B. Diferencias de precios, falta de regulación y la tortilla ¿guerra perdida?
VII. RESULTADOS Y ANÁLISIS: RAZONES PARA LA SIEMBRA DE MAÍZ CRIOLLO
1. Santiago y Sandra: el campesinado y el magisterio
We have shown through our steady-state counterfactual analysis that offering callbacks under a variety of loads can increase service quality and may slightly increase or decrease system throughput. While the above results demonstrate the effects of implementing callback technology in steady state, in actuality call centers routinely experience extraordinary spikes in call volume and reductions in service capacity. For example, banking call centers such as the one in our study may experience unplanned spikes in volume due to outages in service channels such as online banking or planned volume spikes due to advertising efforts. Call centers also routinely schedule off-phone activities such as team meetings, training, and team- building exercises for large groups of agents. Furthermore, they often encounter consistent staffing deficiencies during certain periods of the day due to constraints in the agent schedules. The effect of these disruptions is to create temporary periods where system congestion is higher than normal.
Offering callbacks during periods of temporary congestion caused by volume spikes or reduc- tions in service capacity could prove highly beneficial, particularly as a demand postponement strategy. By allowing managers to shift excess demand to periods of excess service capacity, offering callbacks could increase service quality by reducing online queue congestion while also increasing system throughput by serving callers who otherwise would have abandoned. To determine whether these benefits could be realized in this call center, we conduct a transient counterfactual analysis, where we maintain the same constant arrival rate as the steady-state analysis, but temporarily reduce the staffing levels within the simulation. Specifically, we begin our simulation at 9:00 a.m. with 180 servers in the system. Then at 10:00 a.m., 60 servers drop out of the system for a planned off-phone activity. At 10:30 a.m., the 60 servers reenter the system, where they stay available until the simulation ends at 12:00 p.m.19 This half-hour, off- phone activity creates a temporary shortage in service capacity followed by a period of excess
19We describe how we handle the process of servers dropping out of the system for their off-phone activity. At
the beginning of the simulation, we flag the 60 servers who are scheduled to attend the activity. If, at 9:57 a.m., a flagged server is idle, that server immediately drops out of the system. Otherwise, the flagged server drops out immediately after concluding the current call. All flagged servers then reenter the system at 10:30 a.m. In the rare case that a flagged server is on a call that lasts the entire duration of the off-phone activity, the server misses the activity and stays in the system for the entire length of the simulation.
service capacity. Because we are interested in capturing the benefits of postponing demand by offering callbacks, we simulate the system under the Window policy with a 60-minute upper bound (W(60)) and compare it with the performance of the system under the no callback policy (policy N). Under each policy, we simulate the transient scenario 300 times. While we offer to call callers back within 60 minutes, we find that the system actually becomes uncongested within 30 minutes of the end of the off-phone activity under both policies (N and W(60)). Thus, we illustrate the average performance measures only from 10:00 a.m. to 11:00 a.m.
Result 5: Offering callbacks as a demand postponement strategy during tempo- rary periods of high congestion can increase service quality and system throughput.
We present the performance measures of our transient counterfactual analysis in Table 2.5. We first observe that offering callbacks under policy W(60) significantly increases service quality by decreasing the average waiting disutility (AWD) by 54% (from 0.147 to 0.068). We also find that offering callbacks under this policy significantly increases system throughput by reducing the percentage of lost calls (% Calls Lost) from 3.98% to 1.99%. Hence, the demand postpone- ment strategy of implementing policy W(60) indeed simultaneously increases service quality and system throughput. We remark that Ata and Peng (2018) also demonstrate in their setting that offering callbacks during temporary periods of congestion can substantially reduce system cost by prioritizing callers in the online queue while postponing callbacks until the system is less congested.
Table 2.5: Transient Scenario System Performance from 10:00 a.m. to 11:00 a.m. under Policies N and W(60)
Service Quality Measures Throughput Measures
AWT AWT On AWT Off
Policy Stat AWD (sec.) (sec.) (sec.) % Calls Lost % Aban % NACB
N Mean 0.147 80.3 80.3 0.0 3.98% 3.98% 0.0% L.B. 0.141 77.2 77.2 0.0 3.83% 3.83% 0.00% U.B. 0.153 83.4 83.4 0.0 4.12% 4.12% 0.00% W(60) Mean 0.068 112.8 11.3 101.5 1.99% 0.64% 1.35% L.B. 0.065 107.2 10.6 96.3 1.93% 0.61% 1.32% U.B. 0.071 118.3 11.9 106.7 2.05% 0.66% 1.39%
2.7, which plots several average performance measures for each 15-minute interval from 9:30 a.m. to 11:00 a.m. In the upper-left quadrant we plot the average number of available servers under each policy to illustrate the sudden drop in service capacity from 10:00 a.m. to 10:30 a.m. as 60 servers drop out of the system to attend the off-phone activity. In the upper-right quadrant we plot the average queue lengths under each policy. For policy N, this includes only the average online queue length since callbacks are not offered under this policy. For policy W(60), we stack the online and offline queue lengths into one column. Note how policy W(60) maintains substantially shorter average online queue lengths by channeling callers into the offline queue and postponing demand until capacity is available. This reduction in the average online queue length leads to lower average online waiting times, which increases service quality by reducing the callers’ average waiting disutility. In the lower left-hand quadrant, we plot the average number of calls lost by policy, which includes the abandoned calls under policy N and a stacked column of the abandoned calls as well as the calls lost from callers not answering callbacks under policy W(60). We observe a significant reduction in abandonment under policy W(60), which is only partially offset by the callers who do not answer the callback. Hence, policy W(60) increases system throughput by offering callbacks to callers who otherwise would have abandoned. Finally, in the lower-right quadrant, we plot the average server utilization under both policies. Note that policy W(60) leads to higher utilization since the excess demand created by the off-phone activity is effectively stored in the offline queue until excess capacity is available.
Figure 2.7: Various Average Performance Measure during Transient Scenario by 15-Minute Interval from 9:30 to 11:00 under Policies N and W(60)