ABSTRACT

The banking industry regularly mounts campaigns to improve customer value by offering new products to existing customers. In recent years this approach has gained significant momentum because of the increasing availability of customer data and the improved analysis capabilities in data mining. Typically, response models based on historical data are used to estimate the probability of a customer purchasing an additional product and the expected return from that additional purchase. Even with these computational improvements and accurate models of customer behavior, the problem of efficiently using marketing resources to maximize the return on marketing investment is a challenge. This problem is compounded because of the capability to launch multiple campaigns through several distribution channels over multiple time periods. The combination of alternatives creates a complicated array of possible actions. This paper presents a solution that answers the question of what products, if any, to offer to each customer in a way that maximizes the marketing return on investment. The solution is an improvement over the usual approach of picking the customers that have the largest expected value for a particular product because it is a global maximization from the viewpoint of the bank and allows for the effective implementation of business constraints across customers and business units. The approach accounts for limited resources, multiple sequential campaigns, and other business constraints. Furthermore, the solution provides insight into the cost of these constraints, in terms of decreased profits, and thus is an effective tool for both tactical campaign execution and strategic planning.

Keywords

Database marketing, Cross-selling, Up-selling, Profit Optimization, Assignment Problems, Constrained Optimization. Response Models.

1.    INTRODUCTION

The new mantra of database marketing in banking and financial services is “the right product to the right customer at the right time”. However, a practical and effective implementation of this goal is not easy to accomplish. What makes this particularly difficult is that companies have multiple products and operate under a complex set of business constraints. Choosing which products to offer to which customers in order to maximize the marketing return on investment and meet the business constraints is enormously complex. This paper outlines a framework for solving this problem and presents an example using data from Scotiabank.

Scotiabank is one of North America’s premier financial institutions; it is comprised of Domestic Banking, Wealth Management, International Banking and Scotia Capital groups. The Domestic Bank employs more than 23,000 people and has over 6 million customers. The Wealth Management Group incorporates key personal investment and advisory activities within the Scotiabank Group. Scotiabank is the most international of all Canadian banks, its International Banking Group has more that 21,000 employees and provides retail banking services in over 50 countries. The Scotia Capital Group provides corporate and investment banking on a global basis. Because of its breadth, Scotiabank is able to offer a full suite of financial products to its clients.

Scotiabank has made a deliberate effort to become a customer- focused institution, as opposed to a vertical product driven company. The bank’s formally stated goal is “to be the best at helping customers become financially better off by providing relevant solutions to their unique needs”. A direct consequence of this goal is that marketing campaigns are multiple product campaigns as opposed to single product campaigns. This transforms the data mining and campaign targeting process from a fairly simple application of individual response models into a significantly more complex problem of choosing which product, if any, to offer to which customer and through which channel. The benefit is that campaigns are more customer-focused than in the past.

1.1    Business Problem

The database marketing community has changed significantly over the last several years. In the past, database marketers applied business rules to target customers directly. Examples include; targeting customers solely on their product gaps or on marketers’ business intuition. Marketers have also applied RFM type analysis where general recency, frequency, and monetary measurements as well as product gaps are used to target customers for specific offers. The current approach, which has widespread use, relies on predictive response models to target customers for offers. These models accurately estimate the probability that a customer will respond to a specific offer and can significantly increase the response rate to a product offering. However, simply knowing a customer’s probability of responding to a particular offer is not enough when a company has several products to promote and other business constraints   to consider in its marketing planning.

Marketing departments also face the problem of knowing which product to offer to a customer, not just which customer to offer a product. In practice, many ad hoc rules are used. Prioritization rules based on response rates or estimated expected profitability measures have been used; business rules to prioritize products that can be marketed are sometimes used; and product response models to select customers for a particular campaign are also used. One approach that is easily implemented but, for reasons outlined later, may not produce optimal customer contact plans relies on a measure of expected offer profitability (the estimated probability of response multiplied by the profit given customer response less direct costs) to choose which products to offer customers. However, a shortcoming of this approach is its inability to effectively handle complex constraints on the customer contact plan.

1.1    Business Constraints

Database marketing departments face several types of business constraints. Typically, there are

• restrictions on the minimum and maximum number of product offers that can be made in a campaign,
• requirements   on   minimum   expected   profit   from product offers,
• limits on channel capacity,
• limits on funding available for the campaign, and
• campaign return-on-investment hurdle rates that must be met.

These are a sample of the constraints that marketing departments must meet when executing a campaign. Ad hoc approaches are also typically used in an attempt to meet these constraints.

The opportunity costs of the business constraints are generally not known. Constraints are usually negotiated between marketing, product lines and delivery channel management. If the cost of a constraint was known, then the company could choose to tighten or to relax the constraint by removing or adding more resources. For example, channel capacity could be increased if it were known that there was a significant return on the investment by doing so. Knowledge of the opportunity costs could help evaluate these management decisions. Applications of this will be discussed in the “STRATEGIC USAGES” section of this article.

Ultimately, the database marketer needs a concrete framework to effectively act on “the right product to the right customer at the right time” mantra. The approach we take is to transform the database-marketing problem into an optimization problem that is designed to generate the maximum incremental profit from a limited amount of resources subject to the necessary business constraints. This paper will describe an actionable framework that will satisfy this business problem.

2.    SOLUTION FRAMEWORK

It is helpful in understanding the solution framework to understand the data that are available for marketing campaign planning. Understanding the data will help make the problem more concrete.

2.1    Data

We assume that there has been a thorough analysis of historical marketing campaigns and that accurate response probability models exist for all products in question. The result of these data mining exercises is a data set that contains an expected profit for each product for each customer, where the expected profit is derived from the customer specific probability of response and the profit generated given a customer response. Needless to say, these data sets can be rather large. It is not unusual to have over 5 million customer records in such a data set. Let’s assume that there are 10 products and 1 million customers, and that for each customer and product we have an estimate of the expected profit given that each customer is offered each product.

2.2    Ideal Approach

The ideal approach to solving this problem is to model it as a specialized type of assignment problem. This type of problem is an integer program. It can be unambiguously expressed with a mathematical formulation. Let xij = 1 if customer i is offered product j, and 0 if not; let rij the expected profit of offering customer i product j; let cij the cost of offering customer i product j; let R be the corporate hurdle rate. Then, a very simplified version of the problem can be expressed as finding the xij that satisfy

This formulation captures only the bare elements of the problem. It does not account for multiple campaigns composed of different products, multiple channels, and channel capacity constraints just to name a few possibilities. However, the model can easily be extended to cover most typical business constraints encountered in practice, but the basic formulation remains the same. It is important to note that this ideal formulation is difficult to solve because of its scale. For 1 million customers and 10 products there are 10-million integer variables xij, this yields 210,000,000 possible customer-offer combinations. Using standard optimization methods a problem of this size can, in principle, result in a branch and cut tree of as many nodes. Because of this problems of this size are extremely difficult to solve, so we propose an alternative approach. While not providing a strictly optimal solution, the alternative approach does provide an approximately optimal solution that in preliminary studies has shown to be a good approximation.

2.3    Practical Approach

Although it is not practical to solve problems formulated in this ideal way, it is possible to approximate the ideal formulation and arrive at a formulation that is practical to solve. There are numerous ways to approach this approximation; one approach is to sample from the customer base and use that sample as representative for the optimization. Another approach (and the one that we take) is to aggregate customers based on coefficients rij in the ideal formulation. Aggregation can be considered natural in this setting particularly when we understand that much of the data is consistent and estimated. For example, the cost data cij are most likely to be consistent across customers for a given product. Similarly, the estimated expected profit rij is most likely the result of data mining techniques such as predictive response models. As long as the customer/offer specific response models. As long as the customer/offer specific response rate is represented as a probability, the proposed framework can handle it. Scotiabank uses standard, accepted statistical and data mining approaches to obtain these estimates.

The aggregation process we use involves conversion of the raw data into a form that can be used naturally in a linear programming optimization model. The key is to cluster the raw data rij and use the clusters as the aggregate. Unlike the usual use of clustering, the purpose here is not the identification of customer segments or to differentiate groups of customers, but to aggregate customers into similar groups. This is an important distinction to keep in mind since clustering is most frequently used to distinguish, not to aggregate. If the clusters are internally consistent, then the cluster centroids can be used as representative of the data for all the customers within a single cluster.

This aggregation enables the problem to be reformulated as a linear program so that rather than assigning offers to individual customers, as the ideal integer program does, the program identifies proportions within each cluster for each product offer. This can be accomplished with similar constraints to those of the ideal formulation. Moreover, the linear program is much smaller and much easier to solve. Note however, the solution may require that multiple products are offered to proportions of customers within a single cluster. When that happens, a new problem is defined that is a simple assignment problem at the level of the cluster, where multiple offers are to be assigned within the cluster, and it is relatively easy to solve.

It should also be noted that preliminary experiments solving the approximation with varying numbers clusters indicate that as the number of clusters increases the value of the objective quickly rises then slowly converges to the integer relaxed solution. Further study is needed to identify a good number of clusters that work in a “typical” setting.

2.3.1          Formulation

Consider the following variables defining raw data as input into the solution algorithm. Let yij be the number of customers in cluster i that are offered product j; let r’ij be the estimated expected profit given that customer in cluster i is offered of product j; let c’ij be the cost of offering a customer in cluster i product j; let R be the corporate hurdle rate. Then, a very simplified version of the problem can be expressed as finding the yij that satisfy

Once the yij that satisfy the formulation are found, the optimal proportions that they give must be applied to the customers within the specific clusters. For example, suppose that yij is the total number of customers in cluster i. Then, every customer in that cluster should be offered product j. Alternatively, suppose that for a given i, yij > 0 and yij’ > 0 for j ! j’. Then, yij of customers in cluster i must be offered product j and yij’ of customers in cluster i must be offered product j’. The optimal way to do that is to solve a simple assignment problem using the estimated expected profit rij for the individual customers and not the clusters. It is important to note that some of the constraints may be violated as a result of solving this assignment problem particularly if the cluster centroids used in the linear program formulation are involved in a tight constraint and not consistent within the cluster.

3.    A TACTICAL EXAMPLE

We demonstrate this approach with data from Scotiabank and using existing procedures within the SAS system to implement the formulation described above. The details of the SAS code will not be given.

Eleven unique offers were to be considered: five investment, three lending and three day-to-day banking offers. The investment offers included GICs, mutual funds, Registered Education Savings Program (RESP) and two unique discount brokerage offers. The lending offers included a mortgage and two credit card offers. The day-to-day banking offers included one of two Scotia online banking service offers and a deposit account acquisition. The term campaign is used here to imply one large pro-active customer contact campaign that it comprised of eleven distinct offers, it can be thought of as eleven single product campaigns that are being offered at generally the same time to a non-overlapping set of customers. For the purposes of this paper the detailed product offer descriptions have been suppressed. Approximately 2.5 million customers were included in the potential universe for the campaign.

Ultimately, the goal of marketing campaigns is to produce a positive return on investment for the company that exceeds the corporate investment hurdle rate. Although the timeframe upon which this investment should be measured may be debatable, the goal is fundamental to the bank. To achieve this specific objective, the bank can execute marketing campaigns that are designed to maximize the expected incremental profit through making one of several offers to some of its customers, or potential customers.

3.1    Response Models

The expected incremental profit of a specific offer to a customer is an estimate based on response models and detailed product profitability calculations. These response models are used to estimate the probability that a customer will accept a specific offer. Scotiabank’s data warehouse has detailed account level profitability calculations for all of its products. This profitability information is used to estimate the near term incremental profit given that the customer accepts the specific offer. Once a specific offer is made to a customer there are two possible outcomes: the customer can accept or reject the offer. Using the offer specific response models the probability of both states is known for each customer. The incremental profit for both states is also known; it is zero if the customer rejects the offer and the mean near-term profitability for new accounts of the specific type if the offer is accepted. With this information, the expected incremental profit of the offer can be calculated for each customer/offer combination. The cost of making each offer is also known and is largely dependent on the channel through which the offer is made.

3.2    Channels

Scotiabank has several distribution channels through which campaigns can be executed. The main channels for direct marketing are direct mail, retail branch centres and call centres. For this example we assume that leads sent to the branch officers and call centres are follow-ups from a direct mail piece and that offers designated as direct mail are direct mail only. The use of the branch and call centres for follow-ups has been shown to have a positive effect on the probability of response to the offer when compared to direct mail alone. Of course, the lead delivery costs vary with the channel used. In this example we have used costs per lead of $3.00, $1.50 and $1.00 for the branch, call centre and direct mail only channels respectively.

3.3    Business Constraints

Several practical issues surround the campaign execution process that affects the customer/offer selection process, for this application to be acceptable for implementation these business constraints must be maintained. The following business rules have been translated into constraints that can be applied to the optimization model:

• Campaign costs cannot exceed $1 million.
• The campaign must have a return on investment of at least the corporate hurdle rate. In this example we have used 20%, which is not necessarily the bank’s actual corporate hurdle rate.
• The branch and call centre channels have a certain capacity constraint for timely processing of campaign generated leads. In the example, the call centre can accept up to 500,000 leads, the branch can handle up to 250,000 leads and direct mail is unlimited.
• Product offer minimums are also required to satisfy internal bank objectives. For the purposes of this example we set all offer minimums to 20,000 with two exceptions. The RESP offer, which has an extremely limited eligible universe, had a lower bound of 2,500 and one of the Scotia online offers had a lower bound of 5,000.
• Cannot offer products to customers who already have that product at Scotiabank.
• The standard marketing exclusions, such as credit risk or do not solicit, must also be strictly adhered to.

3.4   Solution

The estimates for customer/offer expected incremental profit, costs and business constraints serve as inputs to the profit optimization phase of the campaign design. The profit optimization phase is independent of the construction of these inputs. This means that as response models, profit estimates or costs are refined as long as the results are represented in the same manner, the optimization phase will be able to accept them as inputs. This property is important as the bank is constantly testing and refining these inputs as the marketplace is ever changing.

3.4.1       Results

The result of this algorithm is an allocation, of a specific offer, or no offer, to each customer. Also output is the associated expected incremental profit by customer making that offer. This solution is a SAS data set that has a customer identifier, the expected return, offer and channel designation. The full data set is 2.5 million records; the table below shows the first 25 records.

Customer Id	Product	Channel	Return
00182723			.
00200688			.
00032937			.
00722119			.
02137391			.
00992639			.
00060721			.
00483601	Offer 2	Direct Mail	0.0005
01164964			.
00025469			.
01008244			.
00179891			.
00410488	Offer 10	Direct Mail	3.1852
01484008			.
00184804			.
03350118			.
00983111			.
00387834	Offer 5	Call Center	13.0782
01100914			.
01507075			.
01559899			.
00309931			.
00657640			.
01095694			.
02075404			.

Figure 1. Sample of the solution dataset.

To better understand the solution, it is useful to look at several charts that summarize the solution and a report that is produced by the algorithm.

3.4.2         Offers

The Offer Frequencies by Channel chart, Figure 2, provides a graphical representation of the distribution of offers by channel and is useful for understanding the solution. This figure shows that few of the contacts have a branch follow-up treatment, some have call centre follow-ups and most have just the direct mail treatment. Offer 10 has the largest quantity of contacts and is spread across the direct mail and call centre channels. The Constraint Report, Figure 3, provides significantly more insight into the nature of the derived solution.

 

 

 

 

 

 

 

 

 

 

 

3.4.3         Business Constraints

The Constraint Report summarizes the constraints applied to the problem as well as outlines the chosen level and marginal costs associated with each constraint.

 

 

 

 

 

 

 

 

 

 

 

•  The last line of the Constraint Report in Figure 3 shows that the objective function, expected profit, was maximized at $3.58 million from an expenditure of $1 million. This results in a 258% return on investment for the campaign.
•  The first two lines in Figure 3, Branch Capacity and Call Center Capacity, summarize the results of the branch and call centre capacity constraints respectively. The branch capacity for follow-up contacts was limited to 250,000 and the call centre to 500,000. In the solution, only 2,180 contacts were assigned to the branch, and 202,258 to the call centre, for follow-up calls. This low quantity of follow-up contacts, at either the branch or the call centre, is due to the conservative estimate of the increased response rates resulting from the follow-up and the significantly higher cost, $3.00 for branch and $1.50 for call centre, as compared to direct mail only, $1.00.
• The third constraint, Campaign Cost, limits the total costs   for the campaign to $1 million. This constraint is in fact tight, meaning that the optimal solution was restricted by the condition.   The marginal value of the constraint is $1.53, this means that an additional $1 spent on the campaign would result in a $1.53 increase in expected profit.
• The offer constraints show the lower bounds for each of the specific offers. Notice that 5 of the 11 products were limited by their lower bounds. In the solution, Offer 9 was only made to 20,000 people. If that constraint were to be decreased by one unit, e.g. to 19,999, then the objective function, expected profit, would be increased by $1.48 – so the cost, on expected profit, of this business constraint is clear.

3.4.4         Profitability

The estimated Profitability by Channel report, Figure 4, clearly reveals the quality of leads that are sent to the respective channels. The branch follow-up leads have a significantly higher expected incremental profit than for call centre follow-up leads or direct mail alone. The call centre is also sent leads that are more profitable than direct mail alone. A few comments about the way that the channel effects were modeled are necessary to fully understand this result.

 

 

 

 

 

 

 

 

 

 

 

 

The differential effects of the various channels enter the response models as a main effect. A call centre follow-up treatment increases the probability of response as compared to no call centre follow-up treatment. The branch follow-up treatment has the same directional effect as the call centre, although larger. As such, everything else the same, the expected profit from making an identical offer to a customer with a branch or call centre follow-up is greater than without the follow-up. As such we would expect to see a higher expected rate of return when applying the additional follow-up treatments, although not at this magnitude. There is some other factor driving the higher rate of return, in fact, it is the channel selection by the optimization routine.

Recall that the channel costs are fixed at $1.00, $1.50 and $3.00 for direct mail only, call centre follow-up and branch follow-up respectively. The incremental expected profit enters the equation through an increase in response rate. For a branch follow-up to be more profitable than an offer made without a branch follow-up it must be the case that:

Prob Branch – Profit – Cost Branch + Prob DM – Profit – Cost DM

. Profit – (Prob Branch , Prob DM ) + Cost Branch , Cost DM

. Profit – (Prob Branch , Prob DM ) + 3 – 1

. Profit – (Prob Branch , Prob DM ) + 2

Either the profit given that the customer accepts (Profit) or the incremental effect on the response probability of the Branch follow-up (ProbBranch – ProbDM) is large enough to overcome the $2 increase in contact costs (CostBranch – CostDM). The greater the difference in this inequality the more beneficial the follow-up contact is. The difference can be large if a highly profitable product is being offered, or the increase in the probability of response is high. For reasonable values of the probability of response from the direct mail (ProbDM), the increase in response rate due to a branch follow-up (ProbBranch – ProbDM) will rise with an increase in the probability of response from the direct mail (ProbDM). From a business perspective, this means that either highly profitable offers and/or customers who are most likely to respond to the offer will tend to be given a follow-up treatment. This is a particularly important result when dealing with the business owners of the call centre and branch channels.

3.5    Summary of Tactical Example

In summary, the solution provided an approximately optimal solution to the ideal capacitated assignment problem. The output is a decision, for each customer as to which, if any, product to offer and through what channel. The campaign expected profit is $3.58 million on $1 million invested, for a return on investment of 258%. Using an ad hoc approach, that utilized response models and near term profit, and met all the business constraints the most profit that could be generated was $2.65 million on $1 million invested for a campaign return on investment of 165%. The boost in the campaign return on investment of 93% is entirely attributable to the quality of the solution produced by the optimization process as opposed to the ad hoc approach.

4.    STRATEGIC USES

Although this technique was developed primarily for its tactical application, as described above, it has some significant strategic applications too. The strategic applications are in the area of capacity planning. Two insights will be discussed, one dealing with campaign budgeting and the other with channel capacity planning.

4.1    Campaign Budget Allocation

In general, campaign budgets are determined prior to the campaign design. The degree of analysis that goes into determining specific campaign budgets, or annual campaign budgets, can vary greatly from institution to institution. As a strategic tool, the optimization technique provides an opportunity to determine the effects of making different budget allocations – in the budgeting process.

For example, in determining how much money to invest in a campaign, it wold be useful to know the marginal return on an additional dollar investment. This is reported at $1.53 in Figure 3 as the marginal value of the $1 million cost constraint. Given the campaign definition, all of the other constraints and the current customer base, investing one more dollar would result in an increase in profit of $1.53. If marginal return on investment is greater than the corporate hurdle rate then that supports an argument to increase the investment.

This technique provides a compelling and empirically based process for altering, positively or negatively, the campaign budget. Of course, other considerations go into budget allocations but the technique could shed light onto the impact of such decisions.

4.2    Channel Capacity Planning

Similarly, channel capacity planning can benefit from this technique. It is understood that in the short run, channel capacity is fixed, although in the long run, or planning mode, channel capacity can be changed.

Again, from the Constraint Report, Figure 3, if it appears as though a specific channel is used to capacity, then we can look at the marginal value of the constraint. The marginal value of these constraints gives the increase in profit, the objective function, given a one-unit increase in channel capacity. With the cost of this increase in capacity quantified, one can determine if the additional investment in the channel is warranted. This also helps to quantify the opportunity costs of having branch staff shift away from non-campaign related work. Again, this is from the perspective of campaign execution, there are other benefits to channel capacity augmentation that would also have to be taken into account, but at least the campaign benefits would be understood.

4.3    Sensitivity Analysis

In addition to using the solution for capacity planning the approach lends itself to analyzing the sensitivity of the solution to changing assumptions. This is particularly useful for evaluating the solution to changes in the business constraints.

For example, a scenario can be optimized with several alternative budget constraints. Figure 5 shows the results from solving a model that is very similar to that discussed above. Each row in the table corresponds to a different limit on the total cost of the campaign. All the other constraints and data in the model are held constant. The first scenario allocates at most $1,000,000 on the campaign, the second $1,250,000 and the third $1,500,000. For each of these scenarios the optimal solution is calculated and the expected return from that solution, the number of offers through the branch, and the ROI, are also shown in the figure.

 

 

 

 

 

 

 

It is interesting to note that for each of these scenarios, as for the example above, the branch capacity is 250,000 offers. However, until the size of the campaign, as measured by the cost of the campaign, reaches $1,500,000, the branch channel is not used to

its capacity. This means that for campaigns with a budget less than $1,500,000 there is excess capacity in this channel and its value is lost. The additional $500,000 expenditure results in an additional $670,000 return with a 34% ROI and no excess branch capacity.

5.    SHORTCOMINGS

The shortcomings of this approach can be broadly classified into two categories: business and technical. From a business perspective there are three perceived shortcomings of this solution. These shortcomings are related to the required inputs to the solution, changes in the types of campaign design decisions, and post analysis, which are performed by the business user. From a technical perspective there are two general shortcomings of this approach. The first is related to the constraints, and the second is related to the acceptable problem size.

5.1    Business Perspective Shortcomings

The approach requires as an input the expected incremental profit associated with each offer and customer combination. Fundamentally this implies the creation, and maintenance, of offer specific probability of response models and detailed profitability measurements. Both of these inputs require ongoing maintenance and enhancement, as the marketplace is ever changing. This requirement does not seem to be too onerous as most organizations that would consider implementing this solution are likely already are producing these inputs.

The types of decisions made by business users will be altered with the adoption of this approach. The business users will be making decisions that affect the objective function and constraints but not directly about how many offers to make of each type or through which distribution channel. At first the business users might resist this solution, as the decisions that are being made by the business are more abstract than those made during the business as usual campaign design process. Although the business users are making more abstract decisions about campaign design, they are actually gaining more effective control over the campaign. Thus the successful adoption of this process would likely involve some amount of business user education with respect to the design of the optimization problem.

The solution was designed to explicitly maximize the expected incremental profitability from running a campaign, not to maximize the overall response rate of the campaign. For instance, a customer might have a higher estimated response probability to a low rate credit card offer than a high rate card. However, the profitability given offer acceptance could be substantially higher for the high rate card. If the goal were to maximize the response rate, then the low rate card would be offered. Whereas, if the goal were to maximize profitability then the offer with the greatest expected profitability would be made. This is only a shortcoming in the sense that business users’ expectations have to be managed as campaign response rates are more easily and quickly measured than is campaign profitability.

5.2    Technical Perspective Shortcomings

As this solution was developed in a linear programming framework the constraints must be expressed as linear functions of the choice variables.   A couple of examples of plausible business constraints that are non-linear and therefore do not work with the linear programming approach are:

• The number of credit card offers must be greater than 20,000 or equal to zero.
• The cost of a specific offer is a function of the number of those specific offers being made.

The scalability of this approach has not been exhaustively explored. Tests have been run on data sets with 2.5 million customers, 11 offers and 3 distribution channels and the solution is generated in an acceptable amount of time and resources. The solution was explicitly designed to scale well to the number of customers. Scalability in the offer and channel dimensions is significantly more expensive than along the customer dimension. Although, the number of distribution channels that a company can utilize in an automated campaign is not too large; the number of possible products that could be offered could grow well beyond eleven.

6.    CONCLUSION

This offer optimization approach provides three significant improvements over other, more standard, approaches to the problem of campaign design.

• First and foremost, the developed solution produces significantly more incremental profit than competing solutions. As demonstrated in the tactical example, the campaign incremental profit is almost twice as high as that of the standard approach.
• Secondly, this technique is designed to implement multiple constraints and therefore affords the business more control over the direct marketing process. Attempting to satisfy several business constraints simultaneously using ad hoc techniques is a very labor-intensive task and generally produces poor results.
• Finally, the additional information that can be presented as a part of this solution can provide the business with more insight into the customer base, product offerings and the effects of the constraints.

This insight can be used to guide the company to craft better investment decisions in order to make future campaigns even more successful.

7.    ACKNOWLEDGMENTS

Our thanks to the three anonymous referees whose suggestions helped clarify our discussion.

8.    REFERENCES

Campell, D., Erdahl, R., Johnson, D., Bibelnieks, E., Haydock, M., Bullock, M., and Crowder, H. (2001), “Optimizing Customer Mail Streams at Fingerhut,” Interfaces, (31), 77-90.

Cohen, M. (2000), “Offer Optimization, Optimizing customer value,” SAS Internal Technical Paper. May 2001.

Cohen, M., and Parks, J. (2000), “Optimizing the Allocation of Cross-Selling Effort,” Proceedings of the DiamondSug 2000 Conference, San Francisco.

Cohen, M. (2000), “Targeted Marketing, Optimizing the customer/event/channel assignment,” SAS Internal Technical Paper. December 2000.

Nemhauser, G., and Wolsey, L. Integer and Combinatorial Optimization, John Wiley & Sons, New York, 1988.

Abstract The year 2000 saw an explosion in the volume of short text messages being   sent to mobile phones. Originally the sole realm of the telecommunication providers, this communication medium is starting to be used by other types of organisation to deliver messages to consumers. Marketers are starting to recognise the potential of this medium for marketing communications. This paper explains how the technology works   and explores potential business applications in marketing.

INTRODUCTION

To date much of the discussion around the wireless Internet has concentrated on the application of Wireless Application Protocol (WAP) technologies. But a survey by Forrester Research1 showed that many WAP sites have a long way to go in meeting the requirements of the consumer for reliability, access and navigation. Short message services (SMS) available on GSM mobile phones on the other hand have grown from strength to strength. Unlike WAP, SMS can be used as a two-way communication vehicle, allowing people or organisations to send and receive short text messages from a mobile phone in near real time. The ready uptake by consumers of SMS will ensure that it becomes an integral part of the marketing mix.

THE TECHNOLOGY BEHIND SMS

The SMS service is actually a network of SMS centres that are connected to each other and can interchange text-based messages. The SMS centres are specially written software packages that can:

• send messages to mobile phones
• receive messages from mobile phones
• receive messages from the Internet
• send messages to other SMS centres
• receive messages from other SMS centres.

SMS centres tend to be owned by telecommunication companies that want to offer SMS services to their customers. The software for the SMS centres is designed and developed by specialist IT companies including Logica, CMG,

 

 

 

 

 

 

 

 

 

 

 

Nokia and Sema. There is no single standard protocol for messaging systems and each of the SMS centre solution providers uses its own protocol.

The SMS business is a key part of their profitability. In fact, for some of these companies the majority of their profit growth came from their SMS divisions.

THE GROWTH IN SMS

The volume numbers for SMS services exploded in 2000. Upwards of 10 billion messages a month are sent globally using the SMS service (see Figure 1).

The key reason for the rapid growth lies in the fact that only recently has the service been truly global. Before 1998, most services were limited to the mobile operator network only. Mobile users could only send messages to other users on the same network. Clearly, that is a very limited service. Mobile operators at that time began to negotiate peering agreements that would permit the SMS centres to send messages to each other when the users were on different networks. Once these agreements were in place, the service multiplied its reach dramatically.

In addition, text messaging was associated with children and students, mainly because the cost of messaging was much cheaper than the cost of voice calls. As more people began to get messages though, especially messages alerting them to new voicemail, they began to understand the non-intrusive, but highly pertinent and time-sensitive information that could be received (and also stored in memory) using SMS.

As people became more comfortable with the service its use soared. Simple applications were deployed, including e-mail notification, information services (lottery numbers, sports scores, etc.), and other time-sensitive applications. Most of these new SMS services were deployed by the telecom operators themselves but the situation is changing as other organisations recognise the potential of this new communication vehicle.

PRICING OF SMS SERVICES

The mobile telecommunication operators typically charge from 2p to £1 for each message sent, with 10p being the average charge. This cost structure may prohibit some organisations from using SMS for marketing purposes. The pricing model will change as other organisations recognise the revenue potential of SMS and move into the market.

DELIVERING SMS SERVICES

The ability of SMS centres to send and receive messages from the Internet means that creating the necessary technical infrastructure to deliver SMS is within the capabilities of most technology-orientated organisations.

There are three key options that are being explored by the early adopters:

• developing static applications that talk to one SMS centre
• sending messages to a service provider that is connected to various SMS centres
• installing a messaging application that talks to all of the SMS

There are a number of key factors that influence the choice of delivery method for the early adopters. These include:

• up-front cost
• ongoing operating costs
• flexibility of platform
• availability of internal IT resource
• quality of service desired
• volumes of messages to be sent/received.

Most have tended to opt for the   service provider option and have then migrated to a messaging platform or gateway once the SMS has proved its worth.

KEY FEATURES OF SMS

Mid-1999 saw the mobile operators beginning to offer simple services over the phone. These services included:

• welcome messages
• e-mail notifications
• voice message notifications
• sports results
• lottery

These services were relatively straightforward and lent themselves naturally to SMS. The ability to connect directly to SMS centres via the Internet has allowed a much more extensive range of services to be developed. These new services exploit some of the beneficial features of SMS. These include their:

• personal nature: the fact that people carry a mobile phone as a personal communication device means that short text messages can be sent directly to an individual at almost any time
• near real-time delivery; a short text message can be sent to a mobile phone almost in real time, if the phone is switched on the individual will immediately be informed about the receipt of the message
• unobtrusive nature: text messages can be read when recipients want and do not therefore interrupt their current activity
• relative low cost: depending on the service provider, the cost of sending an SMS communication (2p–£1) to an individual is much lower than for more traditional media such as direct mail (25–75p) or telephone marketing (£1.00–£2.50)
• simplicity: receiving and sending short text messages from a mobile phone are simple operations that do not require additional software or specialist hardware
• support for two-way communications: unlike WAP, SMS allow messages to be sent both ways
• message forwarding (ability to forward messages to peers): most SMS services allow messages to be forwarded to other individuals and, in some cases, groups of individuals
• location-based potential: the mobile technology allows the location of the phone user to be identified, opening up the potential for location-based

The key to success will be in developing SMS that are timely, relevant and pertinent.

PERMISSION-BASED MARKETING

As with e-mail, the potential for unwanted messages or SPAM is great. This will become particularly true as SMS become more widely available to third parties. The fact that most customers pay directly to retrieve text messages means that the potential impact will be even more pronounced. There is   a strong move by telecommunication companies to move to a self-regulated permission-based approach to SMS.

The sanctity of the mobile phone as a personal communication tool could easily be violated by inappropriate communications. Filtering technologies are already being developed to address this issue.

BUSINESS APPLICATIONS FOR SMS

There are six main types of services (as defined by the eigroup), but that does not mean that there are not many   more variations on the theme of timely, relevant information delivered directly to the user.2 The six ‘SMS’ types are:

• Send Me Stories
• Save Me Somehow
• Search My Server
• Sell Me Something
• Sort My Socialising
• Send Me

Send Me Stories provide a marketing message to a mobile phone user that contains information that is relevant and time sensitive, for example sending details of an in-store promotion to a loyalty card customer.

Save Me Somehow provides a marketing message to a mobile phone user that acts as coupons allowing discounts on specified goods. For example, sending details of a discount on a television to a store card customer. Or   a marketing message to a mobile phone user that reminds them about an important event. For example, confirming that a large transaction has been lodged in a bank account and recommending most appropriate action.

Search My Server provides messages to customers with the objective of stimulating access to a WAP site. For example, stimulating usage of a financial information WAP site after usage has declined or service has been revamped.

Sell Me Something provides messages to customers selling a product or service and allowing them to purchase items directly through text response. For example, sending details of a new CD released by a customer’s favourite band, using previous purchase or customer preference data and allowing the customer to send text message as response and   initiate purchase process. As location tracking becomes widely available, the combination of time and location will prove very compelling in offering products or services to the consumer. Bell Mobility in the USA are currently   exploring location-specific digital couponing, offering discounted products and services to subscribers within a certain radius of participating merchants. The commercial roll-out is expected in 2001. The technology will allow identification of location to within 50   feet.3

Sort My Socialising provides messages that can be forwarded to a customer’s peer group. For example, sending a customer details of a concert or webcast in a form that can be forwarded to friends and including a response mechanism that allows peers to register on WAP or website. Ensuring compatibility with the Internet messaging facilities (e-mail and ICQ), is essential if a full service offering is to be provided.

Send me Signals provides a marketing message to customers that signals an action is required. For example, a local motor dealer sends a message to a customer warning that the customer’s vehicle is due a service, suggesting possible availability.

LOCATION-BASED SERVICES

The technology associated with modern cellular communications allows the location of the mobile device user to be identified to within 50 feet. This has led to the development of a number of location-based services. Although there are a number of data protection and privacy issues that will need to be addressed, a number of organisations in Europe are running pilot projects where the information is being used to drive marketing communications. If this technology becomes more widespread the application of location-based data may become the norm in some sectors, for example the automotive sector.

THE IMPACT OF BLUETOOTH

Bluetooth (a technology named after a 10th century Viking king) is being developed by Ericsson, Nokia, IBM, Toshiba and Intel to allow wireless communication between electronic devices. This technology, which will not really be available until 2002, should increase the power of SMS by facilitating more integrated communication between the mobile phone and other devices. In particular it could facilitate the payment for goods and services using the mobile phone.

CONCLUSION

SMS will grow in popularity as a marketing communication vehicle over the next few years. Its potential is only just being realised within the telecommunications industry and is still relatively unknown outside the sector. But with the creation of a global infrastructure and subsequent explosion in the use of SMS by consumers, it will not be long before other parties start to explore the opportunities that it avails.

The current pricing model used by the telecommunication companies will inhibit SMS use for marketing in the short term. But the revenue potential of the SMS market will attract other providers who will increase the competition and bring down prices. The unique qualities of SMS will make it a powerful weapon in the modern marketing mix.

Abstract The paper explores the advantages and disadvantages of rule-based engines and optimisation techniques in supporting marketing communication decisioning.

INTRODUCTION

In the past, determining which marketing communication a customer received was easy as the choice of channel was very limited. With the growth in active (eg e-mail) and passive (Web) communication channels and the increase in the volumes of campaigns facilitated by marketing automation technologies, however, this simple choice is becoming more difficult to make.

To address this complexity two new classes of technology are emerging — optimisation and rules engines. This paper explores some of the advantages and disadvantages of rules-based engines.

RULES-BASED DECISIONING

This type of technology breaks the communication decision into one or more simple rules that when applied determine the most appropriate marketing communication(s) for a customer. It does not use any specific statistical process, although one or more of the variables used by the rule(s) may   be behavioural model scores.

DEFINITIONS

The following section describes rule-based engines in more detail:

A rule is the smallest unit of business logic. It generally captures a test condition and identifies the action to   take, based on the test evaluation, eg if a customer is male then send communication.

A rule set is a collection of rules, generally grouped together because they relate to a common task, eg if a customer is male and aged between 20 and 30 years then send communication.

A rule flow shows the order in which a particular rule set or series of rule sets are executed. These rule flows are normally represented graphically. The order in which rules or sets of rules are applied can significantly affect the outcome. For example, if gender is male, age is 25 to 30 years, marital status   equals married (rule set 1) then if   product code equals loan, and loan status equals open (rule set 2) send standard cross-sell communication.

Many of the new rule-engine technologies are based on object-orientated programming techniques, in which case the concepts of classes and objects are used to represent various business entities and ease rule definition. They also allow the rule developer to connect the rules with some outside calling application and pass information back and forth.

Variables are used to maintain and pass information within the rule, eg gender.

Enumerations allow the rule developer to define a limited set of values for some object. In some cases, there may be some unit of business logic that involves executing a series of steps in a   rigid, pre-defined order. While this could be implemented as a rule, the function   entity is designed to capture this form of business logic.

BUSINESS APPLICATION

The following section explores some of the business applications of rules-based engines in a customer relationship management   (CRM) context.

Perhaps the most common use of business rules-based engines is the processing of applications for products and services, where there are a large number of complex business rules that have to be applied and where these rules change frequently. For example, processing complex lending rules as part of personal loan application.

For delivery of context-specific information the business rules are used to determine when context-specific information is to be provided to a client or employee as part of a standard process. For example, providing context-specific data in a telemarketing script.

In order to provide personalised content on websites the business rules are used to determine the specific content to   be presented to a customer as part of an interaction on the website. This personalisation could be based on the customer’s segment, attributes profile or behaviour on the site.

For supported selling the business rules are used in combination with product purchase propensity score to determine the ‘next best offer’ as part of a supported selling process at the point of sale.

In the case of lead allocation in   channels the business rules are used to determine how a sales lead is to be allocated to a specific channel or sales agent within the channel. This approach facilitates a consistent treatment of a sales opportunity while ensuring effective use of channel resource. For example, allocation of mortgage sales leads between a retail branch, direct sales force and third-party agent where the individual is not account managed.

Rules-based engines work best when the:

• process is well defined, complex, uses a large number of rules or rule sets and is executed in real time
• business policies or procedures change frequently
• business users want to control changes in the business rules

BENEFITS OF RULES ENGINES

Modern business rules-based engines have the following benefits, they:

• allow non-IT users to create, maintain and monitor the impact of business rules on processes
• allow organisations to automate and streamline processes
• allow organisations to ensure consistency of process
• allow organisations to increase the efficiency of the processes
• provide a clear audit trail of the decisioning processes.

ISSUES WITH RULES ENGINES

The following section explores some of the problems associated with rules engines.

Performance

Where the business process requires the application of a large number of rules or rule sets, performance of the decisioning process can be significantly affected. This is particularly true where the rules engine   is being used in real time and has to access data from an associated data repository (eg operational data store) rather than from the customer interaction (eg Web page interaction).

Batch processing

Most of the recent rules engine technologies have been developed with real-time application of the rules in mind: one of the consequences is that they have been poorly designed for large volume batch processing.

Perhaps one of the obvious potential applications of rules engines was as a core component of campaign management technology, but their success to date has been limited.

Understanding implications of change

Most business rules engines allow the user to change individual rules with ease but it is difficult to assess the impact of these changes. A simple change in one rule or rule flow can radically change the outcome. This can cause significant problems where the rule engines are being used to drive real-time core business applications, eg instant   credit processing.

In some cases the ability to test the impact of changes in the rules, rule sets or rule flows has been achieved by allowing the user to change the rules and to process the same series of events or data. The user can then identify the ‘ideal’ set of rules and rule parameters.

Effective change control and/or authorisation procedures are essential if the business is not to be adversely affected by any changes.

Integration with other business systems

For rules engines to be effectively deployed they often need to be integrated with other operational systems, eg call centre or branch teller systems. In some cases this can be quite difficult, although this issue has been recognised and the newer technologies have addressed the integration issue well. A range of different deployment models is becoming available.

Many rule engines can be run as a service (ie using threads provided by the container application in which it resides) or as a server (ie where it needs to use its own threads since there is no container application). They can be deployed in most environments that support Java including: EJB; J2EE; LDAP,   MTS; RMI; CORBA; JSP,   ASP and MQS series.

A few vendors supply standard deployment APIs that handle the interaction between the rule engine server and the calling application, eg call centre application.

THE USE OF RULES ENGINES FOR OFFER OPTIMISATION

A number of vendors have positioned rules engines as optimisation technologies for managing customer communications, following in the wake of statistical optimisation products like those from niche player Market Switch and mainstream players SAS and SPSS.

With the wider availability of channels that can be used to communicate with and receive communication from customers, has come a significant increase in the complexity of the communication-decision process. The key thing that the optimisation technologies are trying to do is to maximise mathematically the profit generated from marketing and other communications, removing the guesswork from the campaign management process. The problem is that with dozens of potential offers, many communication channels and thousands or millions of potential customers, there could be billions of different offer–customer–channel combinations to choose from.

Optimisation technologies use mathematical algorithms to determine the best combinations of customers, products and communication channels to maximise profit, while at the same time taking into account real-world business constraints such as limited budgets or channel capacity. The output of the optimisation process can then be used to drive communication with a customer. Analytical and modelling tools can be used to understand the customer and build behavioural models. For a given marketing offer, these models can be used to determine a customer’s likelihood of responding to that offer.

The use of product or offer propensity models is becoming commonplace. But these single dimensional models do not take into account the impact of concurrent offers, channel or business constraints. This is where statistical optimisation techniques can help to resolve the problem of maximising return on investment (ROI) across all planned communications. The better solutions take into account interdependencies of financial goals, business constraints and customers’ needs to deliver the offer that is best for the business and satisfies the customers.

The   optimisation   technologies   allow the user to define simply the business constraints, input customer response data and specify which offers and communication channels to consider. The software automatically selects the appropriate   combination based on specified objectives, such as maximising profit, then rolls up the projected results and scores the database for offer distribution via the correct channel.

Business rules engines and other ‘best guess’ solutions that were invented to try to address this problem do not mathematically maximise the profit. These techniques do not consider the complicated interdependencies of multiple offers, are not easily updated with new information and cannot scale to the magnitude of the problem.

The following example compares the results of a rules-based approach to offer optimisation and a statistical optimisation approach. In this scenario:

Customer 1 has two potential offers: Offer A through call centre with an expected value of $100

Offer B through direct mail with an expected value of $75.

Customer 2 has two potential offers: Offer C through call centre with an expected value of $90

Offer D through direct mail with an expected value of $20.

Due to channel capacity constraints only one call-centre contact can be attempted.

Using a typical rules-based approach the following would result:

Customer 1 would receive Offer A through call centre with an expected value of $100.

Customer 2 would receive Offer D through direct mail with an expected value of $20.

Hence the expected return from a rules-based approach would be $120.

Using a statistical optimisation approach seeking to maximise the expected value, the following would result:

Customer 1 would receive Offer B through direct mail with an expected value of $75.

Customer 2 would receive Offer C through call centre with an expected value of $90.

Hence the expected return from the statistical optimisation approach would be $165.

Within the rules-based approach, the process selects the first customer from a ranked list; this does not necessarily deliver the optimal solution. Within the statistical optimisation approach, the process selects the best combination of offers to maximise the optimisation function.

Hybrid solutions for communication optimisation

The emergence of hybrid solutions has been seen where rule-engine technology has been combined with other technologies such as statistical optimisation or neural networks. This approach seems to take the best of both approaches and is gaining in popularity.

CONCLUSION

The use of rules-based engines is gaining ground in a number of areas as they allow business users to manage the evolution of the rules and are simple to understand. Their application in communication decision making is being proposed by a number of vendors and in some cases these technologies are being positioned as optimisation solutions.

They do have a role to play in a number   of areas but they come a poor second when compared to statistical optimisation for communication decision making.

SYNOPSIS

This paper explores the use of social network analysis in the Telco sector. Rather than focusing on the analytical techniques (many of which are proprietary and covered by patents), it focuses on the types of results that have been achieved to date and their business application in marketing.

BACKGROUND

My previous paper,‘The role of social networks in marketing’ in the JDM, stimulated many enquires about the subject, and as a result I have decided to focus this paper on the use of social network analysis in the telecommunications sector. Based on my international experience, this is probably the sector that is the most   mature in the business application of social network analysis in marketing.

DEFINITION

For the purposes of this paper

Social network analysis

Social network analysis is concerned with the analysis of the influence of an individual within a social network on product purchase and service usage.

Social network

A social network is a group of people who are connected through their use of mobile communication services.

Concept

A concept is a behaviour, for example, new product usage or an idea, that moves through the social network.

Wave

A wave is the measured way in which a concept flows through a social network.

Connection

A connection is the link between two individuals within a social network. The link may have a number of attributes to describe it, for example, direction, strength and concept fl ow speed. An individual may be connected to one or more individuals within the social network.

Influencer

An influencer is an individual who stimulates a concept (eg behavioural change) to flow through a network.

INTRODUCTION

The use of social network analysis is not new; it has been used in a number of areas including social science for more than 80 years. But it is only in the last few years that social network analysis has been seen in marketing. I have been working with US and European cellular service providers that have been experimenting with the subject for the last 18–24 months. It is only in the last few months that I have seen solutions going into production, often after pilot programmes.

One of the reasons that we are starting   to see production deployments of social network analysis in the Telco sector is the emergence of technology designed to meet the specific needs of marketing. In many cases this technology was originally developed to meet the needs of the anti- terrorist organisations in the military and security sectors. A natural step for these vendors has been to move into the fraud sector. A few have targeted the marketing arena from the start.

WHY THE EMERGENCE OF SOCIAL NETWORK ANALYSIS IN THE MOBILE TELCO SECTOR?

There are two key business reasons why the Telco sector has been an early adopter of social network analysis. These are

• Availability of data
• Business pain

The following section explores these in more detail.

Availability of good data

The mobile telecommunication sector is unique in that they have access to detailed call records made between individuals. This call data includes

• Caller number
• Caller handset identifier
• Receiver number
• Start time of the call
• End time of the call
• Day of week of call
• Time of day of call
• Duration of call
• Caller cell location
• Receiver cell location
• Receiver on or off network status

Similar types of data are available for text messaging.

This call data significantly simplifies the definition of the social network and influencers.

In addition:-

In the case of the caller (or receiver if on network) the organisation has access to

• Customer profile data
  • Comprehensive for contact customers, limited for pre-paid
• Account data
• Handset data
• Product involvement data
• Services usage data
• Inbound contact history, for example, call centre
• Marketing contact history and response data

This data allows us to understand the flow of concepts through the social network.

This breadth of data makes the Telco sector data rich for social network analysis.

There is one issue that still has to be addressed by most of the vendors and that is the volumes of data that need to be processed. To effectively analyse the flow of concepts through a social network means the solution has to process years worth of call data. In the US this means billions of call records.

BUSINESS PAIN

The second reason that the Telco sector has been the incubator for social network analysis in marketing is the presence of significant business pains.

These include

• Persistent problems with churn in the contract customer segment

Much work has been done to manage churn in the contract customer segments, in terms of both predicting customers who are a risk and developing effective marketing tools to address the churn. But there are still persistent problems with churn.

• Rapidly rising churn in the pre-paid customer segment

The last two years has seen a significant rise in churn in the pre-paid customer segment.The general decline in the price of voice services is further contributing to this problem. Unlike the contract customers the Telcos often have limited and in some cases no personal data

on these pre-paid customers.This has significantly reduced the available retention tools. It has also made it difficult to measure real churn in this sector because customers swap phones within the same provider.

• Need to increase revenue (ARPU) from data and other value added services

As price pressure from the competition on voice services is driving down revenue, the Telcos are looking at data and other value added services to grow monthly revenue per subscriber.

Managing the cost of selling these new products and services is becoming pivotal to success.

These two factors — available data and business pain — have driven the Telcos to explore the use of social network analysis in a number of areas.

HOW IS SOCIAL NETWORK ANALYSIS BEING USED?

The following section explores how social network analysis is being used in this sector.

My colleagues and I have been involved in a number of social network analysis projects and the following are examples of how social network analysis is being used in the Telco sector across the world.

The business applications include

• Improving churn prediction
• Improving customer value management
• Improving churn measurement
• Improving up sell campaign performance

The business applications are described in more detail below.

Improving churn prediction

In this case, the following data types were used to identify the social network for contract customers. A wide range of data were used that included

• Customer
• Account
• Handset
• Product and service usage data
• Detailed call record data
• Event history
• Channel usage

Historical data covering a number of years was used.

The social network analysis process generated a range of social network parameters at the customer and social network level. These parameters were then used as input variables in the current churn modelling process (regression).

The social network parameters

• Increased the model performance by a factor of ten or more
• Appeared in the list of top ten predictive variables in the model
• The retention campaigns saw and even better uplift (10–25 times)

The new model was then used to enhance the churn retention process.

This process was repeated for pre-paid customers and produced even better results.

IMPROVE CUSTOMER VALUE MANAGEMENT

In this case the Telco had a complex process to calculate the historical and near-term future value of a customer. This customer value had been embedded in a number of key business processes including

• Customer retention
• Customer recovery
• Customer experience management in call centre and stores
• Price plan development

As in the previous example a wide range of data were used to define the social networks and to identify customers who where influencers for a number of key concepts.

The customer values for the individual customers were combined to produce a customer level social network value. The customer and social network values were compared for customers on the base. The results showed that over 18 per cent of the customers in the current lowest customer value band were actually part of high value social networks. In addition about 7 per cent of the low value customers were identified as influencers. Across all segments the penetration of influencers varied between 7 and 18 per cent.

The Telco is now in the process of embedding the new social network value and influencer indicators into a number of core customer management business processes. The initial results of the changes in the retention management processes are proving very valuable.

IMPROVING CHURN MEASUREMENT

It had proven hard in the pre-paid segment to accurately measure ‘revolving churners’. These are customers who cancel one product and then replace with a new   product from the same network. In the case of contract customers name and address data can be used to monitor a customer’s purchases over time. In the case of pre-paid customers many organisations do not collect personal details and therefore find it very difficult to track these revolving churners.

One vendor I worked with has developed what they call ‘social network finger printing’, it this case they build the social networks and look to map new customers into existing social networks where one or more individuals has left (cancelled or gone dormant).

In one case they were able to identify 26 per cent of the churning customers as revolvers. The organisation has now created a new category of customer that they monitor on a regular basis.

The consulting partner is currently working with the organisation to develop more effective retention marketing tools for these revolving churners.

IMPROVING UP SELL CAMPAIGN PERFORMANCE

In this case the Telco has used information on influencers to target a range of direct marketing campaigns to influencers only. These campaigns included

• Hand set upgrades
• Price plan migration
• Data packages and bundles
• New hand set launches

The basic principle is not new to anyone who has run viral marketing campaigns. But in this case the proposition is sent to influencers only. The other members of the social networks receive no communication.

In some cases incentives are used and in other campaigns no incentive is used.

The influencer has effectively been used to promote the product or service to other members of the social network.

The results have been quite surprising.

The influencers do promote products and services even if they do not buy themselves

• One influencer stimulates between 2 and 17 other individuals in the social network to respond
• On average one influencer stimulates five–seven individuals in the social network to respond
• The campaign lifecycle (length of time people respond) doubles
• In some cases the response curve is bi-modal as concepts flow through the social network

The overall impact was that return on investment was more than five times better than the current campaigns.

Influencer campaigns are now becoming a standard part of the marketing mix in the Telco.

They are also looking at how to market   to the social network as micro-segments but they are in early stages.

These activities are now being called one to one to many marketing.

CONCLUSION

The marketing teams of the telecommunications sector have been an early adopter of social network analysis, primarily because of the availability of rich data and the desire to address key business pains.

The emergence of new technologies from the anti-terrorist industries that can be scaled to meet the data volumes and demands of this industry is now allowing social network analysis to become a production solution.

It is still early days and there is little experience globally in business applications in marketing. I, however, do see a time in the near future when social network   analysis, social network finger printing and one to one to many marketing are all common terms that we use. I believe social network analysis will have made a positive contribution to marketing in both the Telco and other sectors.