International Journal of Health Care Quality Assurance Improving a patient appointment call center at Mayo Clinic Rohleder Thomas Bailey Brian Crum Brian Faber Timothy Johnson Brandon Montgomery LeTesha Pringnitz Rachel
Article information:
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
To cite this document: Rohleder Thomas Bailey Brian Crum Brian Faber Timothy Johnson Brandon Montgomery LeTesha Pringnitz Rachel , (2013),"Improving a patient appointment call center at Mayo Clinic", International Journal of Health Care Quality Assurance, Vol. 26 Iss 8 pp. 714 - 728 Permanent link to this document: http://dx.doi.org/10.1108/IJHCQA-11-2011-0068 Downloaded on: 31 January 2016, At: 22:47 (PT) References: this document contains references to 24 other documents. To copy this document: [email protected] The fulltext of this document has been downloaded 610 times since 2013*
Users who downloaded this article also downloaded: Yu-Li Huang, (2013),"Ancillary service impact on outpatient scheduling", International Journal of Health Care Quality Assurance, Vol. 26 Iss 8 pp. 746-759 http://dx.doi.org/10.1108/IJHCQA-02-2012-0019 Jeremy Henri Maurice Veillard, Michaela Louise Schiøtz, Ann-Lise Guisset, Adalsteinn Davidson Brown, Niek S. Klazinga, (2013),"The PATH project in eight European countries: an evaluation", International Journal of Health Care Quality Assurance, Vol. 26 Iss 8 pp. 703-713 http://dx.doi.org/10.1108/IJHCQA-11-2011-0065 Mariado Carmo Caccia-Bava, Valerie C.K. Guimaraes, Tor Guimaraes, (2013),"Important factors for success in hospital BPR project phases", International Journal of Health Care Quality Assurance, Vol. 26 Iss 8 pp. 729-745 http:// dx.doi.org/10.1108/IJHCQA-01-2012-0007
Access to this document was granted through an Emerald subscription provided by emerald-srm:277061 []
For Authors If you would like to write for this, or any other Emerald publication, then please use our Emerald for Authors service information about how to choose which publication to write for and submission guidelines are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com Emerald is a global publisher linking research and practice to the benefit of society. The company manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as well as providing an extensive range of online products and additional customer resources and services. Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive preservation. *Related content and download information correct at time of download.
The current issue and full text archive of this journal is available at www.emeraldinsight.com/0952-6862.htm
IJHCQA 26,8
Improving a patient appointment call center at Mayo Clinic Thomas Rohleder and Brian Bailey
714 Received 21 November 2011 Revised 4 April 2012 Accepted 30 April 2012
Process Improvement Office, Vanderbilt University Medical Center, Nashville, Tennessee, USA
Brian Crum and Timothy Faber Health Sciences Research, Mayo Clinic, Rochester, Minnesota, USA
Brandon Johnson Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
Essentia Health, Information Services, Duluth, Minnesota, USA, and
LeTesha Montgomery and Rachel Pringnitz Health Sciences Research, Mayo Clinic, Rochester, Minnesota, USA Abstract Purpose – Contact centers for patient and referring physician are important to large medical-centers such as the Mayo Clinic’s Central Appointment Office (CAO). The aim of this case study is to report the process and results of a major process improvement effort, designed to simultaneously improve service quality and efficiency. Design/methodology/approach – Discrete-event simulation and optimization are used and linked to significant service improvements. Findings – The process improvement efforts led to about a 70 percent improvement in patient service performance as measured by average answering-speed (ASA) and average abandonment rate (AAR). This was achieved without adding additional staff, despite call volume increasing by 12 percent. Evaluating process improvement projects is difficult owing to the “phased” implementation of changes. Thus, there is no true control against which to compare. Additionally, the results are based on a single case study. Research limitations/implications – Evaluation of process improvement projects is difficult due to the “phased” implementation of changes. Thus, there is no true control to compare against. Practical implications – Contact center data and operations research methods, such as discrete-event simulation and optimization, can be integrated with change management, which results in significant process improvements in medical call-centers. Originality/value – Structured quantitative modeling of contact centers can be an important extension to traditional quality and process improvement techniques. Keywords Modelling, Process redesign, Service delivery Paper type Case study
International Journal of Health Care Quality Assurance Vol. 26 No. 8, 2013 pp. 714-728 q Emerald Group Publishing Limited 0952-6862 DOI 10.1108/IJHCQA-11-2011-0068
Introduction and background At 8:01 on February 21, 2011, the Scheduling Operations manager at Mayo Clinic in Rochester, Minnesota received an email from a Central Appointment Office supervisor alerting the operations manager, other supervisors and the analytics team, that The authors thank Randy Swanson for his assistance with the simulation and optimization analysis and Todd Huschka for his assistance with model input analysis.
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
service-levels were well-below target since the office opened an hour earlier. Call volumes were significantly higher than projected. Additionally, an overnight snow storm accounted for seven staff absences. Staff were aware of the performance issues and logged-in to the call handling system, even before their scheduled shift. In the email, the supervisor outlined the steps taken to improve performance, which included changing staff schedules, adjusting agent skills and monitoring real-time call center reports. This response by supervisor and staff followed initiatives to improve performance at the Appointment Information Desk – Referring Physician Service (AID-RPS). In this article, we discuss a case study that led to the quick response discussed above and significant ongoing service improvements. Our primary objective is to show how applying simulation and optimization modeling operations research tools resulted in a better staff match to patient call demand. This results in significantly better service to callers as measured by the average-speed-answer (ASA) and average abandonment rate (AAR). We discuss how using data and analytics were change drivers by ensuring administrators were accountable and provided motivation for change. The Mayo Clinic has nearly one million patient visits per year to its Scottsdale, Jacksonville and Rochester centers. However, there are more interactions with patients via telephone, regular mail correspondence, e-mail, web portals, etc. to facilitate these visits. The Rochester telephone call volume is about 340,000 calls per day. Managing these communications effectively and efficiently is important for ensuring quality care and service experience for Mayo’s patients. One important contact point for the Rochester, Minnesota location is the Central Appointment Office (CAO – its primary purpose is to handle calls from prospective patients and referring physicians. Contact centers like Mayo’s CAO are critical to coordinating patient care (Bentley et al., 2005). Lazarus (1997) notes call centers’ increasing importance to healthcare organization success. Thus, high-quality and timely service are key Mayo call-center strategies. In 2009, owing to increasing call volumes and issues with staff scheduling and task assignment, CAO patient service levels were variable. Caller abandonment rates were up to 22.7 percent in 2008 and 15.7 percent in 2009. These were well above the 5 percent maximum target and underscored that efforts were needed in 2010 to significantly improve service performance. Moreover, the CAO group needed to feel that they could be successful on a daily basis with consistent and planned staffing to match demand. Case study The Mayo CAO has three functions: (1) The Appointment Information Desk (AID) – triages calls from patients and either directs the call to the appropriate department or schedules an appointment with department staff. The AID staff also handle intra-clinic calls from patient service representatives, clinical assistants and medical secretaries. The AID staff receives around 4,500 calls and 450 online requests per week. (2) The Referring Physician Service (RPS) – handles telephone calls, faxes and online requests from physicians referring patients to the Mayo Clinic. The RPS receives around 2,900 calls and requests per week. Some online requests are submitted through a secure portal. Participating provider offices are enrolled in the portal and accept the terms of use. The online portal’s appealing feature is the referring provider’s ability to submit requests any time and trust that the
Improving a patient call center 715
IJHCQA 26,8
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
716
appointment request will be processed, and the appointment status will be communicated to the referring office. (3) The Patient Correspondence Center (PCC) – processes appointment confirmation letters in the institution. All departmental appointment desks are linked to PCC. Around 15,000 letters are received each week. A total of 60 staff engaged in these functions, who typically have a primary skill (AID or RPS) and may or may not be cross-trained for a secondary skill. Staff may also be assigned to the PCC, but will be able to take AID or RPS calls as well. For telephone calls, key CAO service metrics for are average answering speed (ASA) and average abandonment-rate (AAR). As noted in Gustafson’s (1999) article on patient financial service call centers, responding quickly to patient requests drives patient satisfaction. We note that abandonment rates for top retail operations strive to keep the frequency below 2-3 percent. Mayo Clinic’s targets are 30 seconds or less for ASA and 5 percent or less for AAR. The targets are intended to be met each day. For fax and online requests, a 24-hour turnaround time is the goal. A challenge to meet these measures is the highly variable call volume (Figure 1). Calls come in randomly – often bunched; weather, holidays and other events can exacerbate uncertainty for both call volumes and staff availability. Quality and process improvement initiatives The CAO is important for connecting call-center activities to Mayo Clinic’s strategies, clinical practice and managers’ overall plan to improve quality (Stier, 1999). Several components were required for CAO staff to achieve the desired performance.
Figure 1. CAO call volume variability
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
Culture change and collaboration The team decided early that there would be immediate opportunities to improve service and also a longer and more involved process based on their findings. Initially, the team leaders moved from a retrospective service-view to real-time performance monitoring. Measuring service performance in health service call centers and adapting to changing conditions is highlighted by Heinen (2006) who notes that quality needs to be measured from various perspectives to be useful. With the performance data and directive to use that data to drive change, there were efforts to layer staff coverage to ensure larger resource commitments to services offered by unit staff. At the same time, the CAO leadership group collaborated with the Mayo Clinic Office of Access Management (OAM) analytics team and the research-oriented Healthcare Policy and Research (HCPR) staff. Together, these teams began the process of understanding demand, capacity and optimal resource configuration. A key requirement for cultural change was acknowledging the problems. Collaboration with the analytic team required the leadership group to be more transparent with operational struggles and hurdles. By understanding individual performance metrics, the team began educating the team about how appointment coordination affected the overall service. Historically, the service metrics were generated and the operational team provided details and reasons for performance lapses. By understanding how their individual actions affected the overall team performance, along with the leadership group reinforcing team and individual goals, the team began achieving goals based on projections. The group began to shift from feeling the impact and responding why service was or was not meeting targets to managing expected performance. Heinen (2006) notes that looking at individual performance and really understanding metric effects at a detailed level is important for achieving a call center’s desired strategic objectives. Tactics to improve service The AID-RPS management team used four strategies to improve CAO service performance: (1) optimize telecommunications configuration; (2) reduce workload and improve efficiency; (3) improve productivity; and (4) staffing to workload. The first tactic was to optimize the telecommunications configuration to maximize agent utilization and to appropriately prioritize incoming calls. The CAO call center receives different call types: potential patients call AID to schedule appointments or to inquire about Mayo’s services; new or established patients call for information regarding travel, hotels and other general information about their trip; and patients call to cancel or reschedule existing appointments. The RPS receives calls from providers referring patients or to inquire about an existing patient that has been referred. To address these different call types, automated technology directs calls to specific agent groups. An automated attendant allows the caller to select the reason for calling from a menu. Although Mayo has institutional policies that generally prohibit auto-attendants, an exception was made for AID-RPS. Calls to schedule appointments were directed to the
Improving a patient call center 717
IJHCQA 26,8
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
718
core team. Calls to cancel or reschedule appointments were directed to another agent. Generally, cancel calls require less AID-specific knowledge, so these calls also use cross-coverage from other work units. Calls for general information, including travel, received a lower priority. Since most information was available on the website, the queue message was updated to direct patients with general questions to mayoclinic.org. Calls regarding medical records were directed to the medical records department. To address workload and efficiency, agent utilization was analyzed in detail to better understand how productive each patient appointment coordinator (PAC) was each day. Detailed agent-level performance monitoring is important to call center success in healthcare organizations (Ater, 1998; Elwell, 1998). Further, skill assignment by the different call types (AID-RPS) was always present in the unit, skills layering and real-time adjustments were not always reset to planned levels. To optimize skill configuration and to maximize agent utilization, an effort was made to increase cross-training within the unit so that multiple skills could be assigned to an individual PAC. A weighted skill priority method was used to determine the priority given to each skill across all agents. Providing primary skill assignments allowed PACs to feel confident with the calls they took through repetition, but secondary or tertiary assignments allowed enhanced coverage during peak times. To ensure that daily coverage was met, assignments were sometimes changed hourly by the supervisory team with the additional step of resetting the skill assignments to base levels each day for more predictable coverage. Perhaps the most impactful improvement strategy was staffing to workload. The goal was to reassign staff to meet incoming call demand. Staff to workload forecasting was required to project future call volumes. There were several challenges with forecasting AID-RPS call volumes. Changes in the ACD configuration, including adding and removing calls routed to these agent groups made the historical volumes less useful when forecasting future volumes. A moving average time-series forecast was used to project weekly volumes. The forecast compared weekly call volumes from the previous year, adjusted for growth trends. Once the weekly forecast was determined, historical weekday-patterns were used to project daily volumes (Figure 1). Analysis showed that call volumes by weekday as a weekly call volume percent were consistent throughout the year. Historical interval call patterns were used to project call volumes by time (Figure 1). Analyzing historical interval call patterns as a call volume percent for the day was consistent, despite call-volume variability. This forecasting method provided managers with call volume projections by hour. With this information, analysis could be provided to determine the appropriate staffing level throughout the day. Initially, an Excel add-in, Queueing ToolPak 4.0 (Ingolfsson and Gallup, 2011) was used to determine the appropriate hourly AID-RPS staffing levels. This tool has MS Excel functions that perform basic calculations for waiting line/time analysis. There were several limitations identified with this analysis: the queuing logic assumes there is one queue and does not account for an agent covering several queues. A spreadsheet model was developed that determined the average answer speed for different hourly call volume/staffing scenarios based on historical average AID-RPS call-handling time. The model output was the minimum staff required to meet the thirty-second average speed-to-answer performance target. Despite the model’s limitations, managers estimated future staffing requirements by week, day and time. Overall, combining cultural, training and technical components all contributed to process improvement.
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
Simulation and optimization modeling Simulation and other operations research modeling methods are being increasingly applied to healthcare settings. Common applications are in outpatient clinics (Clague et al., 1997; Rohleder et al., 2011) and surgical processes (Lehtonen et al., 2007). Several articles also broadly review the literature on applying operations research to healthcare. For example, Jun et al. (1999) apply discrete-event simulation and van Sambeek et al. (2010) provide a review simulation and analytic methods found in the medical literature. The latter article notes that in most studies, the models developed are not used for decision making; a shortcoming our research addresses. Using quantitative modeling to support decision making in call centers is particularly useful because there are inherent uncertainties. Variability in call arrival and call handling processes makes it difficult to predict performance without sophisticated tools such as discrete-event simulation (DES) (Law and Kelton, 2000). Dileepan and Ettkin (2010) emphasize that using simulation even for relatively small call centers like Mayo is justified owing to poor service’s high cost. While simulation can perform ‘what-if’ analyses to test performance following operational changes, it does not establish optimal decisions and those related to shift scheduling involves several alternatives. To effectively evaluate these alternatives, optimization or efficient search heuristics are required (Hillier and Lieberman, 2005).
Improving a patient call center 719
Simulation model Figure 2 shows the high-level simulation modeling structure that we used. It is based on Mehrotra and Fama’s (2003) general approach. Call forecast values were
Figure 2. Patient call center model overview
IJHCQA 26,8
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
720
based on past data (Figure 1). Call volume rates were modeled as a non-stationary Poisson process, common in other call center models (Aksin et al., 2007). Rates were adjusted every 15 minutes in the model to account for rapidly increasing and decreasing pattern shown in Figure 1. The hourly pattern was consistent by day and month; however, overall volumes did change for longer phases. Therefore, the simulation model was based on a full day and different model parameters were used to account for the daily and monthly seasonal effects. Call handling times were also an important input. Data were collected on call types (e.g. patients calling the AID versus referring physicians calling the RPS) during quality evaluation efforts. Input distributions to use in the simulation model were exposed to Expert Fit (Averill M. Law and Associates, 2002) software, e.g. the AID patient calls were matched to Lognormal (8.5, 7.5) distribution function and the RPS physician calls to the Lognormal (10.8, 8.6) function. Breaking down fundamental agent skills was based on their ability to handle AID and RPS calls. Of the total PACs included in the base model, about 25 percent were assigned AID-only, 25 percent RPS-only and the remaining could handle both AID and RPS calls. These skills were based on agent training and skill levels at the time the model was built. Queues were then assigned to agents by using resource sets that contained the appropriate dedicated and multi-skilled PACs. Initial PAC schedules were based on the CAO’s actual schedules, which had evolved ad hoc over several years resulting in a reasonable match between staffing and call demand. Nonetheless, we observed that the existing schedules were sometimes overstaffed during low-call volume periods and understaffed in other periods. Given the significant call-volumes variability and call handling times, determining the best schedules was a challenging task for the supervisors without significant analytic support. The final simulation-model component was the abandonment process. To properly model caller abandonment, a patience time distribution had to be determined based on historical data. We used the approach described in Mandelbaum et al. (2001), shown to be effective for call centers. We estimated the average time to abandonment at about 13 minutes using historical data (see Appendix 1). Using the data and associated assumptions, we created a computer model using Arena simulation software (Rockwell Automation, 2009). The model was validated by comparing performance measures output by the model to actual system performance. Table I compares two key measures: average answering-speed and caller abandonment rate. For both measures, the confidence intervals for the means from the simulation model output include the mean value based on CAO data. These results support the simulation model’s validity (Law and Kelton, 2000; see Figure 3).
Performance measure
Table I. Model validity
Simulation model (n ¼ 30) Actual results
Average answering-speed, seconds (95% CIs)
Average abandonment-rate (95% CIs)
30.2 (22.4 to 38) 33
4.43% (4% to 4.91%)a 4.35%
*Note: aCI proportion based on the Wilson procedure with continuity (Newcombe, 1998)
Improving a patient call center
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
721
Figure 3. Call-handling times – distribution
Optimal shift assignment Adding flexibility to call-center shifts improves performance (Hannif et al., 2010). Thus, we chose to consider adding more shift options to better match workload and staff. In particular, because call volumes increased rapidly in the morning and decreased quickly starting at about 4 pm, staggering shifts more effectively was required. Also, the original break schedules were not structured in a way to avoid too many staff being away from the telephones at the same time. Therefore, we developed a process to create shift schedules that avoided these problems and assign them more efficiently. A structured approach was used to create shift schedules to cover 7:00 am to 7:00 pm. A total of 52 separate full-time shift schedules starting from 7:00 am until 10:00 am at 15 minute intervals were established. At each quarter hour, four separate schedules were created with non-overlapping break and lunch schedules to provide flexibility. While it might have been possible to select optimal shift start and break times using mathematical modeling, for practical reasons, predefined schedules were used. The optimization formulation was then straightforward: Minimize: J X
Xj
j¼1
where: Xj ¼ An integer variable 0,1,2 . . . representing staff assigned to each shift, j;
IJHCQA 26,8
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
722
Figure 4. Old and new shift schedules (AID)
j
¼ A member of the J set ¼ 52 different shift schedules. For example, shifts 1 through 4 start at 7:00, shifts 5 – 8 at 7:15 . . . and shifts 49 – 52 at 10:00.
Expression 1 objectively minimizes overall staff numbers assigned to answering phones. Two constraints were added to ensure that desired service performance was achieved: average answering-speed # 30 seconds; and abandon rate # 0.05. These constraints were for each day, not hourly. This eased the burden of achieving very tight service levels at beginning and end of each work day when call volumes were low. Nonetheless, for most periods, performance measures were met or exceeded. To solve the model, we used the Optqueste solver available in the Arena simulation modeling tool, which uses a metaheuristic approach employing scatter and tabu searches that do not guarantee optimality but is intended to lead to high-quality solutions. To help ensure good, if not optimal, solutions, we ran many replications (30 simulations for each shift assignment scenario) and forced Optquest to run at least 1,000 shift assignments for each scenario. Different scenarios were run for different weekdays and different times of the year to establish base line shift assignment schedules. Because the assignments changed by weekday and time of year, we analyzed the shifts to identify those selected most frequently. We then restricted the schedules to the 20-25 that were most commonly used. An example assignment is shown in Appendix 2. This overall approach achieved approximate optimal solutions while easing the implementation process by limiting different shifts. Figure 4 shows how the new shift scheduling method just described creates a better match to call workload demand compared to past practice, which matched average workload generally; however, there were periods where capacity significantly exceeded or fell below average workload. The new shift schedules lead to far fewer periods where capacity greatly exceeds or falls below average workload. Generally, capacity will exceed average workload owing to the variability in call arrivals and handling times and the high service-level being targeted. Lunch periods and breaks during certain times also lead to excess capacity in some periods. In the end, our method projected that the CAO could achieve desired service levels with fewer overall staff assigned to call handling.
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
Results The process improvement initiatives changed CAO processes and practices in 2009; however, fully implementing most improvement actions including data-driven modeling discussed in the previous section were in place by early 2010. Thus, it is difficult to determine an exact time at which to perform a before-and-after analysis. Such evaluation challenges are typical in process improvement efforts, where processes are dynamically evolving over time. Nonetheless, comparing performance for 2009 and 2010 provides a reasonable approach. Only selecting a portion of each year (e.g. early 2009 vs late 2010) might entail clearer before and after implementation epochs, but would be confounded with the seasonal call and staffing patterns (e.g. lower winter-month call volumes). Figure 5 compares caller abandonment rates for 2009 and 2010 for the AID call center, where the most significant changes to scheduling were implemented. Performance is consistently better for 2010 (the plot for ASA shows a similar pattern). Table II shows that the mean and variance for ASA and the AAR are all statistically significantly lower in 2010 than 2009, despite call volumes being approximately 12 percent higher in 2010 than 2009 and staffing in 2010 was at or below the 2009 levels (precise FTEs are difficult to determine owing to structural changes, but no additional staff were added). This time-series approach has been used by other researchers to show process improvement initiatives’ statistically significant effects (Chaboyer et al., 2012).
Improving a patient call center 723
Figure 5. Before and after implementation AAR comparison (AID)
Year
Mean
ASA (sec) Variance
AAR Mean proportion
2009 2010
66.56 21.63
1217.17 150.79
0.094 0.028
Notes: t ¼ 8:76 ð p ¼ 0:000Þ; F ¼ 8:07 ð p ¼ 0:0000Þ; Z ¼ 109:65 ð p ¼ 0:0000Þ
Table II. CAO performance 2009 vs. 2010
IJHCQA 26,8
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
724
Conclusions Contact centers are important service resources for large medical centers. In this article, we presented a process-improvement case study at the Mayo Clinic CAO. The initiative’s primary goal was to reduce the average answering-speed and caller abandonment rate without increasing staff levels. The major improvement method applied quantitative data analysis and operations research tools to improve staff scheduling. Analytics helped to lead cultural change, collaboration and other service improvement tactics. These efforts resulted in about a 70 percent improvement in the key response measures. These dramatic improvements were achieved without reducing patient/agent interaction quality (which is continuously monitored). Study limitations include improvements occurring over time; i.e. there is no true control to compare against the improved process. Additionally, we report a single case study at Mayo Clinic. The culture and dynamics in other organizations may not result in the same performance improvements. Nonetheless, improvement performance and its link to the actions taken, support our methods. Major performance improvement started with a cultural change that required Mayo Clinic leaders to face the performance problems head on. Staff at all levels had to take accountability for the CAO’s performance. Facilitating these changes required collaboration with analytics experts and call center staff in an integrative manner so that performance and their drivers could be openly understood and discussed. The cultural change around data and performance led to improvements in telecommunications technology, staff training and skills, and most importantly, determining how many staff needed to be available at what times. To address the last concern, discrete-event simulation and optimization were used to model the call center and determine the best staff assignments. Using these science-based techniques helped to ensure that implementation would be successful rather than using a trial and error process. Call center managers and supervisors were more certain that the new schedules would be successful. This commitment was important given the natural staff-resistance to change. Analyzing Mayo Clinic’s CAO continues, with an eye to further improving performance. Finding creative ways to manage the annual seasonality along with the dynamic effects of patients and Mayo’s customers adopting new technology is an ongoing challenge. However, with the improved organizational culture around the CAO and a growing analytical toolkit, we are confident that Mayo’s contact center will continue to be an asset to the organization. References Aksin, Z., Armony, M. and Mehrotra, V. (2007), “The modern call center: a multi-disciplinary perspective on operations management research”, Production and Operations Management, Vol. 16 No. 6, pp. 665-688. Ater, D. (1998), “Teleservicing the needs of the health-care industry”, Telemarketing and Call Center Solutions, Vol. 16 No. 12, pp. 92-96. Averill M. Law and Associates (2002), ExpertFit v. 7.0, Tuscon, AZ. Bentley, P.J., Turner, V.F., Hodgson, S.A., Drimatis, R. and Hart, J. (2005), “A central role for the health call centre”, Australian Health Review, Vol. 29 No. 4, pp. 435-438. Chaboyer, W., Lin, F., Foster, M., Retallick, L., Panuwatwanich, K. and Richards, B. (2012), “Redesigning the ICU nursing discharge process: a quality improvement study”, Worldviews on Evidenced Based Nursing, Vol. 9 No. 1, pp. 40-48.
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
Clague, J.E., Reed, P.G., Barlow, J., Rada, R., Clarke, M. and Edwards, R.H. (1997), “Improving outpatient clinic efficiency using computer simulation”, International Journal of Health Care Quality Assurance, Vol. 10 Nos 4-5, pp. 197-201. Dileepan, P. and Ettkin, L.P. (2010), “Not just for large companies: benefits of simulation modeling for a small telephone call center”, Production and Inventory Management Journal, Vol. 46 No. 1, pp. 54-63. Elwell, R. (1998), “Healthcare providers quality monitoring pays off”, Telemarketing and Call Center Solutions, Vol. 16 No. 7, pp. 98-100. Gustafson, B.M. (1999), “A well-staff PFS call center can improve patient satisfaction”, Healthcare Financial Management, Vol. 53 No. 7, pp. 64-66. Hannif, Z., McDonnell, A., Connell, J. and Burgess, J. (2010), “Working time flexibilities: a paradox in call centres?”, Australian Bulletin of Labour, Vol. 36 No. 2, pp. 178-194. Heinen, J. (2006), “Improving customer service: taking a strategic approach to measuring contact center performance”, Health Management Technology, Vol. 27 No. 5, pp. 34-36. Hillier, F.S. and Lieberman, G.J. (2005), Introduction to Operations Research, 8th ed., McGraw-Hill, New York, NY. Ingolfsson, A. and Gallup, F. (2011), Queuing ToolPak 4.0, available at: http://apps. business.ualberta. ca/aingolfsson/qtp (accessed July 14, 2011). Jun, J., Jacobson, S. and Swisher, J. (1999), “Application of discrete-event simulation in healthcare clinics: a survey”, Journal of the Operational Research Society, Vol. 50, pp. 109-123. Law, A.M. and Kelton, W.D. (2000), Simulation Modeling and Analysis, 3rd ed., McGraw-Hill, New York, NY. Lazarus, I. (1997), “Putting personal health management on the line”, Managed Healthcare, Vol. 7 No. 7, pp. 53-55. Lehtonen, J.M., Kujala, J., Kouri, J. and Hippelainen, M. (2007), “Cardiac surgery productivity and throughput improvements”, International Journal of Healthcare Quality Assurance, Vol. 20 No. 1, pp. 40-52. Mandelabaum, A. (2001), Empirical Modeling of Call Centers, Technical Report, Technion, Haifa. Mehrotra, V. and Fama, J. (2003), “Call center simulation modeling: methods, challenges, and opportunities”, Proceedings of the 2003 Winter Simulation Conference, ACM, New Orleans, LA, pp. 135-143. Newcombe, R.G. (1998), “Two-sided confidence intervals for the single proportion: comparison of seven methods”, Statistics in Medicine, Vol. 17, pp. 857-862. Rockwell Automation (2009), ARENA v. 13.0, Milwaukee, WI. Rohleder, T., Lewkonia, P., Bischak, D., Duffy, P. and Hendijani, R. (2011), “Using simulation modeling to improve patient flow at an outpatient orthopedic clinic”, Healthcare Management Science, Vol. 14 No. 2, pp. 135-145. Stier, R.D. (1999), “The medical call center: a strategic marketing resource for the future”, Marketing Health Services, Vol. 19 No. 2, pp. 25-28. van Sambeek, J.R.C., Cornelissen, F.A., Bakker, P.J.M. and Krabbendam, J.J. (2010), “Models as instruments for optimizing hospital processes: a systematic review”, International Journal of Health Care Quality Assurance, Vol. 23 No. 4, pp. 356-377.
Improving a patient call center 725
IJHCQA 26,8
726
Appendix 1. Caller abandonment modeling To get the expected values for P, the patience time (time until a caller will abandon) and N - time callers need to wait in a queue, we can use the following expressions: E(P)
¼ E(WA) þ E(WS)(PS / PA).
E(N
¼ E(WS) þE(WA)(PA / PS).
where: E(WS
¼ Average waiting time for patients who are served (only count those that went into a queue).
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
E(WA) ¼ Average waiting time for patients who abandon (again only those who go into queue). PS
¼ Proportion who wait and get served.
PA
¼ Proportion who wait but abandon before being served.
The above approach assumes an exponential patience and waiting-time distribution. This assumption has been validated as reasonable in other call center studies (Mandelbaum, 2001).
Appendix 2. Revised optimal schedules – an example (italicised rows identify shifts omitted from secondary optimization)
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
Start 7:00 am 7:00 am 7:00 am 7:00 am 7:15 am 7:15 am 7:15 am 7:15 am 7:30 am 7:30 am 7:30 am 7:30 am 7:45 am 7:45 am 7:45 am 7:45 am 8:00 am 8:00 am 8:00 am 8:00 am 8:15 am 8:15 am 8:15 am 8:15 am 8:30 am 8:30 am 8:30 am 8:30 am 8:45 am 8:45 am 8:45 am 8:45 am 9:00 am 9:00 am 9:00 am 9:00 am 9:15 am 9:15 am 9:15 am 9:15 am 9:30 am 9:30 am 9:30 am 9:30 am 9:45 am 9:45 am
Shift
M 1 Nov
T 2 Nov
W 3 Nov
Th 4 Nov
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46
0 0 1 2 3 2 0 1 2 0 0 1 2 1 0 0 1 2 0 0 0 1 1 0 2 1 0 1 1 0 0 0 2 1 0 1 0 0 0 0 0 0 0 0 1 0
0 0 1 2 2 2 0 1 1 0 0 0 2 1 0 0 1 0 0 0 0 1 1 0 1 0 0 1 0 0 0 0 2 1 0 1 0 0 0 0 0 0 0 0 1 0
0 0 1 2 2 2 0 1 0 0 0 0 2 1 0 0 1 1 0 0 0 0 1 0 1 0 0 1 0 0 0 0 2 0 0 1 0 0 0 0 0 0 0 0 1 0
0 0 1 2 2 2 0 1 1 0 0 0 2 1 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 2 1 0 1 0 0 0 0 0 0 0 0 1 0
Improving a patient call center
F 5 Nov 0 0 0 1 2 1 0 0 2 0 0 1 0 1 0 0 1 2 0 0 0 1 1 0 1 1 0 1 1 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 1 0 (continued)
727
Table AI.
IJHCQA 26,8
728
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
Table AI.
Start 9:45 am 9:45 am 10:00 am 10:00 am 10:00 am 10:00 am Total
Shift 47 48 49 50 51 52
M 1 Nov
T 2 Nov
W 3 Nov
Th 4 Nov
F 5 Nov
0 0 0 2 0 0 32
0 0 0 2 0 0 24
0 0 0 2 0 0 22
0 0 0 2 0 0 21
0 0 0 0 0 0 20
Corresponding author Thomas Rohleder can be contacted at: [email protected]
To purchase reprints of this article please e-mail: [email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints
This article has been cited by:
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
1. Allyson M. Best, Cinnamon A. Dixon, W. David Kelton, Christopher J. Lindsell, Michael J. Ward. 2014. Using discrete event computer simulation to improve patient flow in a Ghanaian acute care hospital. The American Journal of Emergency Medicine 32, 917-922. [CrossRef]
Article information:
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
To cite this document: Rohleder Thomas Bailey Brian Crum Brian Faber Timothy Johnson Brandon Montgomery LeTesha Pringnitz Rachel , (2013),"Improving a patient appointment call center at Mayo Clinic", International Journal of Health Care Quality Assurance, Vol. 26 Iss 8 pp. 714 - 728 Permanent link to this document: http://dx.doi.org/10.1108/IJHCQA-11-2011-0068 Downloaded on: 31 January 2016, At: 22:47 (PT) References: this document contains references to 24 other documents. To copy this document: [email protected] The fulltext of this document has been downloaded 610 times since 2013*
Users who downloaded this article also downloaded: Yu-Li Huang, (2013),"Ancillary service impact on outpatient scheduling", International Journal of Health Care Quality Assurance, Vol. 26 Iss 8 pp. 746-759 http://dx.doi.org/10.1108/IJHCQA-02-2012-0019 Jeremy Henri Maurice Veillard, Michaela Louise Schiøtz, Ann-Lise Guisset, Adalsteinn Davidson Brown, Niek S. Klazinga, (2013),"The PATH project in eight European countries: an evaluation", International Journal of Health Care Quality Assurance, Vol. 26 Iss 8 pp. 703-713 http://dx.doi.org/10.1108/IJHCQA-11-2011-0065 Mariado Carmo Caccia-Bava, Valerie C.K. Guimaraes, Tor Guimaraes, (2013),"Important factors for success in hospital BPR project phases", International Journal of Health Care Quality Assurance, Vol. 26 Iss 8 pp. 729-745 http:// dx.doi.org/10.1108/IJHCQA-01-2012-0007
Access to this document was granted through an Emerald subscription provided by emerald-srm:277061 []
For Authors If you would like to write for this, or any other Emerald publication, then please use our Emerald for Authors service information about how to choose which publication to write for and submission guidelines are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com Emerald is a global publisher linking research and practice to the benefit of society. The company manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as well as providing an extensive range of online products and additional customer resources and services. Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive preservation. *Related content and download information correct at time of download.
The current issue and full text archive of this journal is available at www.emeraldinsight.com/0952-6862.htm
IJHCQA 26,8
Improving a patient appointment call center at Mayo Clinic Thomas Rohleder and Brian Bailey
714 Received 21 November 2011 Revised 4 April 2012 Accepted 30 April 2012
Process Improvement Office, Vanderbilt University Medical Center, Nashville, Tennessee, USA
Brian Crum and Timothy Faber Health Sciences Research, Mayo Clinic, Rochester, Minnesota, USA
Brandon Johnson Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
Essentia Health, Information Services, Duluth, Minnesota, USA, and
LeTesha Montgomery and Rachel Pringnitz Health Sciences Research, Mayo Clinic, Rochester, Minnesota, USA Abstract Purpose – Contact centers for patient and referring physician are important to large medical-centers such as the Mayo Clinic’s Central Appointment Office (CAO). The aim of this case study is to report the process and results of a major process improvement effort, designed to simultaneously improve service quality and efficiency. Design/methodology/approach – Discrete-event simulation and optimization are used and linked to significant service improvements. Findings – The process improvement efforts led to about a 70 percent improvement in patient service performance as measured by average answering-speed (ASA) and average abandonment rate (AAR). This was achieved without adding additional staff, despite call volume increasing by 12 percent. Evaluating process improvement projects is difficult owing to the “phased” implementation of changes. Thus, there is no true control against which to compare. Additionally, the results are based on a single case study. Research limitations/implications – Evaluation of process improvement projects is difficult due to the “phased” implementation of changes. Thus, there is no true control to compare against. Practical implications – Contact center data and operations research methods, such as discrete-event simulation and optimization, can be integrated with change management, which results in significant process improvements in medical call-centers. Originality/value – Structured quantitative modeling of contact centers can be an important extension to traditional quality and process improvement techniques. Keywords Modelling, Process redesign, Service delivery Paper type Case study
International Journal of Health Care Quality Assurance Vol. 26 No. 8, 2013 pp. 714-728 q Emerald Group Publishing Limited 0952-6862 DOI 10.1108/IJHCQA-11-2011-0068
Introduction and background At 8:01 on February 21, 2011, the Scheduling Operations manager at Mayo Clinic in Rochester, Minnesota received an email from a Central Appointment Office supervisor alerting the operations manager, other supervisors and the analytics team, that The authors thank Randy Swanson for his assistance with the simulation and optimization analysis and Todd Huschka for his assistance with model input analysis.
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
service-levels were well-below target since the office opened an hour earlier. Call volumes were significantly higher than projected. Additionally, an overnight snow storm accounted for seven staff absences. Staff were aware of the performance issues and logged-in to the call handling system, even before their scheduled shift. In the email, the supervisor outlined the steps taken to improve performance, which included changing staff schedules, adjusting agent skills and monitoring real-time call center reports. This response by supervisor and staff followed initiatives to improve performance at the Appointment Information Desk – Referring Physician Service (AID-RPS). In this article, we discuss a case study that led to the quick response discussed above and significant ongoing service improvements. Our primary objective is to show how applying simulation and optimization modeling operations research tools resulted in a better staff match to patient call demand. This results in significantly better service to callers as measured by the average-speed-answer (ASA) and average abandonment rate (AAR). We discuss how using data and analytics were change drivers by ensuring administrators were accountable and provided motivation for change. The Mayo Clinic has nearly one million patient visits per year to its Scottsdale, Jacksonville and Rochester centers. However, there are more interactions with patients via telephone, regular mail correspondence, e-mail, web portals, etc. to facilitate these visits. The Rochester telephone call volume is about 340,000 calls per day. Managing these communications effectively and efficiently is important for ensuring quality care and service experience for Mayo’s patients. One important contact point for the Rochester, Minnesota location is the Central Appointment Office (CAO – its primary purpose is to handle calls from prospective patients and referring physicians. Contact centers like Mayo’s CAO are critical to coordinating patient care (Bentley et al., 2005). Lazarus (1997) notes call centers’ increasing importance to healthcare organization success. Thus, high-quality and timely service are key Mayo call-center strategies. In 2009, owing to increasing call volumes and issues with staff scheduling and task assignment, CAO patient service levels were variable. Caller abandonment rates were up to 22.7 percent in 2008 and 15.7 percent in 2009. These were well above the 5 percent maximum target and underscored that efforts were needed in 2010 to significantly improve service performance. Moreover, the CAO group needed to feel that they could be successful on a daily basis with consistent and planned staffing to match demand. Case study The Mayo CAO has three functions: (1) The Appointment Information Desk (AID) – triages calls from patients and either directs the call to the appropriate department or schedules an appointment with department staff. The AID staff also handle intra-clinic calls from patient service representatives, clinical assistants and medical secretaries. The AID staff receives around 4,500 calls and 450 online requests per week. (2) The Referring Physician Service (RPS) – handles telephone calls, faxes and online requests from physicians referring patients to the Mayo Clinic. The RPS receives around 2,900 calls and requests per week. Some online requests are submitted through a secure portal. Participating provider offices are enrolled in the portal and accept the terms of use. The online portal’s appealing feature is the referring provider’s ability to submit requests any time and trust that the
Improving a patient call center 715
IJHCQA 26,8
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
716
appointment request will be processed, and the appointment status will be communicated to the referring office. (3) The Patient Correspondence Center (PCC) – processes appointment confirmation letters in the institution. All departmental appointment desks are linked to PCC. Around 15,000 letters are received each week. A total of 60 staff engaged in these functions, who typically have a primary skill (AID or RPS) and may or may not be cross-trained for a secondary skill. Staff may also be assigned to the PCC, but will be able to take AID or RPS calls as well. For telephone calls, key CAO service metrics for are average answering speed (ASA) and average abandonment-rate (AAR). As noted in Gustafson’s (1999) article on patient financial service call centers, responding quickly to patient requests drives patient satisfaction. We note that abandonment rates for top retail operations strive to keep the frequency below 2-3 percent. Mayo Clinic’s targets are 30 seconds or less for ASA and 5 percent or less for AAR. The targets are intended to be met each day. For fax and online requests, a 24-hour turnaround time is the goal. A challenge to meet these measures is the highly variable call volume (Figure 1). Calls come in randomly – often bunched; weather, holidays and other events can exacerbate uncertainty for both call volumes and staff availability. Quality and process improvement initiatives The CAO is important for connecting call-center activities to Mayo Clinic’s strategies, clinical practice and managers’ overall plan to improve quality (Stier, 1999). Several components were required for CAO staff to achieve the desired performance.
Figure 1. CAO call volume variability
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
Culture change and collaboration The team decided early that there would be immediate opportunities to improve service and also a longer and more involved process based on their findings. Initially, the team leaders moved from a retrospective service-view to real-time performance monitoring. Measuring service performance in health service call centers and adapting to changing conditions is highlighted by Heinen (2006) who notes that quality needs to be measured from various perspectives to be useful. With the performance data and directive to use that data to drive change, there were efforts to layer staff coverage to ensure larger resource commitments to services offered by unit staff. At the same time, the CAO leadership group collaborated with the Mayo Clinic Office of Access Management (OAM) analytics team and the research-oriented Healthcare Policy and Research (HCPR) staff. Together, these teams began the process of understanding demand, capacity and optimal resource configuration. A key requirement for cultural change was acknowledging the problems. Collaboration with the analytic team required the leadership group to be more transparent with operational struggles and hurdles. By understanding individual performance metrics, the team began educating the team about how appointment coordination affected the overall service. Historically, the service metrics were generated and the operational team provided details and reasons for performance lapses. By understanding how their individual actions affected the overall team performance, along with the leadership group reinforcing team and individual goals, the team began achieving goals based on projections. The group began to shift from feeling the impact and responding why service was or was not meeting targets to managing expected performance. Heinen (2006) notes that looking at individual performance and really understanding metric effects at a detailed level is important for achieving a call center’s desired strategic objectives. Tactics to improve service The AID-RPS management team used four strategies to improve CAO service performance: (1) optimize telecommunications configuration; (2) reduce workload and improve efficiency; (3) improve productivity; and (4) staffing to workload. The first tactic was to optimize the telecommunications configuration to maximize agent utilization and to appropriately prioritize incoming calls. The CAO call center receives different call types: potential patients call AID to schedule appointments or to inquire about Mayo’s services; new or established patients call for information regarding travel, hotels and other general information about their trip; and patients call to cancel or reschedule existing appointments. The RPS receives calls from providers referring patients or to inquire about an existing patient that has been referred. To address these different call types, automated technology directs calls to specific agent groups. An automated attendant allows the caller to select the reason for calling from a menu. Although Mayo has institutional policies that generally prohibit auto-attendants, an exception was made for AID-RPS. Calls to schedule appointments were directed to the
Improving a patient call center 717
IJHCQA 26,8
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
718
core team. Calls to cancel or reschedule appointments were directed to another agent. Generally, cancel calls require less AID-specific knowledge, so these calls also use cross-coverage from other work units. Calls for general information, including travel, received a lower priority. Since most information was available on the website, the queue message was updated to direct patients with general questions to mayoclinic.org. Calls regarding medical records were directed to the medical records department. To address workload and efficiency, agent utilization was analyzed in detail to better understand how productive each patient appointment coordinator (PAC) was each day. Detailed agent-level performance monitoring is important to call center success in healthcare organizations (Ater, 1998; Elwell, 1998). Further, skill assignment by the different call types (AID-RPS) was always present in the unit, skills layering and real-time adjustments were not always reset to planned levels. To optimize skill configuration and to maximize agent utilization, an effort was made to increase cross-training within the unit so that multiple skills could be assigned to an individual PAC. A weighted skill priority method was used to determine the priority given to each skill across all agents. Providing primary skill assignments allowed PACs to feel confident with the calls they took through repetition, but secondary or tertiary assignments allowed enhanced coverage during peak times. To ensure that daily coverage was met, assignments were sometimes changed hourly by the supervisory team with the additional step of resetting the skill assignments to base levels each day for more predictable coverage. Perhaps the most impactful improvement strategy was staffing to workload. The goal was to reassign staff to meet incoming call demand. Staff to workload forecasting was required to project future call volumes. There were several challenges with forecasting AID-RPS call volumes. Changes in the ACD configuration, including adding and removing calls routed to these agent groups made the historical volumes less useful when forecasting future volumes. A moving average time-series forecast was used to project weekly volumes. The forecast compared weekly call volumes from the previous year, adjusted for growth trends. Once the weekly forecast was determined, historical weekday-patterns were used to project daily volumes (Figure 1). Analysis showed that call volumes by weekday as a weekly call volume percent were consistent throughout the year. Historical interval call patterns were used to project call volumes by time (Figure 1). Analyzing historical interval call patterns as a call volume percent for the day was consistent, despite call-volume variability. This forecasting method provided managers with call volume projections by hour. With this information, analysis could be provided to determine the appropriate staffing level throughout the day. Initially, an Excel add-in, Queueing ToolPak 4.0 (Ingolfsson and Gallup, 2011) was used to determine the appropriate hourly AID-RPS staffing levels. This tool has MS Excel functions that perform basic calculations for waiting line/time analysis. There were several limitations identified with this analysis: the queuing logic assumes there is one queue and does not account for an agent covering several queues. A spreadsheet model was developed that determined the average answer speed for different hourly call volume/staffing scenarios based on historical average AID-RPS call-handling time. The model output was the minimum staff required to meet the thirty-second average speed-to-answer performance target. Despite the model’s limitations, managers estimated future staffing requirements by week, day and time. Overall, combining cultural, training and technical components all contributed to process improvement.
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
Simulation and optimization modeling Simulation and other operations research modeling methods are being increasingly applied to healthcare settings. Common applications are in outpatient clinics (Clague et al., 1997; Rohleder et al., 2011) and surgical processes (Lehtonen et al., 2007). Several articles also broadly review the literature on applying operations research to healthcare. For example, Jun et al. (1999) apply discrete-event simulation and van Sambeek et al. (2010) provide a review simulation and analytic methods found in the medical literature. The latter article notes that in most studies, the models developed are not used for decision making; a shortcoming our research addresses. Using quantitative modeling to support decision making in call centers is particularly useful because there are inherent uncertainties. Variability in call arrival and call handling processes makes it difficult to predict performance without sophisticated tools such as discrete-event simulation (DES) (Law and Kelton, 2000). Dileepan and Ettkin (2010) emphasize that using simulation even for relatively small call centers like Mayo is justified owing to poor service’s high cost. While simulation can perform ‘what-if’ analyses to test performance following operational changes, it does not establish optimal decisions and those related to shift scheduling involves several alternatives. To effectively evaluate these alternatives, optimization or efficient search heuristics are required (Hillier and Lieberman, 2005).
Improving a patient call center 719
Simulation model Figure 2 shows the high-level simulation modeling structure that we used. It is based on Mehrotra and Fama’s (2003) general approach. Call forecast values were
Figure 2. Patient call center model overview
IJHCQA 26,8
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
720
based on past data (Figure 1). Call volume rates were modeled as a non-stationary Poisson process, common in other call center models (Aksin et al., 2007). Rates were adjusted every 15 minutes in the model to account for rapidly increasing and decreasing pattern shown in Figure 1. The hourly pattern was consistent by day and month; however, overall volumes did change for longer phases. Therefore, the simulation model was based on a full day and different model parameters were used to account for the daily and monthly seasonal effects. Call handling times were also an important input. Data were collected on call types (e.g. patients calling the AID versus referring physicians calling the RPS) during quality evaluation efforts. Input distributions to use in the simulation model were exposed to Expert Fit (Averill M. Law and Associates, 2002) software, e.g. the AID patient calls were matched to Lognormal (8.5, 7.5) distribution function and the RPS physician calls to the Lognormal (10.8, 8.6) function. Breaking down fundamental agent skills was based on their ability to handle AID and RPS calls. Of the total PACs included in the base model, about 25 percent were assigned AID-only, 25 percent RPS-only and the remaining could handle both AID and RPS calls. These skills were based on agent training and skill levels at the time the model was built. Queues were then assigned to agents by using resource sets that contained the appropriate dedicated and multi-skilled PACs. Initial PAC schedules were based on the CAO’s actual schedules, which had evolved ad hoc over several years resulting in a reasonable match between staffing and call demand. Nonetheless, we observed that the existing schedules were sometimes overstaffed during low-call volume periods and understaffed in other periods. Given the significant call-volumes variability and call handling times, determining the best schedules was a challenging task for the supervisors without significant analytic support. The final simulation-model component was the abandonment process. To properly model caller abandonment, a patience time distribution had to be determined based on historical data. We used the approach described in Mandelbaum et al. (2001), shown to be effective for call centers. We estimated the average time to abandonment at about 13 minutes using historical data (see Appendix 1). Using the data and associated assumptions, we created a computer model using Arena simulation software (Rockwell Automation, 2009). The model was validated by comparing performance measures output by the model to actual system performance. Table I compares two key measures: average answering-speed and caller abandonment rate. For both measures, the confidence intervals for the means from the simulation model output include the mean value based on CAO data. These results support the simulation model’s validity (Law and Kelton, 2000; see Figure 3).
Performance measure
Table I. Model validity
Simulation model (n ¼ 30) Actual results
Average answering-speed, seconds (95% CIs)
Average abandonment-rate (95% CIs)
30.2 (22.4 to 38) 33
4.43% (4% to 4.91%)a 4.35%
*Note: aCI proportion based on the Wilson procedure with continuity (Newcombe, 1998)
Improving a patient call center
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
721
Figure 3. Call-handling times – distribution
Optimal shift assignment Adding flexibility to call-center shifts improves performance (Hannif et al., 2010). Thus, we chose to consider adding more shift options to better match workload and staff. In particular, because call volumes increased rapidly in the morning and decreased quickly starting at about 4 pm, staggering shifts more effectively was required. Also, the original break schedules were not structured in a way to avoid too many staff being away from the telephones at the same time. Therefore, we developed a process to create shift schedules that avoided these problems and assign them more efficiently. A structured approach was used to create shift schedules to cover 7:00 am to 7:00 pm. A total of 52 separate full-time shift schedules starting from 7:00 am until 10:00 am at 15 minute intervals were established. At each quarter hour, four separate schedules were created with non-overlapping break and lunch schedules to provide flexibility. While it might have been possible to select optimal shift start and break times using mathematical modeling, for practical reasons, predefined schedules were used. The optimization formulation was then straightforward: Minimize: J X
Xj
j¼1
where: Xj ¼ An integer variable 0,1,2 . . . representing staff assigned to each shift, j;
IJHCQA 26,8
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
722
Figure 4. Old and new shift schedules (AID)
j
¼ A member of the J set ¼ 52 different shift schedules. For example, shifts 1 through 4 start at 7:00, shifts 5 – 8 at 7:15 . . . and shifts 49 – 52 at 10:00.
Expression 1 objectively minimizes overall staff numbers assigned to answering phones. Two constraints were added to ensure that desired service performance was achieved: average answering-speed # 30 seconds; and abandon rate # 0.05. These constraints were for each day, not hourly. This eased the burden of achieving very tight service levels at beginning and end of each work day when call volumes were low. Nonetheless, for most periods, performance measures were met or exceeded. To solve the model, we used the Optqueste solver available in the Arena simulation modeling tool, which uses a metaheuristic approach employing scatter and tabu searches that do not guarantee optimality but is intended to lead to high-quality solutions. To help ensure good, if not optimal, solutions, we ran many replications (30 simulations for each shift assignment scenario) and forced Optquest to run at least 1,000 shift assignments for each scenario. Different scenarios were run for different weekdays and different times of the year to establish base line shift assignment schedules. Because the assignments changed by weekday and time of year, we analyzed the shifts to identify those selected most frequently. We then restricted the schedules to the 20-25 that were most commonly used. An example assignment is shown in Appendix 2. This overall approach achieved approximate optimal solutions while easing the implementation process by limiting different shifts. Figure 4 shows how the new shift scheduling method just described creates a better match to call workload demand compared to past practice, which matched average workload generally; however, there were periods where capacity significantly exceeded or fell below average workload. The new shift schedules lead to far fewer periods where capacity greatly exceeds or falls below average workload. Generally, capacity will exceed average workload owing to the variability in call arrivals and handling times and the high service-level being targeted. Lunch periods and breaks during certain times also lead to excess capacity in some periods. In the end, our method projected that the CAO could achieve desired service levels with fewer overall staff assigned to call handling.
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
Results The process improvement initiatives changed CAO processes and practices in 2009; however, fully implementing most improvement actions including data-driven modeling discussed in the previous section were in place by early 2010. Thus, it is difficult to determine an exact time at which to perform a before-and-after analysis. Such evaluation challenges are typical in process improvement efforts, where processes are dynamically evolving over time. Nonetheless, comparing performance for 2009 and 2010 provides a reasonable approach. Only selecting a portion of each year (e.g. early 2009 vs late 2010) might entail clearer before and after implementation epochs, but would be confounded with the seasonal call and staffing patterns (e.g. lower winter-month call volumes). Figure 5 compares caller abandonment rates for 2009 and 2010 for the AID call center, where the most significant changes to scheduling were implemented. Performance is consistently better for 2010 (the plot for ASA shows a similar pattern). Table II shows that the mean and variance for ASA and the AAR are all statistically significantly lower in 2010 than 2009, despite call volumes being approximately 12 percent higher in 2010 than 2009 and staffing in 2010 was at or below the 2009 levels (precise FTEs are difficult to determine owing to structural changes, but no additional staff were added). This time-series approach has been used by other researchers to show process improvement initiatives’ statistically significant effects (Chaboyer et al., 2012).
Improving a patient call center 723
Figure 5. Before and after implementation AAR comparison (AID)
Year
Mean
ASA (sec) Variance
AAR Mean proportion
2009 2010
66.56 21.63
1217.17 150.79
0.094 0.028
Notes: t ¼ 8:76 ð p ¼ 0:000Þ; F ¼ 8:07 ð p ¼ 0:0000Þ; Z ¼ 109:65 ð p ¼ 0:0000Þ
Table II. CAO performance 2009 vs. 2010
IJHCQA 26,8
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
724
Conclusions Contact centers are important service resources for large medical centers. In this article, we presented a process-improvement case study at the Mayo Clinic CAO. The initiative’s primary goal was to reduce the average answering-speed and caller abandonment rate without increasing staff levels. The major improvement method applied quantitative data analysis and operations research tools to improve staff scheduling. Analytics helped to lead cultural change, collaboration and other service improvement tactics. These efforts resulted in about a 70 percent improvement in the key response measures. These dramatic improvements were achieved without reducing patient/agent interaction quality (which is continuously monitored). Study limitations include improvements occurring over time; i.e. there is no true control to compare against the improved process. Additionally, we report a single case study at Mayo Clinic. The culture and dynamics in other organizations may not result in the same performance improvements. Nonetheless, improvement performance and its link to the actions taken, support our methods. Major performance improvement started with a cultural change that required Mayo Clinic leaders to face the performance problems head on. Staff at all levels had to take accountability for the CAO’s performance. Facilitating these changes required collaboration with analytics experts and call center staff in an integrative manner so that performance and their drivers could be openly understood and discussed. The cultural change around data and performance led to improvements in telecommunications technology, staff training and skills, and most importantly, determining how many staff needed to be available at what times. To address the last concern, discrete-event simulation and optimization were used to model the call center and determine the best staff assignments. Using these science-based techniques helped to ensure that implementation would be successful rather than using a trial and error process. Call center managers and supervisors were more certain that the new schedules would be successful. This commitment was important given the natural staff-resistance to change. Analyzing Mayo Clinic’s CAO continues, with an eye to further improving performance. Finding creative ways to manage the annual seasonality along with the dynamic effects of patients and Mayo’s customers adopting new technology is an ongoing challenge. However, with the improved organizational culture around the CAO and a growing analytical toolkit, we are confident that Mayo’s contact center will continue to be an asset to the organization. References Aksin, Z., Armony, M. and Mehrotra, V. (2007), “The modern call center: a multi-disciplinary perspective on operations management research”, Production and Operations Management, Vol. 16 No. 6, pp. 665-688. Ater, D. (1998), “Teleservicing the needs of the health-care industry”, Telemarketing and Call Center Solutions, Vol. 16 No. 12, pp. 92-96. Averill M. Law and Associates (2002), ExpertFit v. 7.0, Tuscon, AZ. Bentley, P.J., Turner, V.F., Hodgson, S.A., Drimatis, R. and Hart, J. (2005), “A central role for the health call centre”, Australian Health Review, Vol. 29 No. 4, pp. 435-438. Chaboyer, W., Lin, F., Foster, M., Retallick, L., Panuwatwanich, K. and Richards, B. (2012), “Redesigning the ICU nursing discharge process: a quality improvement study”, Worldviews on Evidenced Based Nursing, Vol. 9 No. 1, pp. 40-48.
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
Clague, J.E., Reed, P.G., Barlow, J., Rada, R., Clarke, M. and Edwards, R.H. (1997), “Improving outpatient clinic efficiency using computer simulation”, International Journal of Health Care Quality Assurance, Vol. 10 Nos 4-5, pp. 197-201. Dileepan, P. and Ettkin, L.P. (2010), “Not just for large companies: benefits of simulation modeling for a small telephone call center”, Production and Inventory Management Journal, Vol. 46 No. 1, pp. 54-63. Elwell, R. (1998), “Healthcare providers quality monitoring pays off”, Telemarketing and Call Center Solutions, Vol. 16 No. 7, pp. 98-100. Gustafson, B.M. (1999), “A well-staff PFS call center can improve patient satisfaction”, Healthcare Financial Management, Vol. 53 No. 7, pp. 64-66. Hannif, Z., McDonnell, A., Connell, J. and Burgess, J. (2010), “Working time flexibilities: a paradox in call centres?”, Australian Bulletin of Labour, Vol. 36 No. 2, pp. 178-194. Heinen, J. (2006), “Improving customer service: taking a strategic approach to measuring contact center performance”, Health Management Technology, Vol. 27 No. 5, pp. 34-36. Hillier, F.S. and Lieberman, G.J. (2005), Introduction to Operations Research, 8th ed., McGraw-Hill, New York, NY. Ingolfsson, A. and Gallup, F. (2011), Queuing ToolPak 4.0, available at: http://apps. business.ualberta. ca/aingolfsson/qtp (accessed July 14, 2011). Jun, J., Jacobson, S. and Swisher, J. (1999), “Application of discrete-event simulation in healthcare clinics: a survey”, Journal of the Operational Research Society, Vol. 50, pp. 109-123. Law, A.M. and Kelton, W.D. (2000), Simulation Modeling and Analysis, 3rd ed., McGraw-Hill, New York, NY. Lazarus, I. (1997), “Putting personal health management on the line”, Managed Healthcare, Vol. 7 No. 7, pp. 53-55. Lehtonen, J.M., Kujala, J., Kouri, J. and Hippelainen, M. (2007), “Cardiac surgery productivity and throughput improvements”, International Journal of Healthcare Quality Assurance, Vol. 20 No. 1, pp. 40-52. Mandelabaum, A. (2001), Empirical Modeling of Call Centers, Technical Report, Technion, Haifa. Mehrotra, V. and Fama, J. (2003), “Call center simulation modeling: methods, challenges, and opportunities”, Proceedings of the 2003 Winter Simulation Conference, ACM, New Orleans, LA, pp. 135-143. Newcombe, R.G. (1998), “Two-sided confidence intervals for the single proportion: comparison of seven methods”, Statistics in Medicine, Vol. 17, pp. 857-862. Rockwell Automation (2009), ARENA v. 13.0, Milwaukee, WI. Rohleder, T., Lewkonia, P., Bischak, D., Duffy, P. and Hendijani, R. (2011), “Using simulation modeling to improve patient flow at an outpatient orthopedic clinic”, Healthcare Management Science, Vol. 14 No. 2, pp. 135-145. Stier, R.D. (1999), “The medical call center: a strategic marketing resource for the future”, Marketing Health Services, Vol. 19 No. 2, pp. 25-28. van Sambeek, J.R.C., Cornelissen, F.A., Bakker, P.J.M. and Krabbendam, J.J. (2010), “Models as instruments for optimizing hospital processes: a systematic review”, International Journal of Health Care Quality Assurance, Vol. 23 No. 4, pp. 356-377.
Improving a patient call center 725
IJHCQA 26,8
726
Appendix 1. Caller abandonment modeling To get the expected values for P, the patience time (time until a caller will abandon) and N - time callers need to wait in a queue, we can use the following expressions: E(P)
¼ E(WA) þ E(WS)(PS / PA).
E(N
¼ E(WS) þE(WA)(PA / PS).
where: E(WS
¼ Average waiting time for patients who are served (only count those that went into a queue).
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
E(WA) ¼ Average waiting time for patients who abandon (again only those who go into queue). PS
¼ Proportion who wait and get served.
PA
¼ Proportion who wait but abandon before being served.
The above approach assumes an exponential patience and waiting-time distribution. This assumption has been validated as reasonable in other call center studies (Mandelbaum, 2001).
Appendix 2. Revised optimal schedules – an example (italicised rows identify shifts omitted from secondary optimization)
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
Start 7:00 am 7:00 am 7:00 am 7:00 am 7:15 am 7:15 am 7:15 am 7:15 am 7:30 am 7:30 am 7:30 am 7:30 am 7:45 am 7:45 am 7:45 am 7:45 am 8:00 am 8:00 am 8:00 am 8:00 am 8:15 am 8:15 am 8:15 am 8:15 am 8:30 am 8:30 am 8:30 am 8:30 am 8:45 am 8:45 am 8:45 am 8:45 am 9:00 am 9:00 am 9:00 am 9:00 am 9:15 am 9:15 am 9:15 am 9:15 am 9:30 am 9:30 am 9:30 am 9:30 am 9:45 am 9:45 am
Shift
M 1 Nov
T 2 Nov
W 3 Nov
Th 4 Nov
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46
0 0 1 2 3 2 0 1 2 0 0 1 2 1 0 0 1 2 0 0 0 1 1 0 2 1 0 1 1 0 0 0 2 1 0 1 0 0 0 0 0 0 0 0 1 0
0 0 1 2 2 2 0 1 1 0 0 0 2 1 0 0 1 0 0 0 0 1 1 0 1 0 0 1 0 0 0 0 2 1 0 1 0 0 0 0 0 0 0 0 1 0
0 0 1 2 2 2 0 1 0 0 0 0 2 1 0 0 1 1 0 0 0 0 1 0 1 0 0 1 0 0 0 0 2 0 0 1 0 0 0 0 0 0 0 0 1 0
0 0 1 2 2 2 0 1 1 0 0 0 2 1 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 2 1 0 1 0 0 0 0 0 0 0 0 1 0
Improving a patient call center
F 5 Nov 0 0 0 1 2 1 0 0 2 0 0 1 0 1 0 0 1 2 0 0 0 1 1 0 1 1 0 1 1 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 1 0 (continued)
727
Table AI.
IJHCQA 26,8
728
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
Table AI.
Start 9:45 am 9:45 am 10:00 am 10:00 am 10:00 am 10:00 am Total
Shift 47 48 49 50 51 52
M 1 Nov
T 2 Nov
W 3 Nov
Th 4 Nov
F 5 Nov
0 0 0 2 0 0 32
0 0 0 2 0 0 24
0 0 0 2 0 0 22
0 0 0 2 0 0 21
0 0 0 0 0 0 20
Corresponding author Thomas Rohleder can be contacted at: [email protected]
To purchase reprints of this article please e-mail: [email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints
This article has been cited by:
Downloaded by University of Arizona At 22:47 31 January 2016 (PT)
1. Allyson M. Best, Cinnamon A. Dixon, W. David Kelton, Christopher J. Lindsell, Michael J. Ward. 2014. Using discrete event computer simulation to improve patient flow in a Ghanaian acute care hospital. The American Journal of Emergency Medicine 32, 917-922. [CrossRef]
