Campaign Optimization:
Maximizing the Value of Interacting with Your Customers

by Kurt Thearling

Published in the Relationship Marketing Report, August 2001


Introduction

In most marketing organizations, there are a wide variety of ways to interact with customers and prospects.    Besides the many possible offers that can be made, there are now multiple communication channels (direct mail, telemarketing, email, the web) that can be used.   The process of marketing campaign optimization takes a set of offers and a set of customers, along with the characteristics and constraints of the campaign, and determines which offers should go to which customers over which channels at what time. 

The process of using campaign optimization begins with a set of marketing offers.    The offers are created independently of the optimization process and typically have a number of financial characteristics (costs, etc.).     Each offer also has a predictive model associated with it that produces a score when a customer record is fed to the model.   The score can take a number of forms, including the probability that the customer will respond to the offer or the expected value of the offer for the customer.    The models themselves are built using a tool such as SAS Enterprise Miner and are incorporated into the campaign.

Using data mining for campaign optimization begins with data about purchasing behaviors for existing customers.     Most forms of optimization are nor much different than the analysis required for single product customer acquisition.     Each of the different possible offers is evaluated as if it were a single product offering.    The key is to then optimize the product offerings across all customers so that the offer (or offers) that a customer receives provides the greatest benefit for both buyer and seller.

Consider the following example.   Assume that you are a marketing manager for a financial services company.   You have the following products available for your customers:

·         Value checking account

·         Plus checking account

·         Standard credit card

·         Gold credit card

·         Platinum credit card

·         Primary mortgage

·         Secondary mortgage

Of these products, you are responsible for marketing the two mortgage products to your customers.  Your goal is to find out which customers might be interested in a mortgage offering.  Beyond this goal, you are also interested in maximizing the profitability of the mortgage product marketing campaigns. You have already done some thinking about your customers and their motivations in this area and came up with a couple scenarios, which you presented to your boss when pitching this new campaign:

1.  Customers preparing to buy a new home.   These customers might be building up cash reserves in their checking and/or savings account in order to put together a down payment.

2.  Customers preparing to re-finance an existing home.   These customers might be paying off credit card debt (thus making them more acceptable from a risk point of view) and hold a mortgage whose interest rate is higher than the current interest rate.

3. Customers preparing to add a second mortgage.   These customers might have increasing credit card debt, an on-time payment history for both their credit cards and existing mortgage (which means that they are a good risk), and enough equity in their house to cover the outstanding credit card balance.

Before you can begin to optimizer offers, you need to determine what kinds of offers you are interested in making available to your customers.   In this example, there are three possible offers (new first mortgage, re-finance of first mortgage, or a second mortgage), only one of which will be made to a customer (assuming that any offer is made at all).  Once the offers are determined, the next step is to collect the data to support the analysis. In recent years this has often meant the access to a data warehouse that holds all of your customer information but any kind of historical data repository should allow you to being the process of analyzing customer data to look for marketing opportunities.  The model that will be created as a result of the data mining process will predict the probability that a customer will sign up for a mortgage with your company.   By ranking the customers by their predicted probability, you will be able to identify the best prospects for your mortgage products.

The historical data contains all information that you have about your customers and their mortgage purchases, including demographic and account level information (age, income, marital status, and zip code) as well as transactional information (recent balances, number of purchases, types of purchases).   Your experience might also tell you to include external macroeconomic information from the time of the mortgage that could be relevant to the decision making process (e.g., the average mortgage rate at that time, housing starts, consumer confidence, etc.).   In the end you will end up with a collection of several hundred (or thousand) pieces of information about each customer at various points of time in their relationship with you.

It should be noted that some information might not be available, either because you don’t know (e.g., marital status), it doesn’t make sense (e.g., the outstanding first mortgage balance for someone who doesn’t already have a first mortgage with your company), or the data is missing (e.g., if they were new customers).   Depending on the data mining technique that is used to analyze the data set, these incomplete records might or might not be included in the eventual analysis.  In the case where customers are excluded from the analysis, they should not be ignored completely since they might be used in a different analysis with a different time frame or prediction target.

Data Analysis and Modeling

The optimization process begins by modeling desired responses to marketing offers over particular channels.  Modeling is the process whereby data mining algorithms analyze the historical offer data, creating mathematical functions (the models) that can be used to predict customer responses for a specific offer/channel combination.   In the example we are going to consider below, all of the offers will be sent out via a single channel (direct mail) but it is easy to extend this scenario such that each offer could be sent out via multiple channels.  

The simplest kind of modeling is to use the average historical response rate for an offer and assign it to all customers eligible for the offer.   If the mortgage offer has traditionally received a 3.2% response rate, on average any new prospect receiving the offer will responds at the same rate.   This kind of modeling is very simple and can be used to get started in optimization.  But once you have some experience, it makes sense to move onto more complicated data mining algorithms to improve the accuracy of the response predictions. 

The process of data mining can be broken down into sub-processes, each of which involves creating models for each of the different offers.   At this point the analysis for each offer is independent of the other offers.    There might be some overlap in the customers you use to carry out the analyses but the actual model building processes would be independent.  For example, one customer might buy a house with a mortgage and then later add on a second mortgage.

Once the offer models have been generated, each can be applied to new customer data in order to make predictions about those customers.    The scores are simply the outputs of the models and might correspond to the predicted probability that a customer will purchase a specific mortgage product two months in the future.   Since we have three different offers there will be three different scores for each customer.    This ends up producing a matrix of scores, with one row for each customer and one column for each offer score.  The final step of the process is the optimization of the scoring matrix, which selects which of the multiple offers will be made to each customer.

Scoring

Once you have generated the three for the three different offers, it is time to apply them to new customer data to determine which mortgage offers to send to which people.   Deciding which customer to score with a particular model might require some thought since there is usually some sort of eligibility criteria used to pre-select customers for consideration of a particular offer.   For example, you might score only those customers with the re-finance model who you know don’t not have a house.   This might mean that you would score a customer with the “New Mortgage Model” even though they might have already signed a mortgage with another company, though you don’t know it.   You might also use a set of risk criteria in order to remove from the process those customers who are considered to have a high risk of non-payment.

In the end you will generated a matrix like the one below.   For each customer, you will have three different scores.   In some cases, the customer/score entry is NULL due to the fact the customer was not eligible to receive an offer. 

Customer

New Mortgage Score

Re-finance Score

Second Mortgage Score

Tom Adams

0.2422

0.4926

0.0872

George Castle

0.8600

0.4465

0.0982

Betty Hanson

NULL

0.9700

0.4453

Robert Marcus

0.7854

NULL

NULL

Carol Smith

0.5063

NULL

NULL

Beverly Thompson

0.8210

0.5014

0.6386

Bill Samuels

NULL

0.5057

0.9177

Terry Jones

0.2226

0.1352

0.0888

Chris Peters

0.2928

0.1732

0.5244

 

Turning Scores into Value

Once the scores are available, the next step is to convert each of the scores into profitability values.    In the simplest approach, each offer has an economic value associated with it (and each offer’s value can be different).   This value is the average over the potential customers and is usually determined by looking at characteristics of existing customers in the historical dataset.   For example, the value (to the company) for a new home mortgage might be, on average, $6000 per customer.   Some customers might be worth more, and some less, but on average they are worth $6000.  

Since the previously computed offer scores are the probabilities that a customer will respond to a particular offer, multiplying the economic values for each offer by the model scores will generate the expected average economic value for each customer / offer combination.   For each customer the offer with the highest expected economic return is highlighted.  

These financial characteristics of an offer can get complicated, with numerous facts about each customer used to evaluate the value of a particular offer/customer combination.    For example, you might want to compute the net present value (NPV) of a customer, then evaluate their risk of filing bankruptcy and factor it into risk adjusted NPV, and then combine this with the expected value for the mortgage offering.  In some cases, it might be possible to use data mining technology to model the value directly, skipping the response probability stage.  Assuming that the models are reliable, this would increase the accuracy of the optimization process.

Constraints

Before the optimization process can go any further, any constraints that the optimization process must handle need to be specified.   Constraints put limits on the marketing campaign based on external factors. For example, there could be a budget limitation for the campaign, which limits the number of offers that could be made.   This would result in more inexpensive offers than would otherwise be made.

Some possible constraints include: 

·     Global Constraints.   Examples include maximum campaign budget,  maximum number of offers, maximum number of responses, and minimum number of responses.

·     Per-offer Constraints.   Examples of per-offer constraints include minimum number of offers, maximum number of offers, minimum number of response, maximum number of responses, maximum total offer cost, and minimum offer net response value.

·     Per-channel Constraints.   Examples of per-channel constraints include minimum number of outbound

·     Per-customer Constraints.  Examples of per-customer constraints include minimum number of offers per customer, maximum number of offers per customer, minimum number of responses per customer, maximum number of responses per customer, minimum response value per customer, maximum number of equivalent offers per customer, maximum number of offers per time interval, and maximum number of offers on a specific channel per customer.

·     Per-household Constraints. Examples of per-household constraints include minimum number of offers per household, maximum number of offers per household, minimum number of responses per household, maximum number of offers per household, minimum response value per household, maximum number of equivalent offers per household, maximum number of offers per time interval, and maximum number of offers on a specific channel per household.

·     Geographic Constraints.  Examples include minimum and/or maximum numbers of offers per geographic region.   You might want to make sure that your offers are distributed evenly across your sales territories.

·     Segment Constraints.   Minimum and/or maximum number of offers per segment of your customer base.     Because of regulatory conditions, you might need to make sure that a particular segment (e.g., by race, gender, etc.) of your customer base is not underrepresented in a marketing campaign. 

Sometimes it is not possible to satisfy all of the external constraints that are specified.  It might be that some constraints are contradictory or simply that the characteristics of the customers do not allow for the constraints to be met.  For example, one constraint might be that customers selected for a particular offer be from each of the 50 United States and that each state have at least 1000 recipients of the offer.   If there are only 500 potential customers from North Dakota in the pool, it will be impossible to select 1000 to receive the offer, regardless of the specified criteria.

In the event that all constraints cannot be met, the process can either “hard fail” or it can fail gracefully.   In a “hard fail,” the process aborts when it is determined that all constraints cannot be met.   The user is then informed of the problem and they can re-formulate the constraints and try again.   In the case of a graceful failure, the optimization process attempts to meet as many of the constraints as possible despite the fact that it is not possible to satisfy all of them.   The constraints may be weighted to that the process takes into consideration their relative importance.

Optimization

The optimization operates on the constraints and the values that were generated by combining the scores with the financial information about each offer.     The primary objective the optimization is to maximize the overall value of the marketing campaign without violating any of the user specified constraints.

Taking our earlier example further, assume that a new home mortgage is worth $6,000 per customer, a second mortgage is worth  $500, and a re-financed mortgage is worth $2000.   Also assume that there are two constraints, that each offer can be assigned to the same number of customers as the other offers, and that each customer can receive only one offer.    The table below shows the values of each offer for each customer as well as the offer that the optimization selected for that customer. 

 

Customer

New Mortgage Score

Re-finance Score

Second Mortgage Score

Tom Adams

$1,452.94

$985.28

$435.95

George Castle

$5,160.14

$893.04

$490.80

Betty Hanson

NULL

$1,939.97

$2,226.57

Robert Marcus

$4,712.41

NULL

NULL

Carol Smith

$3,037.68

NULL

NULL

Beverly Thompson

$4,925.71

$1,002.75

$3,192.89

Bill Samuels

NULL

$1,011.48

$4,588.70

Terry Jones

$1,335.85

$270.34

$443.90

Chris Peters

$1,757.01

$346.34

$2,621.97

 

Notice that the most valuable offer is not always selected.  In fact, for four customers (George, Betty, Beverly, and Terry) the most valuable offer was not selected because of the constraints.




[ Data Mining Page ] [ White Papers ] [ Data Mining Tutorial ]