Results & Reporting

This page captures the modeling outcomes, workforce optimization feasibility, and the final dashboard deployment used to translate these forecasts into operational staffing decisions.

Executive Summary

  • Forecasting Excellence: The Short-Term forecasting model (< 7 Days) achieved highly accurate results (WMAPE: 3.11%), driven predominantly by immediate autoregressive signals. The Long-Term ensemble (>= 7 Days) effectively captured macro-seasonal trends with a WMAPE of 13.12%.
  • Optimization Feasibility: The abandonment-aware workforce optimizer successfully found feasible staffing solutions for 100% of the tested half-hour intervals (April 14-18, 2025), keeping average wait times under 5 seconds while maintaining occupancy targets.
  • Operationalization: An end-to-end Workforce Management Dashboard has been deployed, allowing managers to monitor SLA compliance, run performance simulations, and interactively schedule shifts based on model outputs.

Metrics Dashboard

Date Model/Version Dataset Metric(s) Result Notes
2025-11-15 Short-Term LGBM (< 7 Days) Sequential Test WMAPE / R² 3.11% / 0.9979 Driven by lag features & weekly cyclicality.
2025-11-15 Short-Term LGBM (< 7 Days) Sequential Test MAE / RMSE 28.55 / 58.88 Highly sensitive to preceding 30-min volume.
2025-11-15 Long-Term Ensemble (>= 7 Days) Sequential Test WMAPE / R² 13.12% / 0.9706 Relies on historical averages & YoY baselines.
2025-11-15 Long-Term Ensemble (>= 7 Days) Sequential Test MAE / RMSE 120.25 / 220.71 Captures macro-seasonal/operational states.
2025-04-18 WFO Binary Search April 14-18 Vol Feasibility 100% Found solutions for all 120 half-hour intervals.

Workforce Optimization Results

The optimizer was evaluated over the work week of April 14, 2025, through April 18, 2025. Under the default constraint set (SLA target = 0.80, Wait Time limit = 60s, Occupancy cap = 0.85):

  • SLA Compliance: Mean predicted compliance was 99.85% (minimum interval-level SLA of 90.66%).
  • Wait Times: Mean predicted wait time was 0.07 seconds (maximum interval-level wait time was 4.19 seconds).
  • Agent Occupancy: Tightly bounded between 83.47% and 84.97%.

Constraint Sensitivity Analysis The optimizer’s sensitivity to the occupancy constraint was tested by varying the occupancy cap:

  • Relaxing the cap from 0.85 to 0.90 reduced total agent-slots required by ~5.6%.
  • Tightening the cap from 0.85 to 0.75 increased staffing requirements by ~13.3%.

Workforce Management Dashboard & Visualisations

Interactive dashboard components translate our forecasts and queueing models into actionable staffing decisions.

1. Main Dashboard

The dashboard provides a high-level overview of operational performance, forecasted demand, and workforce activity for rapid situational awareness.

Main dashboard interface Figure 1: Main dashboard interface showing weekly performance metrics, demand forecast, and agent productivity summary.

Weekly Overview & Forecasts Dynamically calculates week-to-date SLA, average wait times, occupancy, and total calls. It displays a seven-day forecast bar chart differentiating past, current, and future days.

Weekly call demand forecast Figure 2: Weekly call demand forecast visualized across the seven days of the selected week.

Top Performing Agents Highlights individual productivity with real-time status, calls handled, handle time, and utilization rates.

Top-performing agents table Figure 3: Top-performing agents table displaying real-time status and productivity metrics.

2. Performance Simulator

An interactive environment for evaluating staffing requirements under different operational targets, integrating demand forecasts with Erlang-based queueing calculations.

Target Metrics Configuration Managers can use slider controls to adjust SLA targets, maximum waiting time thresholds, and occupancy limits to evaluate alternative strategies in real-time.

Performance simulator controls Figure 4: Performance simulator controls allowing managers to adjust SLA, waiting time, and occupancy targets.

Daily Demand and Staffing Forecast Visualizes predicted call volume (area chart) and required staffing (bar overlay) at a 30-minute granularity over a 24-hour period.

Daily demand and staffing forecast visualization Figure 5: Daily demand and staffing forecast visualization at 30-minute granularity.

3. Shift Scheduler

Translates predicted staffing requirements into operational shift assignments while respecting workforce constraints.

Overview & Metrics Users select the week/day via an interactive calendar to view coverage summaries, available agents, staffed time slots, and schedule completion status.

Shift scheduler overview Figure 6: Shift scheduler overview showing week and day selection, staffing coverage summary metrics, and schedule editing controls.

Interactive Timeline & Agent Assignment A 24-hour timeline allows managers to drag blocks to create shifts. Managers can drag agents directly from the availability lists onto the timeline. Schedules automatically lock once the date begins.

Interactive shift scheduling timeline Figure 7: Interactive shift scheduling timeline showing agent shifts and required staffing levels.

Available and unavailable agent lists Figure 8: Available and unavailable agent lists used for shift assignment.


Narrative Reports & Stakeholder Communication

  • Methodology Handover: The comprehensive modeling pipeline and queuing theory mathematics have been documented in the Methods tab.
  • Tooling Access: The UI is currently deployed for Call Center Shift Managers.
  • Status Updates: Weekly syncs are held with the Operations steering committee to review forecasting accuracy and schedule adherence.

Next Steps & Open Questions

  • Model Drift Monitoring: Establish automated alerts if the Short-Term model’s WMAPE degrades beyond 5% for three consecutive days.
  • User Adoption: Track manager engagement with the Performance Simulator to ensure the binary search outputs are trusted over legacy heuristic scheduling.
  • Feature Expansion: Investigate incorporating real-time shrinkage (unplanned absences) into the immediate intraday autoregressive forecast.

Back to top

© 2025 UC San Diego - Data Science Capstone

This site uses Just the Docs, a documentation theme for Jekyll.