2000 Systems Engineering Capstone Conference • University of Virginia SCORECARD DEVELOPMENT FOR A DEBT COLLECTION SYSTEM Student team: Michael Autuoro, Hope Breskman, John Congable, Arin Shapiro, Rachel Yoo Faculty Advisor: William T. Scherer Associate Professor, Department of Systems Engineering Client Advisor: Douglas Fuller First Select Corporation Strategic Operations 4460 Rosewood Drive P.O. Box 9053 Pleasanton, CA 94566 Doug_Fuller@providian.com KEYWORDS: agent, gaming, metrics, Promise To Pay, Refusal To Pay, Right Party Connect ABSTRACT First Select Corporation (FSC) is a young collections company that needed to develop a set of performance measurements, or metrics, to assist in operating an effective call center where its primary business operations take place. The metric development process started with the first phase of topdown analysis. The team researched to become acquainted with metrics through previous studies and case examples. Meetings with FSC management took place in the early stages to establish overall FSC goals. Performance measurements that would support those goals were brainstormed, using knowledge gained from research. After the Capstone team received relevant call center data, the second phase of bottom-up analysis begun, primarily consisting of data mining and scorecard development. A scorecard format facilitates the evaluation process with its graphic layout and concise, specific, and significant data. The scorecard was divided into four employee levels within FSC: management, division, team, and agent. The scorecard can be traced from the top-level down to the agent level. Outliers in the management level scorecard might be detected and further investigated to determine what behavior might have caused that phenomenon. Therefore, data from lower levels might help upper-level management to pinpoint the cause of the underperformance of FSC as a company. INTRODUCTION Providian Financial Corporation is the sixth largest credit card provider in the world. Providian targets high-risk customers and, in exchange, charges higher rates than its competitors. Specifically, Providian attributes much of its success to its unique collections strategy. Its excellence in collections led to the development of First Select Corporation (FSC), a wholly-owned subsidiary that is completely dedicated to debt collection. Its primary business operations take place in its call center located in Louisville, KY. First Select not only collects Providian customers’ debts but also buys debts from other institutions at a reduced cost. In the past few years, Providian has steadily increased 30-40% in market value. Keeping up with the rapid growth of Providian, First Select’s call center in Louisville expanded to over 300 employees in one year of development. One consequence of this rapid growth has been the failure to use a standard set of complete metrics to regularly measure the call center’s performance. In order to progress, FSC management, who make executive decisions, must be capable of evaluating its performance and knowing which areas require improvement based on the available data. The purpose of this Capstone project was to provide an accurate and useful instrument with which FSC can measure its performance, target its weaknesses, and exploit its strengths. 63 Scorecard Development For A Debt Collection System OVERALL PROCESS REVIEW The Capstone project was divided into two phases. The first phase provided an opportunity to get acquainted with the client, First Select Corporation, and understand the problem definition. Conceptualization was the approach to metric development in this topdown phase of the project. In the spring semester, the Capstone team received data from FSC and was able to perform data analysis to begin the second phase. Research took place throughout the entire course of the project. Particularly valuable was the research about the process for developing these metrics, the characteristics of successful metrics, and, in phase two, the existing measurements of performance for call centers. Metrics were then generated from the available data in the second phase, after cleansing and scrubbing the call center data. The combined efforts of both phases produced the final deliverable to the client: the scorecard. BACKGROUND RESEARCH The Capstone team chose to use Sprague’s five traits used as the foundation for metric development during the project duration. It was important to understand the value of each characteristic and repercussions of lacking those traits. When metrics were brainstormed in the initial phase of the project, the aforementioned significant attributes were held as guidelines for metric selection. SCORECARD DEVELOPMENT It was established by the FSC Capstone team that a scorecard would be the most effective means to display various metrics for each employee level. Scorecards facilitate efficient and effective performance review. This was particularly important for higher-level management so that spotting trouble areas would not be a time-consuming, arduous task. The employee structure within FSC can be broken into the levels shown in Figure 1. A scorecard was customized for each of the general four levels: management, division, team, and agent. Measurements must support the high-level goals. The overall goals of FSC were used to determine which metrics would support them. Obviously, goals are unique to the enterprise and the individuals who make it up. Therefore, it is universally agreed that metrics must be unique to the organization they are evaluating. “It’s a question of using the right metric for the right job” (Carter, 14). Research has shown that more and more businesses are driving improvement through the direction that measurements of performance can provide. Bob Frost of Measurement International says, “executives require significantly better measures to direct strategy and business performance than ever before” (Frost, 34). All levels throughout a company can benefit from effective measures. The challenge is to make the measurement effective by maintaining high standards during the metric development process. There are certain vital attributes a performance measure must have to be useful. According to Gail Sprague of Vanguard Communications, there are five main principles for effective measures: 1. Simplicity 2. Drive the right behavior 3. Measurable and Available 4. Accountability associated with control 5. Support high level business goals Source: “Three Steps for Creating Effective Call Center Measures and Reports” by Gail Sprague 64 President Senior Vice President Call center operations manager Dialer operations manager Quality Assurance manager Division managers Team Leaders Agents Figure 1. FSC Management Hierarchy Management Level In providing top-level executives with information on call center performance, the following were the primary objectives: 2000 Systems Engineering Capstone Conference • University of Virginia 1. 2. The metrics must be succinct and comprehensible. The metrics must quickly pinpoint abnormalities in the system. The metrics must incorporate the experience and intuition of an executive employee. All metrics were divided into inbound and outbound performance due to the distinct differences in the respective customer characteristics for each. Although a top priority in metric development was to find a correlation between potential metrics and money brought in the door, lack of payment information precluded this study. With only one month of summary information, analyses focused on creating time-series representations of data and recognizing deviations from normal behavior. “Division” refers to an account’s status in the collection process. The data must be divided according to division because accounts in different divisions are not comparable to each other. A section manager oversees all of the accounts that are in his specific division. Section managers will be most interested in viewing data on a division level. The same metrics on the management level scorecard are included for the division level. If outliers are detected on the management level scorecard, the division level might explain some of the distinctive and seemingly peculiar behavior of a single division. 3. As a first step in identifying metrics, several interviews with top-level executives took place. At these meetings, call center information was prioritized and each executive’s preferred method of analysis was reviewed. This information led to a set of preliminary metrics and relationships that were aligned with the intuition and expertise of the leaders of the company. In conducting this set of interviews, information was obtained that could not be garnered otherwise. For example, several executives mentioned that they often gauge the quality of a day by calculating the ratio of Promise to Pays to Total Dialer Hours. Historically, this ratio has had little variance from day to day and therefore, can be used to evaluate a day’s performance. The second stage in the development of the toplevel metrics was to verify and validate their usefulness. In doing so, the Capstone team generated specific call center scenarios and rated each metric on how well it indicated that scenario. Specifically, call-center technological problems, dialer malfunction, and employee gaming situations were hypothesized. In rating each metric on how well it predicted the aforementioned scenarios, the scorecard was further developed. As a result of an additional iteration of this twostage process the Capstone team completed the toplevel scorecard, which consists of the following metrics: 1. 2. 3. 4. 5. Total number of Promise to Pays achieved per day Total number of Dialer Hours per day The number of Right Party Connects/Dialer Hour per day. The percentage of Right Party Connects that were Promises to Pay, per day. The percentage of Inbound Calls that are abandoned by customers in the queue. Division Level Most of the data in the division level is presented in a time series format so that the section managers can view trends and patterns in the data, as well as individual anomalies. In addition, each division has a summary graph that shows how his division is performing compared to the others. All five of the high-level metrics are of interest at the division level. Team Level The team level of the scorecard compares different teams in the same business aspect of First Select Corporation, i.e. First Select Front End. For our template scorecard, teams were randomly assigned and varied in size from ten to fifteen agents. Similar to the relationship between the management and division level, outliers detected on the division level might be explained by performance on the team level within a particular division. The team level is utilized by supervising managers when comparing the productivity of different teams of agents. For example, if a team is under-performing, the unit manager may notify that team’s leader. Metrics used for the team level are the same as those used on the individual agent level, just average for the team. The identification and acknowledgement of team success is critical in business today. The team level scorecard denotes these qualities and generates a greater sense of teamwork in the office. In addition to their own success, agents might be encouraged to focus on the success of the team. This approach might reduce the existence of a competitive atmosphere, thus reducing the chances of cheating. Having members 65 Scorecard Development For A Debt Collection System working and measured as a team will promote morale and reward those groups of agents that have the same goals as the management of First Select Corporation. Agent Level While it is important for FSC to focus on teamwork and avoid individual competition, evaluation on the agent level is a necessity for numerous reasons. Being the bottom rung of our scorecard, the agent level can be used by team and unit managers to compare an agent’s performance to that of their peers. Knowing individual agent performance will allow First Select to isolate the superior employees from the weaker ones. Consistently low performers might be given more training for future improvement, while strong agents might be given more responsibility or aid the struggling representatives. developed by the FSC Capstone team, using % PTPs in Outbound Calls as the measurement of performance. % PTPs are the percentage of connected calls in which the customer promised the agent a certain payment amount to contribute to the debt owed. When an FSC employee on the management level receives his scorecard, he will first look for unusual behavior, according to what has already been established as the norm (typically the average value). For example, a graph of % PTPs for all divisions for a three-week period will be displayed (each week is connected by a line): )dnuobtuO(PTP % %0.43 The metrics used in evaluating the agent level were the following: %0.23 %0.03 Promises to Pay per Right Party Connect Refusals to Pay per Right Party Connect (RPC) Talk Times per time logged onto the dialer Update Times per time logged onto the dialer Outbound Right Party Connects per Dialer Hour (RPC/DH) %0.82 %0.62 %0.42 %0.22 %0.02 0 0/ 8 / 1 0 0/ 9 / 1 0 0/ 0 1 / 1 0 0/ 1 1 / 1 0 0/ 2 1 / 1 0 0/ 3 1 / 1 0 0/ 4 1 / 1 0 0/ 5 1 / 1 0 0/ 6 1 / 1 0 0/ 7 1 / 1 0 0/ 8 1 / 1 0 0/ 9 1 / 1 0 0/ 0 2 / 1 0 0/ 1 2 / 1 0 0/ 2 2 / 1 0 0/ 3 2 / 1 0 0/ 4 2 / 1 66 0 0/ 5 2 / 1 The following is a step-by-step application of the drilldown functionality of the scorecard for FSC 0 0/ 6 2 / 1 DRILLDOWN APPLICATION 0 0/ 7 2 / 1 Although each metric was analyzed and presented individually to compare agents to one another, a star chart was also developed to provide an extensive summary for each agent shown. Agents might quickly realize what metrics by which they are being measured, and thus manipulate, or game, the system to give the appearance of better performance. Therefore, the ability to look at numerous metrics in one step in a star chart provides First Select with the ability to detect this gaming phenomenon, target weaknesses, and exploit strengths. 0 0/ 8 2 / 1 Since there is limited data regarding the dollars collected by each agent, PTPs are the best surrogate to measure each agent’s contribution to the dollars brought in the door. In the case of RPC/DH, only outbound performance was valid under the assumption that almost all inbound calls are RPCs and an agent has virtually no control over the number of outbound RPCs he or she works. etaD Graph 1. % PTP from Outbound Calls The reader sees that the difference between January 10th and January 11th is noticeably large. First, he will notice that Jan. 10th is a Monday and Jan. 11th is a Tuesday. Because this graph is a comprehensive presentation of all divisions, he might want to look at the behavior within each division on those particular days. Graph 2 shows the % PTP Outbound calls within one particular division, in this case, RCY_FSC. The % PTPs on Jan. 10th and Jan. 11th are similar to the relationship shown in Graph 2. A division manager who is responsible for the agents and teams within his division might wonder what has caused the low performance on January 13 th. The % PTPs for the RCY_FSC division for that day is almost two standard deviations away from the mean value. P TP % 1. 2. 3. 4. 5. 2000 Systems Engineering Capstone Conference • University of Virginia Graph 4. %PTP Outbound RCY_FSC Agent AR1AAMIN % PTP Outbound: RCY_FSC 40.00% 35.00% %PTP Outbound RCY_FSC Agent AR1AAMIN Date % Calls Abandoned Minus 1 Sigma Minus 2 Sigma Graph 2. % PTP Outbound RCY_FSC In order to understand what has caused the pattern of behavior in the above graph for the RCY_FSC division, the division manager might talk to the team leaders. The team leaders can each look at those particular days on Graph 3 and describe what happened on the team level. % PTP Outbound RCY_FSC Team #1 25.00% Percent 20.00% Plus 1 Sigma Plus 2 Sigma Overall Average SCORECARD SYSTEM REQUIREMENTS There are two potential types of graphs on each scorecard. One is the star chart, which displays multiple metrics and the other are “up-down” graphs. The “up-down” graphs display the individual data on the x-axis and the percentage on the y-axis. Each points is shown as a bar raised or dropped from the mean value. Lines showing plus and minus one and two standard deviations allow the reader to understand the general variation of the data. 15.00% Management Level The management level scorecard includes comprehensive data for the combined behavior of all divisions according to various time series. 10.00% 5.00% 28-Jan 27-Jan 26-Jan 25-Jan 24-Jan 21-Jan 20-Jan 19-Jan 18-Jan 14-Jan 13-Jan 12-Jan 11-Jan 10-Jan 0.00% Date % Calls Abandoned Plus 2 Sigma Plus 1 Sigma Minus 2 Sigma Minus 1 Sigma Overall Average Graph 3. %PTP Outbound RCY_FSC Team #1 The agent level scorecard is used to notice agent behavior over a period of time, rather than for one agent on a particular day. First a particular agent might perform on average below the norm. The team leader would then look at that agent’s performance over a period of time using Graph 4. Division Level The scorecard for each division is displayed according to a time series. Because of the diverse characteristics of each division, it is instructive to observe data separated by division. These graphs are shown in a time series format. Team Level The team level are similar to the agent level. Various teams within one division can be displayed on one graph to compare performance. Additionally, the performance of one team across a time series can provide additional information of the performance specific to that team. 67 28-Jan 27-Jan 26-Jan 25-Jan 24-Jan 21-Jan 20-Jan 19-Jan Minus 1 Sigma Overall Average 18-Jan Plus 1 Sigma Minus 2 Sigma 14-Jan % Outbound PTP Plus 2 Sigma 13-Jan Date 10-Jan 1/28/00 1/27/00 1/26/00 1/25/00 1/24/00 1/23/00 1/22/00 1/21/00 1/20/00 1/19/00 1/18/00 1/17/00 1/16/00 1/15/00 1/14/00 1/13/00 1/12/00 1/11/00 10.00% 1/10/00 15.00% 12-Jan 20.00% 45.00% 40.00% 35.00% 30.00% 25.00% 20.00% 15.00% 10.00% 5.00% 0.00% 11-Jan 25.00% Percent Percent 30.00% Scorecard Development For A Debt Collection System Agent Level Agent level analysis should typically encompass a time frame, which permits enough data to be gathered for accurate statistical results. For example, one might think that an agent with a PTP conversion rate of 90% within the time frame of analysis is performing extremely well. Statistically speaking however, that conversion rate is not meaningful when the agent was only passed ten RPCs and converted nine of them to PTPs. Therefore, an agent’s total number of RPCs must be taken into consideration when looking at the PTP conversion rate for an agent. The data used for the duration of this project involved roughly three weeks of agent data. In most cases, enough data was present to provide a significant statistical analysis, but in some cases, more data would be necessary. For this reason, agent analysis should incorporate at least one month of data. BIOGRAPHIES CONCLUSIONS John Congable is a fourth year Systems Engineering major from Richmond, VA, concentrating in Management Information Systems. When he is not working on this Capstone Project, he can be found beating his drums. Next year John will be working for First Select Corporation in Pleasanton, CA. The purpose of this Capstone project was to provide First Select Corporation with an effective means for performance evaluation. A scorecard for each level of employment within FSC according to various time series and employee series increases the value of information contained in the graphs and tables of performance. By consolidating this information for comparison, management at FSC can use their time efficiently. REFERENCES “Best Practices: Best Practices in Call Center Management, Operations, and Technology”. ProSci Benchmarking Report, 1999. Carter, John. “Are you using the right metrics for continuous improvement?” Production. Vol. 106, Sept. 1994, 14-15. Faulkner and Gray. “If the Call Center Fits…” Business and Management Practices. Collections & Credit Risk. July 1999, vol. 4 no. 7 p. 65-71. Frost, Bob. “Performance Metrics: the new strategic discipline.” Strategy & Leadership. Vol. 27, May-June 1999, p. 34-36. Sprague, Gail. “Three Steps for Effective Call Center Measures and Reports.” ACD Call Center: Online Learning Center. Vanguard Communications. 68 Michael Autuoro is a fourth year Systems Engineering major from Brooklyn, NY, concentrating in Economic Systems. Mike has an incredible base of knowledge in various New York sport teams and general computer information; his wisdom is astounding. After graduating, Mike plans on either entering one of the distinguished law schools he has been accepted to or entering the world as a professional in one of many possible fields. Hope Breskman is a fourth year Systems Engineering major from Philadelphia, PA, concentrating in Management Systems. Besides being incredibly active in the Jewish community at UVA, Hope is also an avid reader. This summer she will begin her financial career with BlackRock Financial in New York City. Arin Shapiro is a fourth year Systems Engineering major from Corning, NY, concentrating in Economic Systems. Besides dedicating his life to this Capstone Project, he works at a local restaurant on the Corner and encourages his teammates with a motivational cheer at every meeting. Upon completion of the Systems Engineering Curriculum, he plans to work as an IT consultant for James Martin in Northern Virginia. Rachel Yoo is a fourth year Systems Engineering major from Atlanta, GA, concentrating in Biomedical Management. When not participating in Honor meetings, she hones her formatting skills. After graduating, Rachel will be living in San Francisco, CA, while working for A.T. Kearney as a Business Analyst.