Scheduler and Appointment Management

The scheduler is the optimization core of WFM. It assigns tasks to resources and time windows while balancing feasibility and cost.

1. Scheduler Architecture

tSM Scheduler is based on OptaPlanner (AI constraint solver) and uses a VRP-TW planning approach.

Primary objectives:

• satisfy hard constraints (skills, time windows, dependencies, due dates),
• optimize soft constraints (travel, workload balance, cost, preferred policies),
• support continuous replanning as demand and availability change.

2. Routing and Traffic Services

Routing/traffic data is used for realistic travel-time estimation.

Design points from source specification:

• distance matrix supports up to large planning sets (up to 20k origins/destinations in scheduler context),
• offline traffic profiles are maintained in 24 hourly slots x 7 days,
• an average traffic profile is used for quick term proposals,
• destination buffers are precomputed (30 km default, optional 50 km for long-distance cases),
• fixed matrix fallback supports resilience for long-distance estimation.

Supporting map capabilities include geocoding, reverse geocoding, route display, POI search, and custom layer overlays.

3. Appointment Term Negotiation

Appointment negotiation requires near-real-time candidate generation.

Target from source material:

• offer multiple valid terms in approximately 5 seconds.

Reference timing split:

Step	Typical Time
Routing-time calculation	~1 s
Candidate solution copies	~1 s
Constructive heuristics	~2 s
Optimization pass	~2 s

Configurable appointment strategies:

• strict optimization during negotiation,
• balanced load distribution,
• quota-driven work type allocation,
• reserved emergency gaps.

4. Constraint and Performance Considerations

Vehicle routing is NP-hard; practical operation balances:

• number/complexity of constraints,
• response-time KPI,
• solution quality.

Operational guidance:

• validate each new rule for performance impact,
• keep hard constraints minimal and explicit,
• use soft constraints for preference ranking,
• evaluate hardware scaling against business value.

5. Sizing and Partitioning Guidance

Source guidance indicates strongest performance around:

• approximately 10k active tasks per scheduler instance,
• approximately 8-core class machine for solver and move threads.

Partitioning approaches:

• geofencing by territory,
• temporal split (e.g., selected windows),
• hybrid strategy with nearby-selection heuristics.

The best global optimization quality is usually achieved when all related tasks/resources remain in one solver domain.

6. Customer Appointment Flow

Appointment flow is activated when process/task requirements indicate customer coordination.

1. Scheduler computes candidate terms.
2. User/dispatcher reviews costs and details.
3. Selected slot is confirmed and reserved.
4. Process continues with scheduled tasks.

Calendar view with ranked terms:

Appointment term detail:

Additional calendar/offer visualizations:

7. Dispatcher-driven Appointment Negotiation

Dispatcher can trigger appointment negotiation directly from dispatch context.

Task selection in dispatch view:

Offered slots from scheduler:

Slot detail in Gantt context:

Slot confirmation action:

8. Dynamic Questionnaire Before Scheduling

WFM can execute configurable pre-appointment questionnaires.

Use cases:

• gather missing technical/customer context,
• branch by answers,
• update workforce order before final scheduling.

Questionnaire examples:

9. Operational Controls

• maintain separate planner modes for standard and emergency operation,
• use dry-run simulation before committing major dispatch interventions,
• tune term-negotiation and full-optimization weights independently,
• monitor KPIs: term response time, due-date hit ratio, travel cost, and replanning churn.