AI-Powered Automatic Shift Scheduling System for Contact Centers

Contact center operators work in three shifts, and the load fluctuates every 15 minutes. Manually creating a weekly schedule is a pain: it takes 8–16 hours of hourly tweaking, and there are still overtime or underload issues. We, a team of AI engineers with 10+ years of experience in MLOps and schedule orchestration, have developed over 50 systems for contact centers. Our AI-based schedulers solve the task in 5–15 minutes, taking into account SLA, labor law, skills, preferences, and budget. In one project for a network of 400 operators, overtime decreased by 28% and preference satisfaction reached 72%. Order a preliminary analysis of your project — we will evaluate it in 2 days.

How does the AI system collect data?

The scheduler uses load forecasts in 15–30 minute intervals, HR system data (contracts, vacations, skills, preferences), and labor law and SLA constraints. All data is imported from WFM systems.

Optimization methods

We use a combination of Integer Linear Programming (ILP) with solvers CBC and Gurobi for fixed horizons, Constraint Programming (CP-SAT) for complex hard constraints, and Reinforcement Learning for adaptive adjustments. CP-SAT from OR-Tools outperforms ILP on problems with logical conditions — for example, "if night shift, then next day shift after 24 hours" — without linearization. In our projects, CP-SAT reduces schedule cost by 5–10% under the same constraints.

Why CP-SAT outperforms ILP?

Constraint Programming SAT allows expressing logical conditions without linearization, simplifying the model and speeding up computation. Additionally, CP-SAT is built into OR-Tools, providing a convenient API for combinatorial problems. For fairness, we use the Gini coefficient for undesirable shift distribution — CP-SAT supports such constraints natively.

How does the system handle operator preferences?

A technical solution for fair scheduling includes:

• Uniform distribution: each operator gets ±5% of night, weekend, and holiday shifts
• Preference satisfaction: target over 70%
• Bidding system: operators bid on shifts, the scheduler maximizes total satisfaction while maintaining SLA

Preference satisfaction vs. cost minimization is a multi-criteria problem. The dashboard displays a Pareto front with different trade-offs. The bidding system is implemented using a weighted maximum matching algorithm, guaranteeing Pareto optimality.

Scheduling process

Horizon: 1–4 weeks.

Pipeline:

1. Import load forecast from WFM
2. Load HR data
3. Generate a set of possible shifts
4. CP-SAT optimization
5. Post-processing: fairness check
6. Publish schedule

Intra-day adjustments: When an operator is unexpectedly absent, an RL agent trained on historical replacement data finds a replacement in seconds. This allows reacting to sick leaves, lateness, and other emergencies.

Mobile app for operators

Operators can view schedules, request shift swaps, and time off through the app. Supervisors spend 60–70% less time on organizational issues. The system sends push notifications about upcoming shifts and changes.

Metrics

• Build time: < 10 min for 200 operators
• SLA coverage: > 98% of intervals within ±10% of target
• Overtime reduction: 15–25%
• Preference satisfaction: > 65%
• Stability: < 15% changes within 48 hours
• Fairness: Gini coefficient for undesirable shifts < 0.2

What is included in the work?

• Analysis of current scheduling and requirements gathering
• Development of mathematical model (ILP/CP-SAT/RL)
• Integration with WFM, HR, and mobile apps
• Testing on historical data and A/B pilot
• Documentation, training, 1-month warranty support

Get a detailed savings estimate for your contact center. Contact us for a consultation — we will evaluate your project in 2 days. Find out how an AI scheduler can reduce your operational costs.

Timelines: basic scheduler with ILP/CP-SAT — 6–8 weeks. Full system with mobile app and intra-day adjustments — 4–5 months.

When does a time series forecasting model fail in production?

The CFO requests a quarterly sales forecast. An analyst builds SARIMA on three years of data, achieves MAPE 8.3% on the test set, and deploys. Two months later, the metric in production jumps to 23%. The root cause: the model was trained on pre‑COVID data, tested on a stable period, but production hit a promotion and supply chain disruption. Data leakage plus distribution shift—perfect notebook numbers, a broken forecast in reality. We have seen this pattern dozens of times across retail, fintech, and IoT. Our team has delivered more than 50 forecasting projects over 5+ years.

Incorrect cross-validation. Standard train_test_split for time series creates data leakage: the model sees future values during training. The correct approach is TimeSeriesSplit or walk‑forward validation with an expanding window.

Multiple seasonality. Hourly electricity consumption has three seasonalities: daily (24h), weekly (168h), yearly (8760h). SARIMA handles only one. Prophet can handle multiple but scales poorly to thousands of series.

Missing values and anomalies. A missing sensor reading is information (the sensor turned off), not NaN. Linear interpolation destroys this signal. Proper handling depends on the missingness mechanism.

Cold start. A new SKU in a 50,000‑item assortment has no history, yet a forecast is needed. Standard approaches fail; cross‑learning or feature‑based methods are required.

Why is model selection critical for your data?

Prophet (Meta) – a solid start for business data with clear seasonality and holidays. Fast setup, interpretable, built‑in outlier detection. Fails on irregular patterns and does not scale beyond ~10k series without parallelization.

Gradient boosting on features (LightGBM, XGBoost) – often underestimated. Engineer lags (t‑1, t‑7, t‑28), rolling means, day‑of‑week, holidays. The model trains on all series simultaneously, solving cold start via transfer learning. MAPE in retail often beats neural nets with proper feature engineering.

TFT (Temporal Fusion Transformer) – a transformer designed for interpretable forecasting with covariates. Built‑in variable selection, temporal attention, quantile outputs. Available in pytorch‑forecasting. Requires ~10,000+ records per series for stable training.

PatchTST – splits the series into patches (like ViT for images), capturing local patterns better than classic transformers. Excellent for long‑horizon forecasting (96–720 steps ahead).

N‑HiTS, N‑BEATS – attention‑free neural architectures, faster than TFT, competitive accuracy. N‑BEATS won the M4/M5 benchmarks for tasks without covariates.

How do we deploy TFT in production?

A typical pipeline via pytorch‑forecasting:

training = TimeSeriesDataSet(
    data,
    time_idx="time_idx",
    target="sales",
    group_ids=["store", "sku"],
    min_encoder_length=max_encoder_length // 2,
    max_encoder_length=max_encoder_length,  # 120 days
    min_prediction_length=1,
    max_prediction_length=max_prediction_length,  # 28 days
    static_categoricals=["store_type", "category"],
    time_varying_known_reals=["price", "promo_flag"],
    time_varying_unknown_reals=["sales"],
    target_normalizer=GroupNormalizer(groups=["store", "sku"], transformation="softplus"),
)

A common mistake: the default target_normalizer (StandardScaler) breaks predictions for series with zero values (no sales on weekends). GroupNormalizer with transformation="softplus" is the correct choice for count data.

Case study: retail demand forecasting

A chain of 120 stores, 8,000 SKUs, 28‑day forecast horizon. The original system: SARIMA per series, MAPE 18.4%, retraining cycle – 6 hours. We replaced it with TFT on PyTorch + pytorch‑forecasting: a single model for all series, MAPE 11.2%, retraining – 40 minutes on an A10G. Feature importance via variable selection revealed that day_before_holiday influences more than the holiday date itself. Annual savings on inference alone exceeded $50,000.

Step‑by‑step configuration

1. Data collection and preparation. Handle missing values (mark NaN, interpolate only for technical failures), aggregate to required frequency, engineer covariates (holidays, promotions, prices).
2. Create TimeSeriesDataSet. Set group_ids (store + SKU), time index, forecast horizon. Choose target_normalizer based on target distribution.
3. Train a baseline. Prophet or LightGBM first – to understand complexity.
4. Train TFT. Use TemporalFusionTransformer with loss=QuantileLoss(), tune learning rate and hidden layer sizes.
5. Validate and interpret. Walk‑forward test, analyze variable selection, build attention heatmaps.

How to properly evaluate forecast quality?

RMSE alone is misleading – it over‑penalizes large values. Our standard set:

• MAPE – interpretable, unstable near zero.
• sMAPE – symmetric, avoids division by small numbers.
• MASE (Mean Absolute Scaled Error) – normalized relative to a naive seasonal forecast, ideal for comparing series of different scales.
• Pinball loss – for probabilistic forecasting, inventory management.

We guarantee a model card with these metrics on the validation set and walk‑forward results on at least 6 months of history.

What deliverables do you receive?

• Documentation of chosen architecture and hyperparameter rationale.
• Reproducible training and inference pipeline (Docker + CI/CD + Airflow/Prefect).
• Committed code with unit tests for key components.
• Team training: retraining, output interpretation, deployment of new versions.
• 3 months of post‑delivery support (consultations, bug fixes, fine‑tuning).

The model is deployed via FastAPI or Triton Inference Server. Retraining is scheduled (e.g., weekly) via Airflow with drift validation and automatic rollback if metrics deteriorate.

Process and timeline

We start with EDA: visualization, ADF test, STL decomposition, analysis of missing values and outliers. This takes 2–3 days but often reveals systemic data issues that block forecasting. Then we build a baseline (naive seasonal, Prophet), engineer features for LightGBM, and select a neural architecture if needed. Walk‑forward validation with a realistic horizon. Deployment via API with automatic retraining scheduled via Airflow or Prefect.

Timeline: MVP forecast on one data type – 3–6 weeks. Hierarchical forecasting system with automation – 2–5 months. Cost is calculated individually based on data volume, number of series, and required accuracy.

Our team consists of certified ML engineers (AWS ML Specialty, GCP Professional ML Engineer) with 5+ years on the market and over 50 completed forecasting projects. Contact us for a free analysis of your data – we will assess the task and provide initial recommendations within 1–2 days. Request a consultation to ensure your forecasts work in production, not just in a notebook.