Final Report - Optimizing Work Distribution for NP Orders
- 1. Optimizing Work Distribution for NP Orders Submitted to: John Boosey Metering Services Indianapolis Power & Light Prepared by: Nate Dotzlaf Senior Business Systems Analyst BCforward Brian Kaiser Lead Engineer, Asset Management Indianapolis Power & Light September 11, 2013
- 2. Page 1 Executive Summary Problem Statement IPL implemented the Meter Mobile Data project in February of 2011, a part of which included the move to Computer-Aided Dispatched (CAD) to schedule and track work crews. 18 months after implementation, five of the six types of work have stopped using the job routing system. This project identifies several reasons for the causes behind inefficient job routing of the CAD system and provides a number of solutions to improve the performance of the system. Problem-Solving Approach A cross-functional team was formed with representatives from the Operations Technology and Metering departments, and led by Brian Kaiser and Nate Dotzlaf. Examination focuses on the performance of the NP order processes since the implementation of Meter Mobile Data and utilizes the Six Sigma DMAIC problem-solving approach to identify opportunities for improvement. First, brainstorming sessions and Ishikawa diagrams were used to establish potential reasons for the discontinued use of the job routing functionality. Next, using 12 months of historical data, Pareto analysis was conducted to identify the largest contributor of errors and R-Charts were used to measure process capability. Additional brainstorming sessions were conducted to identify key process input variables from which we were able to identify several recommendations for improvements. Major Findings and Recommendations A major finding of our project is that the majority of the problems associated with job routing are attributable to legacy route management practices. Historically, mapping districts were allocated based on population density and qualitative worker input. Every year, districts would be reviewed and redistributed based on population changes. As the complexity of this process increased, the redistricting process has been dropped for several years. In addition, the routing system algorithms initially were set based solely on geographic distance whereas mapping districts (and hence billing cycles) followed only loose geographic patterns and allowed only a small time window for work completion. Furthermore, the CAD system was designed to only perform a single optimization pass, rather than utilize more robust multi-pass routing algorithms. It is our position that the inefficiencies inherent in the current system are due to a combination of factors and that no single fix will solve the majority of the problems. That said, changes have been ordered to include the mapping district code in the data given to the routing system and preliminary analysis indicates positive results. Other improvements planned: 1. Re-draw the billing districts using quantitative variables (population density, historical workload, distance) to reduce variability in workload 2. Reduce the overall workload by implementing new technology to reduce the need for cutoffs (phone calls, text alerts, smart meters) 3. Adjust wrench time to reflect historical data 4. Incorporate more robust routing algorithms 5. Re-design processes to encourage collection of process data to promote future improvements
- 3. Page 2 1.0 Improvement Opportunity: Define Phase In February 2011, Indianapolis Power & Light (IPL) went live with an expansion of its Computer-Aided Dispatch (CAD) system into the Metering department. A key part of the project was to implement automated job creation and dispatching for high-frequency, low-variability jobs such as disconnect and reconnect of customer service. The process was outlined in the SIPOC diagram below. Figure 1: SIPOC Diagram The initial results were mixed; crew routes were seen to overlap each other, visit the same location twice in a day, and other obvious inefficiencies. Within a couple months, the automated dispatching had been disabled on all work types except for one: disconnect of customer service (cuts). As a result, much of the expected gains in efficiency from the project were never realized. The goal of this project was to design and develop a complete set of solutions to address root causes behind the inefficiencies and problems in the cut job process. Success was defined as decreasing the average cost of a cut, as well as increasing the accuracy of crew routing (projected number of jobs completed in a day vs. actual). Another metric will be satisfaction and trust in the system by dispatchers and crews, as measured by comments made in feedback sessions. The project was limited to analyzing the NP-07 job process in CAD and related processes in other systems. It was assumed that the recommendations would also help to make automated dispatching an option for other work types. Analysis would involve not only the automated processes but the manual steps involved in creating and dispatching work. The physical completion of the work was excluded from this project.
- 4. Page 3 2.0 Current State of the Process: Measure Phase 2.1 Current Performance Level The team utilized SQL queries from the CAD database to collect all recorded NP-07 dispatches spanning a single year from October 1, 2011 to September 30, 2012. Because the system was implemented in February 2011, the first six months were omitted to avoid learning curve issues and allow for point-to-point comparisons. Initial results yielded 137,613 records; however, it was necessary to remove a large number of records to prevent skewing of the data analysis results. Primary causes for removal: 1. Dispatcher updating of records (11,837): Dispatchers often update records for a number of reasons including: worker request, worker error, and end-of-day system adjustments 2. Test-user account (36,014): A system account (test-user) is utilized for a number of processing tasks and by a number of different users. There is no reliable way to either confirm that the records itself represents a real job or to attribute the results to a particular user. 3. Zero drive time (53,628): The current tracking system allows workers to run a number of routes and log all data entry after the fact. Accordingly, a large number of records showed drive-time durations of zero. Even though plausible reasons exist for this result – for instance several routes at a single apartment complex – the relative importance of this factor in the overall cost and performance measurements necessitated removal. After cleansing, 36,134 total records were deemed fit for further analysis, or 26.3% of the original sample set. 2.2 Identification of Key Variables Key product/Process Output variables: 1. Total Drive Time – a. Measured from either: i. Shop location to first stop ii. Job completion to next stop b. Affected by: i. Distance ii. Traffic Conditions 2. Total Wrench and Reporting Time- CAD routing assumes an average time of 8 minutes per location. a. Total Wrench Time – the total time starting when a cut-off man reaches his destination, completes the cut-off, and return to his truck. b. Total Reporting Time – The total time required for the cut-off man to log service information and accept the next location. 3. Total Job Time – Total time required to complete a single job including drive, wrench, and reporting time. 4. Cost Key product/Process Input variables: 1. Distance to Location
- 5. Page 4 a. Distance can vary a great deal but the CAS system creates requests based on billing districts (20 sections of the service territory). In general, this somewhat limits the total daily driving distance, but because the cut-off crews service a large geographical area, first and last segments can involve much longer distances. Other factors that can affect driving distance are: i. High risk accounts: preference is placed on completing these jobs despite longer driving distances ii. Socio-economic factors iii. Housing density iv. Road conditions/routing 2. Cut-off Man Type a. Contractor – Contractors work on an as needed basis and consequently carry a much lower cost per hour. Contractors are limited in what types of cutoffs they are allowed to perform and are thought to be slower due to their unfamiliarity with the routes. However, historically they are much more likely than experienced fulltime Cut-off Men to follow the driving directions given by the CAD system and are permitted to use consumer GPS devices to facilitate their work. b. Fulltime Employee – Fulltime employees are always available to service cutoff requests regardless of demand, resulting in a much higher cost per cutoff. Fulltime employees face fewer restrictions in the type of jobs they can perform and are much more familiar with the city and with the cut-off routes. Figure 2: Cut-off Man Histogram 3. Service Type a. Voltage – Service types range from 120V to 480V and dictate which Cut-off Man is able to complete the job. Contractors are not permitted to service higher (and more dangerous) voltages; these are instead reserved for fulltime employees. However, 120V services represent the majority of cutoff requests. 411 4 2 2892 178155 1 30 2586 716 1905 3473 523 239287 2022 1424 3752 3124 1 1876 1 3565 3066 1213 1078 1610 0 500 1000 1500 2000 2500 3000 3500 4000 Cut-off Man Histogram
- 6. Page 5 + Figure 3: Voltage Type Histogram 4. Rate Code a. Rate Code can also indicate the type of service. However, we saw that the large majority (88%) were standard residential services (RH and RS). Figure 4: Rate Code Histogram 2.3 Identification of Target Performance Levels or Project Goals. • Introduction of MES routing will lower the average cost per cut for contractors and average cost per cut for full time employees. (Employee ratios are beyond the scope of this project) • Decrease the DPMs for driving time (estimated vs. actual time) • Changing estimated wrench time will create more accurate routing – would allow for better planning and future performance evaluation 528 426 35082 13 82 1 2 0 5000 10000 15000 20000 25000 30000 35000 40000 Voltage Type Histogram 2 3 4 3074 3 2 10225 4 21489 46 9 6 2 68 1197 0 5000 10000 15000 20000 25000 CB EH ES RC RD RE RH RI RS RT RU RV RW SH SS Rate Code Histogram
- 7. Page 6 3.0 Analysis and Findings: The Analyze Phase Workflow Analysis To begin to analyze the cut-off process, a workflow diagram of the entire process was developed: Figure 5: NP process diagram
- 8. Page 7 It revealed that while each job was handled almost identically, the dependencies upon various computer systems to deliver accurate data were high. Next, brainstorming sessions involving the project team and key stakeholders were held to identify potential causes of cut jobs taking longer than estimated. Figure 6: Cause and Effect
- 9. Page 8 This level of analysis showed the scale of the requirements to complete a cut job. For this project, emphasis was placed on the Methods, System, and Job Routing branches of the diagram. Driving and Wrench Time Analysis Pursuing the System and Job Routing areas, the first area targeted for analysis was the accuracy of the computer-generated routing in the CAD system. If data like estimated driving time, estimated wrench time, etc. are inaccurate, then the routing will be inaccurate even if the other aspects of the process are correct. Sample data was analyzed in several directions: union vs. non-union personnel; by crew; by area of service territory; by day; and against common mapping products such as Google Maps and Bing Maps. The first stage of analysis verified driving times generated by the CAD system against those generated by Bing Maps. The assumption was made that a professional tool, Bing Maps, was the most accurate driving time estimate system available due to the resources and scale of Microsoft. Results showed that CAD-generated driving times were sufficiently accurate (less than 1% variance) when compared to Bing Maps and there were no identifiable factors of error found. Therefore, the team concluded that drive times assigned by the CAD system when dispatching work to crews should be an accurate estimate of the time needed to complete and would not contribute significantly to total drive-time errors. The second stage of analysis was determining the average wrench time. An estimate of 8 minutes was set at the initial configuration of the system. After over two years of using the system, the data pulled should be an accurate representation of actual wrench time needed to complete cut jobs. Figure 7: Xbar-R Analysis of Reported Wrench Time 2101189116811471126110518416314212111 30 20 10 0 -10 Sample SampleMean __ X=3.00 UCL=12.68 LCL=-6.67 2101189116811471126110518416314212111 80 60 40 20 0 Sample SampleRange _ R=3.64 UCL=11.88 LCL=0 1 1111111111111 11 111 1 1 1 111111 1 1 1 1 1 1 1111111 1 1 1 11 1 1 1 111 1 1 1 111 1111 11 1111 1 1 111 1 1 11 1 1 11 1 11 1 1 1111 1 1 1 1 1 1 11111 111 1 11 1 11 1 11 1 111111 11 11 1 1 111 1 1 1 1 11 1 1 1 1 1 1 11111111111111111111111111111111111111111111111111111111111111111111111111111 11 11 1111111111111111111111111111111111111111111111111 111 1 1111 11111 111 1111111111111111111111111111111111111111111111111111111111111111111111111111 1111111 11111111111111111111 11 11 1 1 1 1 11 1111 1 111 1 1 11111111111 11 1 1 11111 1 11 11 1 1 11 1 111 11111111111111111111111111 1 1 1 1 1 1 111 11 1 1 1 1 1 111111 111 1 1 1 1 11 111 1 1 1 111111111 1 111 1 1 1 1 1 1 1 1 1111 1 1 1 11111 11111111111111111 1 111111111111111111111111111111111111111 1 1 1 111111111 1 1 1 11 1 11111111111 1 11 1 11 11 1 1 111 1 1 1 11 1 1 1 1 111 11 111 1 11 1 1 11 1 11111 1111 1111111 111 1 1 1 11111111 1 1 111 1 1 1 1 1 1 1 11 1 1111 1 1 1 1 1 1 1 111111 1 11 1 1 11 1 11 1 1 1111111 1 1 1 1 1 11 1 1 1 1 11 1 11 1 1 1 1 11 1 1 1 1 1 11 1 1 1 11 1 11 11 1 11 1 11111 1 1 1 1 1 1 1 1 1 1 1 11 1 11 1 1 1 1 1 1 111 1 1 1 11 1 1 1 1 1 1 111 1 1 1 1 111111111111 1111 1 1 11 11 1 11111 1 1 11111 1 11111111111111111111111111111 11 111 1 1 11 11111111111111 11111 11 1 11111111 11111111111 1 111 1 11 1 1 1 1 1 1 1 1 1111 1 1111 1 111 1 1 1 1111 1111 1 1 11 11 1 11 1 11 111111 11 111111 111 11 1 111 1 1111111 11111 1 1 1 11 1 11 111111 11 1 1 1 11 1 1 1 1 1 1 1 1 111 1111 1 1 1 1 111 1 11 11 1 1 11 11 11111111111111 1 111111 11111 1 1111111 1 1 1 1 11 1 1 1 11 1 11 1 1 11 1 1 1 1 111 11 11 1 1 1 1111 111 1 1 1 1 11 1 1 1 1 1 1 1 1 11 11 111 1 11 1 1 1 1 1 11 1 1 11 1 111111 1 111 1 11 11 1 1 111 1 1111111111 1 1 111 111111 1 1 1 1 1 1 1 1 1 1 11 1 1 11 1 1 1 1 11 1 1111 1 11 11 1 1 1111 11 1 1 1 1 1 1 1 1 11111111 1 1 Xbar-R Chart of ReportedWrenchTime Tests performed with unequal sample sizes
- 10. Page 9 Through the analysis, we determined that an average of 3.00 minutes was required to complete a cut-off job. However, with an R value of 3.64, it’s clear that the process measured through wrench time has a very low capability. Therefore, the recommended estimated wrench time will need to take this variability into account. Job Routing Through interviews with crew members and administrators, the team learned that much of the frustration around inaccurate job routing stemmed from the switch from MES-based routing in the paper method to hub-and-spoke routing in the computerized method. MES is a system whereby each service address is given a unique routing number. The unique number is a combination of billing district, route number, and address. This system was developed to route the meter readers as efficiently as possible in the era of paper. Interviews with stakeholders indicated a general consensus that MES-based routing prevented route overlap and distributed job more evenly – and was consequently superior to the CAD system. To test this theory, simulations were run comparing each method with historical samples of jobs. Figure 8: Job Distribution using hub-and-spoke routing method
- 11. Page 10 Figure 9: Job Distribution using MES-based routing The MES-based routing yielded much better grouping of jobs on each crew than the hub-and-spoke routing method, with a result of less total driving time. In addition, the jobs selected for dispatch were closer together, also helping to lower the total driving time. All of our test cases suggested that the MES-based routing will return either the same or more efficient routing than the hub-and-spoke routing. Workload Variability A key data variable – that wasn’t apparent during early analysis – that emerged out of subsequent analysis was the number of customers in each billing district. Historically, when there were physical meter readers who visited each residence every month, billing districts were adjusted each year to help eliminate variance caused by population changes. Since the introduction of Automated Metering Reading in the early 1990s balancing the billing districts became less important and yearly adjustments were forgotten. Further review shows that after 20 years, the number of customers in each billing district has changed dramatically:
- 12. Page 11 Figure 10: Customers per Billing District Because the balance of the billing districts, in terms of population density, geographic location, and socio-economic factors is no longer consistent with the rigid schedule set by monthly billing cycles the system will inherently have a high degree of process variability. Furthermore, unstable workload cycles place a burden on managers who must constantly balance workload and available labor on a daily or weekly basis, the result is a surplus of labor on some days and a shortage on others. 19000 21000 23000 25000 27000 29000 31000 33000 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 Customers Billing District
- 13. Page 12 4.0 Recommendations: The Improve Phase Through several group meetings, a total of six options were identified as possibilities to improve the cut process. A. Route using MES address from CAS This recommendation, involving minor modifications in CAS and CAD, will switch from the hub-and-spoke dispatching to a sort using the MES address. The expected result will be jobs dispatched in the same manner as before CAD was implemented. While we don’t anticipate large performance improvements, the business requirement of completing as many jobs as possible in a billing district before moving to the next district will be better met. B. Use ladder and access data from CAS to differentiate jobs While ladder and access information is currently available for some premises, they are not keyed upon in the current dispatch process. This recommendation will create a new work order type in CAD based on known ladder access issues. Jobs requiring a two person ladder crew will only be dispatched when an appropriate crew is available. In addition, a new work order type in CAD will indicate when a job requires special access from a customer, making it more apparent that the customer will need to be contacted prior to beginning the job. This recommendation will require more significant development and testing time on both the CAS and CAD sides. C. Route using independent vehicle routing program This recommendation will involve adding a third-party Vehicle Routing program to the work order creation process. This program will replace the hub-and-spoke dispatching in CAD with a more efficient, multiple-pass algorithm that will minimize driving time and maximize work time for each day’s jobs. Significant improvements on the current system are expected, but adding another system to the process will increase development, testing, and maintenance needs. D. Adjust expected times according to historical data With over two years of data now available, analysis has shown that the wrench and reporting times estimated by the CAD system could be more accurate. This is a low-risk, low-gain recommendation, but will result in dispatching a more accurate number of jobs to each crew. E. Integrate communication processes with cut process While the initial focus of this project stemmed from inefficiencies with the cutoff process, it is our opinion that preventing the dispatch of a work crew in the first place offers a higher ROI. In fact, several programs such as TeleVox notifications and early adoption of remote cutoff meters are proceeding and show promise. Additionally, web based texting services such as Twilio offer a low cost way to increase customer interaction. Overall, the group feels that a more holistic analysis of customer delinquency and the steps leading to cutting service is needed. However, these approaches could require significant analysis and possible investment.
- 14. Page 13 F. Re-implement regular balancing of the billing districts Balancing of the billing districts, in terms of population density, geographic location, and socio-economic factors would reduce a large amount of variability in the process, resulting in a more consistent workload. Further benefits could include increased regularity in labor forecasts and cash flow cycle improvement. Since recommendation F was already implemented by the Metering group after discovery, it was not evaluated. The remaining recommendations were evaluated using six weighted categories: 1. Easy to Implement (100%) 2. Easy to Maintain (50%) 3. Significant Impact (75%) 4. Minimal Cost and Time (50%) 5. Senior Management Support (100%) 6. Solution will be Automated (75%) The results are as follows: Figure 11: Solution Ratings As shown, the two recommendations are to pursue routing using MES order (16.75) and to adjust wrench time according to historical data (17.125), in addition to restarting the billing district adjustment practice. 5.0 Monitoring and Control: The Control Phase The first recommendation, routing using MES order, was implemented in Production on August 9, 2013. After a week of data, it is suggested that the second change, adjusting the estimated wrench time, go into effect. Re-instating the billing district adjustment process is ongoing, but requires involvement of other groups and thus will take several weeks to accomplish. Control of the process will be maintained through weekly and monthly reporting. The metrics of this project were defined as the average cost to complete a cut job, and the accuracy of crew routing (number of jobs scheduled vs. actual number of jobs completed). The expected results are: 1. Reduce average cost of a cut by 2% 2. Achieve 90% accuracy for crew routing A format for the suggested reports can be found in Appendix A.
- 15. Page 14 To continue the process of improvement, complete analysis should be scheduled in six months. In addition to reviewing the effectiveness of the implemented improvements, the other options identified above should be re-evaluated at that point. 6.0 Conclusion This project was initiated to address problems in the automated routing of service cut jobs at Indianapolis Power and Light. Inefficiencies in routing such as route overlap were undermining the improvements expected with the new automated dispatching system. Each step of the process was analyzed using Six Sigma tools and methodology. Most effort was focused on the Methods, System, and Job Routing categories of issues identified in the cause and effect diagram. Out of a potential five options identified, two were chosen to improve the process: use of the MES order in job routing; and adjustment of estimated wrench time according to historical data. Expected results of (1) a 2% reduction in average cost per cut, and (2) 90% accuracy for crew routing were set and will be monitored through weekly and monthly reports. The final result was a combination of small improvements to the process with detailed identification of further advancement opportunities. As a vital process to IPL’s business, even minor refinements can yield a significant gain in the bottom line.
- 16. Page 15 Appendix A: Reporting samples Average Cost of Cut by Week Ending Rolling over previous 9 weeks Accuracy of Driving Estimates Rolling over previous 9 weeks $19.00 $19.20 $19.40 $19.60 $19.80 $20.00 $20.20 $20.40 $20.60 $20.80 $21.00 Average Cost per Cut 80% 82% 84% 86% 88% 90% 92% 94% 96% 98% 100% Accuracy