  to develop a more targeted campaign and increase conversion and ROI.

The objective is to understand and predict customers' profiles who are more likely to buy the marketed product based on a small scale pilot campaign. Once the company has this insight the marketing campaign would target only the customers with the highest probability of conversion, increasing the return on investment. As a first step, we can perform simple data exploration to better understand existing associations between various variables such as income, education, past purchase history, and response rate among pilot customers.

Fig 1. Upper Chart: Pilot sample distribution based on marital status and the response.
Lower Chart: Pilot sample distribution based on education and the response. “1” purple shows the number of customers who accepted the offer.

Simple charts showing distribution of response among different customer profiles, such as the charts for the education and marital status above, give us a high level insight of who is positively and who is negatively reacting to our campaign. To better understand how each customer characteristic is associated with their response outcome, we can perform a correlation analysis to estimate the association between the response and all other characteristics.

Fig 2. Correlation between customer response and various other customer characteristics. 

The correlation analysis shows interesting insights, for example, NumCatalogPurchase (number of purchases made using catalog) has the highest positive correlation with the response rate. Whereas, Recency (number of days since the last purchase) has the largest negative correlation. Furthermore, we can see that response rate is also positively driven by MntMeatProducts (amount spent on meat products in the last 2 years) and MntWines (amount spent on wines in the last 2 years) among other factors and negatively driven by Dt_Customer_year (date of customer’s enrollment with the company) and Teenhome (number of teenagers in customer’s household) among others (Fig 2).

The next step will be to perform customer segmentation to better understand the similarities between customers and what characteristics make each group unique allowing us to improve our targeting. Clustering analysis (Fig 3) clearly shows 3 major groups based on the pilot results. Our Deep Learning based segmentation also automatically generates explanations for each group (Fig 3 lower graph).

Fig 3. Marketing campaign customer clustering to identify similar groups based on various available parameters. 

The customer segmentation shows that cluster 0 (blue) has a total of 577 customers from the pilot campaign out of which only 171 responded positively to the marketing campaign, representing 29.6%. On the other hand, cluster 1 (orange) and cluster 2 (green) have much lower response rates: cluster 1 = 9.0%, 86 positive responses out of 953; and cluster 2 = 10.8%, 77 positive responses out of 710. Thus, a robust segmentation analysis uncovers actionable insights enabling the company to improve its ROI for the main marketing campaign by only targeting customers similar to the high performing group (cluster 0 (blue) in Fig 3) based on the pilot results.

As a next step, we can leverage the pilot results to predict exactly which customers in our marketing list will most likely accept our offer. This will allow the company to narrow down the list of customers and increase conversion rate significantly. Fig 4 shows how we can easily use all metrics from the pilot data to predict the response for the new customers (the list that was not used in the pilot). All we need to do is select the appropriate analytics (Classification), target variable (Response), and variables used for predicting the target.

Fig 4. Prediction results showing which customers will respond positively or negatively to the marketing campaign along with the probabilities. Zero means negative and one means positive response.

Our automated machine learning (AutoML) will train several ML models on the pilot campaign data and then use the best model to predict which of the customers in our marketing campaign list are more likely to accept the offer if the company targets them. The prediction also generates various model performance metrics to show how good the results are and what parameters are important for the prediction and by what order (Fig 5).

Fig 5. Under the Performance tab, we can find various performance metrics showing how well is our model performing and what are the important variables for our predictions.

After performing correlation, segmentation, and classification analytics, the company already has deep insights on the customers groups and relevant characteristics to allow us to precisely target the potential customers who are more likely to positively respond to the marketing campaign. These actionable insights would save the company significant resources and increase marketing ROI.

Example Dataset

The dataset consists of 2240 customers data and their response to a marketing campaign conducted by a retail company with the goal of understanding and predicting customer behavior in order to maximize the return on investment in a marketing campaign.

AcceptedCmp1: 1 if customer accepted the offer in the 1st campaign, 0 otherwise

AcceptedCmp2: 1 if customer accepted the offer in the 2nd campaign, 0 otherwise

AcceptedCmp3: 1 if customer accepted the offer in the 3rd campaign, 0 otherwise

AcceptedCmp4: 1 if customer accepted the offer in the 4th campaign, 0 otherwise

AcceptedCmp5: 1 if customer accepted the offer in the 5th campaign, 0 otherwise

Response: 1 if customer accepted the offer in the last campaign, 0 otherwise

ResponseWithMissingValues: Same variable as Response but with some missing values for a prediction

Complain: 1 if the customer complained in the last 2 years

DtCustomer: Date of customer’s enrollment with the company

Education: customer’s level of education

Marital: customer’s marital status

Kidhome: number of small children in customer’s household

Teenhome: number of teenagers in customer’s household

Income: customer’s annual household income

MntFishProducts: amount spent on fish products in the last 2 years

MntMeatProducts: amount spent on meat products in the last 2 years

MntFruits: amount spent on fruits in the last 2 years

MntSweetProducts: amount spent on sweet products in the last 2 years

MntWines: amount spent on wines in the last 2 years

MntGoldProds: amount spent on gold products in the last 2 years

MntDealsPurchases: number of purchases made with discount

MntCatalogPurchases: number of purchases made using catalog

MntStorePurchases: number of purchases made directly in stores

MntWebPurchases: number of purchases made through company’s web site

MntWebVisitsMonth: number of visits to the company’s web site in the last month

Recency: number of days since the last purchase

Actable AI
Actable AI Technologies LTD
UK Company Number: 12669336
Malta Company Number: C99699