The Challenge
An energy provider's call centre team spent 2 days per month on manual Excel-based forecasting for staffing decisions. With 30-40% forecast error, they chronically over-staffed (wasting budget) or under-staffed (poor customer experience). Static monthly forecasts quickly became outdated, and the team couldn't optimise intra-day staffing across shifts.
Our Solution
6-week AI Value Showcase delivering production-ready ML forecasting system with two-model architecture:
Model 1: Daily Demand Forecasting
Ensemble ML approach (SARIMA, Prophet, XGBoost) predicting total call volume per day with confidence intervals.
Model 2: Intra-Day Shape Distribution
Time-series model distributing daily volume across 15-minute intervals, enabling tactical shift optimisation.
Rapid Delivery Phases:
- Discovery: 2+ years of historical data extraction, feature engineering (seasonality, weather, billing cycles)
- Model Development: Dual-model training with ensemble approach and confidence intervals
- Production Build: Automated pipeline with visualization dashboard
- Pilot & Deployment: Testing with planners, refinement, full team roll-out
Results
Time Reduction
From 2 days to <10 minutes per forecast
Accuracy Improvement
MAPE reduction vs. manual forecasts
Staffing Efficiency
Reduced overstaffing while maintaining service
Key Impact
- Workforce planning team shifted from forecast production to strategic analysis
- 15-minute interval granularity enabled precise shift optimisation
- Executive leadership gained confidence in AI/ML—tangible proof for broader adoption
- Dynamic forecast updates (daily vs. monthly) enabled response to changing conditions
- 6-week delivery demonstrated value of rapid AI deployment vs. lengthy projects