Workforce optimization (WFO) is how a contact center matches staff to demand and measures and improves the quality of their work. It usually combines forecasting, scheduling, quality management, performance management and analytics.
IBM defines contact center WFO as a strategy that combines technology and people management. Workforce management (WFM), which forecasts demand and builds schedules, is one part of it.
What WFO covers
Forecasting. Expected calls and handle time by month, day and half hour. A 2003 review of call center research by Noah Gans, Ger Koole and Avishai Mandelbaum describes the common method. Planners take monthly totals from history and split them by weekday and time of day.
Staffing and scheduling. A queueing model, usually Erlang C, turns each interval's forecast into agents needed. Scheduling then builds shifts to cover the day, and rostering assigns people to them.
Gross-up for shrinkage. The review notes that planners add time for breaks, training, meetings and absence on top of agents on the phones.
Intraday management. Comparing actual calls and staffing with the plan, and moving breaks or training when the day changes.
Quality management. Scoring a sample of calls against a scorecard. The review gives classic questions, such as whether the agent used the customer's name. Speech analytics can score far more calls than a sample.
Performance management. Coaching and targets built on these measures.
How AI agents change the staffing math
A worked example shows why fewer calls do not always mean proportionally fewer agents. Take 100 calls in a half hour with a three-minute handle time. To answer 80% within 20 seconds, Erlang C asks for 14 agents.
Now an AI agent resolves 30 of those calls. If the other 70 still take three minutes, the target needs 10 agents.
Routine calls are often the short ones, though. If the remaining 70 average four minutes, the target needs 13 agents, one fewer than before.
Pool size matters as well. The review explains that large queues need proportionally fewer extra agents than small ones. A smaller human queue loses some of that efficiency.
Forecast two streams. It helps to forecast AI-handled and agent-handled volumes separately, each with its own handle time.
Plan for handoffs. Calls that move from the AI agent to a person arrive mid-conversation, with context. Their handle time may differ from a fresh call.
Re-run the model. As the AI agent takes on new call types, the mix of human calls changes. Staffing based on last quarter's handle time can drift.
Quality management when AI agents take calls
AI conversations need quality review too. Transcripts, the decisions taken, and every amount or disclosure given can be checked against the rules that should have applied.
An AI audit trail records what the agent said and decided, which makes that review possible. AI quality monitoring can check AI conversations at scale, leaving human reviewers more time for complex calls and coaching.
Frequently asked questions
What is the difference between WFO and WFM?
Workforce management (WFM) covers forecasting, staffing, scheduling and intraday management. WFO includes WFM and adds quality management, performance management and analytics. Vendors draw the line in slightly different places.
What is shrinkage in workforce management?
Shrinkage is the share of paid time when scheduled agents cannot take calls: breaks, training, meetings, coaching and absence. Planners add it on top of the number Erlang C gives. Agent schedule adherence reports from the ACD help estimate it.
How do you forecast call volume for a contact center?
Most teams start from history, split monthly totals by weekday and half hour, and adjust for known events such as billing cycles or campaigns. Handle time needs its own forecast, since it changes with the call mix. Forecasts are worth checking against actuals every week.
How do you plan staffing when an AI agent handles part of the calls?
It helps to forecast the AI-handled and agent-handled streams separately, with separate handle times. The staffing model then runs on the agent stream alone. The remaining calls are often longer and more complex, so the plan is worth reviewing as the AI agent's scope grows.
How we approach this at Zowie
Zowie's AI agents work on voice, chat, and email, with over 100 million conversations a year across our clients. On every channel, the language model speaks. A separate, deterministic Decision Engine returns every amount, rate, offer, disclosure, and escalation from your rules, with an audit log for your quality team. To try it on your own call mix, you are welcome to book a demo.
Explore: Inbound call center, Automatic call distributor (ACD), Quality assurance, Speech analytics, Call center metrics, Average handle time (AHT)