Scheduling Optimisation for Contact Centres

What problem are we actually solving?

Leaders want stable service levels without overtime or burnout. Planners need rosters that cover volatile demand and complex skills. Agents want fair, predictable shifts. Scheduling optimisation turns a forecast into shifts, breaks, and assignments that minimise cost and fatigue while meeting target wait times. Queueing science explains why misfit hurts: as utilisation nears full capacity and variability rises, waits explode, so even a “small” coverage gap drives outsized pain.¹ Contact-centre research treats scheduling as a constrained optimisation problem with objectives such as service-level attainment, cost, fairness, and preference satisfaction.² ³

What does “good scheduling” mean in operational terms?

Good scheduling converts interval workload into staffed minutes with the right skills, concurrency, and breaks. It protects coverage (supply meets demand by interval), stability (no whipsawing between queues), and wellbeing (reasonable spans, rest, and rotation). Modern WFM stacks start with forecasted arrivals and handle time, transform workload using Erlang C or simulation targets, then solve a shift-assignment problem with constraints: min/max shift length, start-time granularity, break rules, skill qualifications, and fairness goals.³ ⁴ Call-centre surveys show that pairing realistic shrinkage and intraday control with optimised rosters outperforms “manual patching” by a wide margin.⁵

Which optimisation methods actually work for contact centres?

Practitioners use three families of methods and often blend them.

  • Integer/Linear Programming (IP/LP). Formulates coverage as decision variables over candidate shifts and solves to optimality or near-optimality. Column generation accelerates large problems by generating only promising shifts.³ ⁴

  • Metaheuristics. Genetic algorithms, tabu search, and simulated annealing explore big solution spaces with complex constraints (bids, fairness, relief breaks) and yield high-quality schedules fast.³

  • Simulation–optimisation. Uses a queueing simulator to evaluate schedule candidates under variability and routing logic, then searches for the best solution. This is valuable for multi-skill, omnichannel operations where Erlang underestimates variance.²

Academic reviews conclude that hybrid approaches—IP for base shifts, metaheuristics for breaks and fairness, simulation for validation—consistently deliver the best practical results.³ ⁴

How do we translate demand into optimised shifts step by step?

1) Stabilise inputs. Lock a rolling forecast with interval workload, skills mix, shrinkage assumptions, and occupancy targets. Use percentiles, not just means, for volatile queues.¹ ⁵
2) Generate candidates. Create legal shift templates (e.g., 6–10 hours), start times, and break patterns that comply with policy and law.
3) Solve coverage. Use IP to select the minimal-cost set of shifts that meets or exceeds demand by interval, subject to skills and concurrency caps.³
4) Place breaks intelligently. Optimise break timing against the interval profile to avoid simultaneous trough-to-peak transitions. Metaheuristics work well here.³
5) Assign agents. Respect skills, preferences, consecutive-night rules, and fairness constraints such as max weekends.² ⁴
6) Validate in simulation. Check service-level attainment and abandon risk under realistic variability and routing. Fine-tune where the model is optimistic.²
7) Publish and protect. Freeze key intervals, define change windows, and set intraday playbooks for callbacks, break moves, and cross-queue shifts.⁵

What constraints and objectives should be explicit in your model?

  • Coverage: meet target offered load by interval per skill/queue.

  • Cost: minimise paid hours, overtime, and premium differentials.

  • Quality: maintain first-contact resolution by protecting access to the first capable resolver (intent-based routing implications).⁶

  • Wellbeing: limit consecutive late/early turns, enforce minimum rest, cap occupancy bands, spread weekends equitably.

  • Fairness & preferences: support shift bidding, honor seniority bands, and incorporate soft constraints with penalties to avoid brittle rosters.³ ⁴

How to handle multi-skill and omnichannel without overstaffing

Multi-skill routing increases effective capacity but complicates scheduling. The accepted approach is to model skill groups and cross-train to cover correlated peaks. Telephone call-centre studies show that carefully calibrated pooling reduces total staffing while keeping waits stable.² Messaging and email add concurrency; schedules must set safe simultaneous-conversation caps, since aggressive caps inflate errors and reduce FCR. Validate concurrency assumptions in simulation; do not assume linear gains.²

Where do breaks, lunches, and meetings fit in an optimised schedule?

Break placement makes or breaks service. Evenly spaced breaks can create hidden troughs. Better practice staggers breaks across micro-intervals and shifts low-urgency activities into shoulder periods. Literature on “tour scheduling” places lunches away from dominant peaks while meeting regulated rest windows.³ Add a coaching ringfence so learning survives; centres that protect coaching weekly see more stable handle times and lower repeat contacts.⁵

What is shrinkage and how does it flow into scheduling?

Shrinkage is the share of paid time not available to handle contacts: leave, training, coaching, meetings, system downtime, and unplanned absence. Build shrinkage bottoms-up by team and month, split planned vs unplanned, and apply it before optimisation so required heads reflect reality. Understated shrinkage generates chronic service misses and overtime.⁵

How to evaluate schedule quality beyond “meets coverage”

Use a paired scorecard:

  • Leading: interval coverage fit (over/under), schedule efficiency (% paid vs productive time), adherence, and occupancy distributions.

  • Lagging: service level, ASA, abandon rate, FCR, overtime hours, and attrition.
    Studies emphasise aligning scheduling metrics with customer outcomes; chasing average handle time in isolation degrades FCR and inflates repeats, which defeats the roster’s purpose.² ⁷

How intraday optimisation keeps rosters effective after the bell

No schedule survives first contact with reality. Intraday teams compare live arrivals and AHT to plan, then apply the lightest correction first: move breaks, enable virtual-hold callbacks, rebalance skills, and flex concurrency. Evidence shows callbacks above queue thresholds reduce abandonment and perceived wait while smoothing peaks—crucial when staffing is fixed.⁸ Reforecast the remainder of day hourly and protect coaching where possible.

How to build fairness and agent choice without breaking coverage

Preference-aware rostering increases engagement and retention. Metaheuristic schedulers can incorporate soft constraints for preferred days, shift swaps, and compressed weeks with penalties when violated.³ ⁴ Self-scheduling models let agents pick from pre-approved tours up to a coverage cap; simulation confirms whether flexibility degrades service. Many centres use bidding windows and preference weights to balance equity and coverage.

What are the most common scheduling mistakes—and fixes?

  • Planning to averages. Fix by scheduling to forecast plus an error percentile on critical intervals; wait time is nonlinear.¹

  • Underestimating shrinkage. Fix by reconciling planned vs actual monthly and adjusting templates.⁵

  • Ignoring intraday shape. Fix by optimising break patterns and start times to the interval profile, not just totals.³

  • Overloading concurrency. Fix by validating in simulation and monitoring quality/FCR impacts.²

  • Chasing AHT over FCR. Fix by protecting first-capable routing and measuring repeats; resolution beats speed.⁷

A 60-day rollout plan for scheduling optimisation

Days 1–15: Baseline and templates.
Baseline 90 days of interval demand, AHT, shrinkage, and skills. Define legal shift templates, start-time granularity, and break rules.

Days 16–30: Solve and simulate.
Run an IP base schedule; add metaheuristic break placement. Validate with simulation for service level, abandon, and occupancy bands.² ³

Days 31–45: Preference & fairness.
Add soft constraints for bids and weekends; tune penalties to hit fairness KPIs without losing coverage.

Days 46–60: Operationalise.
Publish intraday playbooks (break moves, callbacks, cross-queue moves), adherence coaching, and monthly shrinkage reconciliation. Track schedule efficiency, overtime, and FCR alongside service level.⁵ ⁷ ⁸

What outcomes should executives expect

Expect higher schedule efficiency, fewer peak-hour breaches, and reduced overtime. Expect improved agent satisfaction due to fairer rotations and protected coaching. Expect lower abandonment and steadier FCR as breaks stop colliding with peaks and first-capable coverage improves. These gains come from aligning optimisation, intraday control, and wellbeing—three legs of a stable stool.² ⁵ ⁸

FAQ

What is the fastest lever to improve schedules this quarter?
Optimise break placement against the interval profile and enable callbacks above defined wait thresholds. You will cut peak abandons without adding staff.³ ⁸

When should we use integer programming vs heuristics?
Use IP (often with column generation) to choose base shifts for coverage. Use metaheuristics for break placement, fairness, and preference handling, then validate in simulation.³ ⁴

How do we avoid overstaffing with multi-skill agents?
Model skill groups and pool carefully where demand is correlated. Validate with simulation; multi-skill gains are real but not linear.²

How much shrinkage should we carry?
Build it bottoms-up and reconcile monthly. Apply shrinkage before optimisation so rosters reflect true availability.⁵

Does pushing concurrency always help?
No. Aggressive concurrency raises error rates and harms FCR. Set caps by channel and confirm in simulation before scaling.² ⁷

What metrics prove the schedule is working?
Interval coverage fit, schedule efficiency, adherence, occupancy distribution (leading) and service level, ASA, abandon, overtime, and FCR (lagging). Tie metrics to decisions weekly.² ⁷

Sources

  1. Queueing theory and service variability (Kingman’s result overview) — Wikipedia contributors, 2025, Wikipedia. https://en.wikipedia.org/wiki/Kingman%27s_formula

  2. Telephone call centers: Tutorial, review, and research prospects — A. Mandelbaum, N. Gans, G. Koole, 2003, Manufacturing & Service Operations Management / Operations Research. https://pubsonline.informs.org

  3. A review of workforce scheduling and rostering: classification and computational issues — S. van den Bergh, J. Beliën, P. De Bruecker, E. Demeulemeester, L. De Boeck, 2013, European Journal of Operational Research. https://www.sciencedirect.com/science/article/pii/S0377221713005164

  4. Workforce scheduling: A literature review — A. T. Ernst, H. Jiang, M. Krishnamoorthy, D. Sier, 2004, European Journal of Operational Research. https://www.sciencedirect.com/science/article/pii/S0377221703004090

  5. Workforce Management Best Practices — NICE, 2024, nice.com resource. https://www.nice.com/resources/workforce-management-best-practices

  6. Intent-based Routing in the Contact Center — Genesys, 2024, Genesys Blog. https://www.genesys.com/blog/post/intent-based-routing

  7. First Contact Resolution: Definition and Approach — ICMI, 2008, ICMI Resource. https://www.icmi.com/files/ICMI/members/ccmr/ccmr2008/ccmr03/SI00026.pdf

  8. Optimal scheduling in call centers with a callback option — B. Legros, 2016, European Journal of Operational Research. https://www.sciencedirect.com/science/article/abs/pii/S0166531615000930

  9.  

Talk to an expert