Call Center Workforce Management: A Practical WFM Guide to Forecasting, Scheduling and Adherence

Most WFM guides are written for 500-seat operations with a planning team and a dedicated tool. If you run 10 to 60 agents, you probably have a supervisor doing the rota in a spreadsheet on Thursday afternoon, and a service level that collapses every Monday morning for reasons nobody has time to dig into.

This guide is for you. It walks through the whole cycle with one worked example, from eight weeks of history to a Monday rota, an intraday re-forecast and an adherence score, with every number shown so you can redo it with your own data. If your agents work from home or across sites, the forecasting still applies the same way on virtual call center software.

What Is Call Center Workforce Management?

Workforce management is how you make sure enough people, with the right skills, are available in every interval of the day to meet your service target, without paying for idle time. It is not a single report or a scheduling screen. It is a loop you run every week:

1. Forecast contacts and average handle time for each 15- or 30-minute interval.
2. Staff: turn the forecast into agents needed on the phones, then into people to schedule.
3. Schedule: build shifts, breaks, lunches, training and part-time cover that match the requirement curve.
4. Manage intraday: watch actuals against plan and react when the day departs from the forecast.
5. Review: measure forecast accuracy, adherence and shrinkage, and feed what you learn into next week's forecast.

Skip the review step and the loop never improves. Skip intraday management and a good forecast still turns into a bad day.

How Do You Forecast Call Volume?

For a small or mid-size center, a disciplined forecast from your own history beats any sophisticated model you cannot maintain. You need three things from your phone system: offered calls (answered plus abandoned, not just answered) per interval, average handle time per interval, and at least 8 weeks of both. A year of history lets you add seasonality.

Use offered calls, not answered calls. If 15% of Monday-morning callers hung up, a forecast built on answered calls bakes the understaffing into next week.

Step 1: The weekly base

Take the last 8 weeks of offered calls: 5,480, 5,620, 5,510, 5,700, 5,590, 5,650, 5,540 and 5,710. The total is 44,800, so the weekly base is 5,600 calls. Remove or correct weeks distorted by an outage or a one-off event before averaging.

Step 2: Seasonality

Last year, the same week took 6,100 calls against an average week of 5,700, a seasonal index of 6,100 ÷ 5,700 = 1.07. The forecast for the coming week is 5,600 × 1.07 = 5,992, which we round to 6,000 calls. Without a year of history, skip this step and lean harder on known events: a billing run, a price change, a marketing campaign, public holidays or Ramadan hours. Add these as explicit adjustments you can check afterwards, not as a vague "plus a bit".

Step 3: Day-of-week shares

From the same 8 weeks, work out what share of each week's calls fell on each day. In our example, the center is open Monday to Saturday:

Step 4: Interval curves

Each day has its own shape. Monday usually peaks earlier and higher than Friday, so build a curve per day of the week, not one curve for all days. For Monday, average the share of the day's calls that arrived in each half-hour across the last 8 Mondays. A few of Monday's 20 half-hours, open 08:00 to 18:00:

Forecast handle time the same way. AHT often runs longer in the morning, when complex issues arrive, and shorter late in the day. In this example we use 300 seconds (5 minutes, including after-call work) for every interval to keep the maths readable. If you are working to bring it down, the guide to reducing average handle time shows how to do it without hurting resolution.

How do you know if the forecast is any good?

Measure it. Forecast error per interval = |actual − forecast| ÷ actual. Average that across the intervals of the week (mean absolute percentage error). Track it at the interval level, not the weekly total: a week can be 1% off overall and still be 20% wrong at 10:00 on Monday, which is the interval that hurts. Small queues are naturally noisier, so expect larger interval errors with 30 calls an interval than with 300.

How Many Agents Do You Need per Interval?

Staffing turns each interval's forecast into people, in two steps. We use the same method as our Erlang C calculator and the full walk-through in how many call center agents do I need.

1. Agents on the phones. Traffic in Erlangs = calls × AHT in seconds ÷ interval length in seconds. Erlang C finds the smallest number of agents whose predicted service level meets the target, here 80% of calls answered within 20 seconds, with planned occupancy capped at 90%.
2. People to schedule. Divide by (1 − shrinkage) and round up. We use 30% shrinkage. Divide, never add: 21 × 1.3 = 27.3 leaves you short, while 21 ÷ 0.70 = 30.

Two lessons hide in this table. First, staffing is not linear: the 10:00 peak has 2.5 times the calls of 08:00 but needs only 2.1 times the agents, because larger groups absorb random arrivals better. Second, the margin is thin at the peak. With 20 agents instead of 21 at 10:00, the predicted service level falls from 84.0% to 75.0%. One late arrival or one long break costs you the target, which is why adherence matters so much.

On shrinkage: Call Centre Helper reports that shrinkage normally comes out between 30% and 35% (checked October 2026). Measure your own. And do not count the same time twice: if your schedule already places each break in a specific interval, take breaks out of the shrinkage figure you divide by for that interval, and keep only the time you cannot place, such as sickness and short-notice absence.

How Do You Build a Call Center Schedule?

The staffing table gives you a requirement curve: how many people you need in each half-hour. Scheduling is the job of covering that curve with real shifts that people can live with and that your labour rules allow.

Call center scheduling best practices

• Stagger start times. Instead of everyone on 09:00–17:00, start groups at 08:00, 08:30, 09:00 and 10:00. The curve rises through the morning, so should your headcount.
• Use part-timers for the peak. A 10:00–14:00 part-time shift covers the Monday peak and lunch cover without paying for a quiet late afternoon. Part-timers also give you flexibility when volume shifts.
• Place breaks in the valleys. Schedule breaks and lunches where the requirement curve dips, and stagger them so no more than a few people leave at once. In our example, lunches belong between 12:30 and 14:00, not across the 10:00 peak.
• Put training and coaching in low intervals. Off-phone activities are planned shrinkage. Book them where you have spare cover, and mark them as movable so they can be cancelled on a busy day.
• Plan time-off before you build shifts. Approved leave comes off the available pool first. Set a limit on how many people can be off per day, based on the forecast, so a busy week does not get approved into a crisis.
• Publish early and keep swaps simple. Agents who know their shifts two or more weeks ahead show up more reliably. Allow shift swaps between people with the same skills, with a supervisor check.
• Check the gap per interval. After building the rota, compare scheduled people against the requirement for every interval. Over-staffed intervals are wasted money; under-staffed ones are your service level misses waiting to happen.

If you route by skill, for example language or product line, plan each skill group separately, or at least check that each interval has enough agents with each skill. Skills-based routing improves the match between caller and agent, but it also splits your pool into smaller groups, and smaller groups need proportionally more staff for the same service level.

Schedule Adherence vs Conformance: What Is the Difference?

A perfect schedule only works if people follow it. Two numbers tell you whether they do, and they answer different questions.

A worked adherence example

An agent is scheduled for 480 minutes of activities: phones, a break at 12:00, a training session in the afternoon. During the day they:

• log in 6 minutes late (6 minutes out of adherence);
• start their 12:00 break at 12:10, so they are on the phones for 10 minutes when they should be on break, then on break for 10 minutes when they should be back on the phones (20 minutes);
• stay in training 15 minutes longer than scheduled (15 minutes);
• stay logged in 20 minutes after their shift ends to finish calls.

Minutes out of adherence: 6 + 20 + 15 = 41. Adherence = (480 − 41) ÷ 480 = 91.5%. Time worked: 480 − 6 + 20 = 494 minutes. Conformance = 494 ÷ 480 = 102.9%.

High conformance with lower adherence means the right amount of time, at the wrong times. That is exactly what breaks the peak: in our staffing table, one person missing at 10:00 drops the service level by nine points, even if they make up the time at 17:30.

Handle adherence fairly. Allow a small tolerance for a call that runs past the start of a break, look for patterns by interval before you look at individuals, and fix schedule design problems (a break placed at the peak, a training session that always overruns) before treating them as discipline problems. There is no universal adherence target that holds for every center; set yours from a few weeks of your own baseline.

How Do You Manage the Day Intraday?

Forecasts are wrong some of the time. Intraday management is what you do about it before the service level is gone. Check actuals against the plan at least every 30 minutes, and agree in advance what you will do at each level of variance, so supervisors act instead of debating.

Worked example: Monday is running hot

By 10:00, offered calls since opening are 16% above forecast. Re-forecast the next interval: 99 × 1.16 ≈ 115 calls. At 5 minutes each, Erlang C now needs 24 agents on the phones for 80/20, three more than the 21 planned. With 21 agents and 115 calls, the predicted service level falls to 47.7% and the average speed of answer to roughly 97 seconds; abandons will follow.

Work through the levers in order of cost:

1. Move off-phone activities. Postpone the 10:00 coaching session and the training block. If two people were in training, that is two agents back on the phones at no extra cost (22 agents: 64.3%).
2. Delay breaks by 15 to 30 minutes for those not yet on break, staying within your labour rules.
3. Offer callbacks. A callback that keeps the caller's place turns peak demand into work you can do in the afternoon dip. See callback scheduling.
4. Pull in cross-skilled staff from back-office work or a quieter queue.
5. Offer overtime (or an early start tomorrow if the pattern is a weekly one).
6. Route overflow to an AI voice agent for routine requests, if you have measured that it resolves them (more below).

The reverse matters as much. When volume runs well below forecast, offer voluntary time off (VTO), move people to training, outbound follow-ups or ticket backlog, and pull forward the coaching you postponed. Paying for idle time is a planning failure, just as a missed service level is.

Where Do AI Voice Agents Change the Plan?

An AI voice agent changes the forecast, not just the schedule. If it fully handles some calls, those calls leave the human queue; if it handles calls outside opening hours, you may not need a thin early or late shift at all. Two honest rules make the difference between a real saving and a spreadsheet one.

Plan on measured resolution, not containment. A call the AI answered is not a call it resolved. Count only calls that ended with a confirmed outcome (a booking saved, a question answered with no callback needed) on your own traffic, over a few weeks. Calls the AI hands over still reach your queue, often with the caller already partly through the conversation.

Forecast two streams. Forecast AI-resolved calls and human-handled calls separately, each with its own interval curve and handle time. Routine calls tend to be short, so the calls left for agents can be longer on average; update AHT, not just volume.

The maths, using our Monday peak: if the AI voice agent were measured to fully resolve 20 of the 99 calls at 10:00, the human forecast would fall to 79. Erlang C then needs 17 agents instead of 21, and 25 people scheduled instead of 30. On DialerBee's managed AI rate of $0.60 per started connected minute, 20 calls of 2 started minutes each would cost $24 for that half-hour. Whether that beats four agent-intervals depends on your labour cost, which the call center cost calculator works out from paid hours. If the measured resolution rate turns out to be 10%, not 20%, the saving halves, so start with a pilot and plan on what you measured.

After hours is often the easiest place to start, because the alternative is usually voicemail. The after-hours answering guide compares the options, and the AI receptionist page shows what DialerBee's agent handles and what it hands to your team.

AI voice agent demo — simulated caller: the replies come from DialerBee's production voice agent, the caller is synthetic and the clinic is fictional. It answers questions and books a callback, the kind of routine call that comes off the human forecast.

Transcripts are on the AI voice agent page.

Which Workforce Management KPIs Should You Track?

Track a small set every week, at the interval level where it matters. The full list of formulas is in our call center KPIs guide.

Spreadsheet or WFM Tool: Which Do You Need?

A spreadsheet is a perfectly good start. The question is when the time it takes, and the errors it hides, cost more than a tool.

The biggest practical gain of a tool is not the maths, which a spreadsheet can do. It is that forecasting, schedules and adherence use the same data as your queues, so nobody re-keys exports, and adherence is measured from what agents actually did rather than what someone remembers.

How DialerBee Supports Workforce Management

DialerBee includes workforce management with shifts, time-off, schedule adherence and forecasting, inside the same platform as your queues. The supervisor tools add agent activity timers showing how long each agent has been in ready, in wrap-up, on break or on a call, and coaching rules that surface an agent who has been in one state too long, such as wrap-up. Those states are the raw material for adherence and shrinkage.

For intraday management, callers can request a callback and keep their place in the queue, and the AI voice agent can answer inbound calls routed to it, book after checking your real calendar and hand off to a human agent. Skills-based routing is enabled per queue. See the full call center software overview for everything else in the platform, and pricing for plans.

Frequently Asked Questions

What is call center workforce management?

Workforce management (WFM) is the process of having the right number of people, with the right skills, available at the right times to meet your service targets at a sensible cost. It runs as a cycle: forecast contact volume and handle time, calculate the staff needed per interval, build schedules, manage the day as it happens, then review what was forecast against what happened.

How do I forecast call volume?

Start from at least 8 weeks of history per interval. Average the weekly totals, adjust with a seasonal index from the same week last year, split the week by day-of-week shares, then split each day by its interval curve. Add known events, such as a billing run or a campaign, as separate adjustments, and compare every forecast with actuals so the shares keep improving.

What is the difference between schedule adherence and conformance?

Adherence measures whether agents did the right thing at the right time: minutes in the scheduled activity divided by scheduled minutes. Conformance measures whether they worked the right amount of time: time worked divided by time scheduled. An agent who takes lunch an hour late can have 100% conformance and poor adherence.

What is a good schedule adherence percentage?

There is no universal number, and published benchmarks vary by source. Set your target from your own baseline after a few weeks of measurement, allow a small tolerance (for example, a few minutes when a call runs past the start of a break), and look at the intervals where non-adherence clusters rather than ranking individual agents.

How many agents do I need for my call volume?

Convert each interval's calls and average handle time into Erlangs, use Erlang C to find the smallest number of agents that meets your service level, then divide by (1 minus shrinkage) and round up. For 99 calls in 30 minutes at 5 minutes each, with an 80/20 target, you need 21 agents on the phones and 30 people scheduled at 30% shrinkage.

Can I do workforce management in a spreadsheet?

Yes, for a single queue, one site and up to roughly 15 to 20 agents with simple shifts. Spreadsheets break down when you add several queues or skills, part-timers, frequent shift swaps, intraday re-forecasting and adherence measured from real agent states. That is the point to move into a WFM tool connected to your contact center data.

How do AI voice agents affect workforce planning?

An AI voice agent can take routine and after-hours calls, which lowers the human forecast for those intervals. Plan only on the share of calls it fully resolves, measured on your own calls, and remember that the calls it hands over still reach your queue. Forecast AI-resolved and human-handled calls as two separate streams.

Does DialerBee include workforce management?

Yes. DialerBee includes workforce management with shifts, time-off, schedule adherence and forecasting, inside the same platform as your queues, plus agent activity timers that show time in ready, wrap-up, break and on a call. Those states are what adherence is measured against.

The Bottom Line

Good workforce management in a small or mid-size center is not about software first. It is a weekly habit: forecast offered calls per interval from your own history, staff each interval with Erlang C and an honest shrinkage figure, build shifts that follow the curve, act early when the day runs hot or cold, and measure what happened so next week is better. Add AI only on the share of calls you have measured it resolving.

If you want forecasting, shifts, time-off and adherence in the same platform as your queues and your AI voice agent, book a demo, or start a 14-day free trial with your own SIP trunk or PBX.