The gap between plan and reality

The first three articles in this series covered the planning side of WFM — forecasting demand, building schedules, and calculating the staffing numbers that underpin both. This article covers what happens after the plan meets reality.

No forecast is perfect. No schedule survives contact with a week of live operations unchanged. Agents call in sick. Volume arrives 30% above forecast on a Tuesday morning with no warning. A product incident generates a surge of contacts that nobody predicted. A key specialist is pulled into an urgent cross-functional meeting during your busiest hour.

The question is not whether your plan will diverge from reality — it will, every day. The question is how quickly you detect the divergence, how effectively you respond to it, and how systematically you use what you learn to improve next week's plan.

That is what this article is about.

Schedule adherence: the foundation of real-time management

Before you can manage real-time deviations, you need a baseline to deviate from. That baseline is the schedule — and the metric that measures how well agents are following it is schedule adherence.

Schedule adherence is the percentage of scheduled time that agents spend doing what their schedule says they should be doing, when it says they should be doing it. An agent scheduled to be available for contacts from 9:00am to 12:00pm who logs in at 9:15am, takes a break 20 minutes early, and spends time in an unscheduled offline state after lunch is not adherent — even if they handle a full day's worth of contacts.

The formula is straightforward:

Adherence % = (Time spent in scheduled activity) ÷ (Total scheduled time) × 100

Industry benchmark for a well-run operation is above 90%. Below 85% is a signal that adherence is actively undermining your staffing model — you may be rostering enough agents on paper but not getting the productive coverage those numbers assume.

Why adherence matters more than it appears

Adherence is often treated as a discipline issue — agents not following their schedule — rather than an operational metric. That framing misses its true significance.

When adherence breaks down, your staffing model breaks down with it. Your Erlang C calculation assumed a certain number of productive agents available at each interval. Your shrinkage budget accounted for planned unavailability. What it did not account for is agents being offline at times when they were scheduled to be available.

The compounding effect is significant. If 20 agents are scheduled for a peak interval but 4 are logged off for unscheduled reasons, you effectively have 20% less capacity than planned — not for the whole shift, but precisely during the period when you needed it most. SLA performance drops sharply during that interval, and the queue that builds during the shortage takes time to clear even after agents return.

In operations with tight SLA commitments — particularly in regulated industries where breaches carry financial penalties — adherence gaps during peak intervals are one of the most common root causes of SLA failures that look, on the surface, like a staffing problem.

What drives poor adherence

Understanding why adherence breaks down is more useful than simply measuring that it has. The causes fall into three categories.

System and process causes are the most overlooked. If agents have to navigate multiple systems to log in, if after-contact work routinely takes longer than the scheduled wrap time, or if the ticketing system frequently crashes and forces unplanned offline states, adherence will be poor regardless of how motivated your agents are. Fix the friction first before attributing adherence gaps to behaviour.

Management causes are the next layer. Managers who routinely pull agents off the queue for ad-hoc meetings, who allow informal break extensions to become normal, or who haven't clearly communicated why adherence matters are contributing to the problem. Adherence culture starts with management behaviour.

Individual behaviour causes are the third category — and the one most managers focus on first, which is why adherence problems often persist. An agent who consistently logs in late or takes extended breaks is a coaching conversation. But if ten agents across different teams are showing the same pattern, the cause is almost certainly systemic rather than individual.

Real-time management: responding to what's actually happening

Real-time management — sometimes called intraday management — is the practice of monitoring live operational performance against plan and making adjustments in the moment to protect SLA performance.

It requires two things: visibility and decisiveness. Visibility means knowing, in near real-time, what your queue looks like, how many agents are available, what your current service level is, and how far you are from your SLA target. Decisiveness means having the authority and the playbook to act on that information without waiting for approval.

The real-time management toolkit

Real-time managers — whether that's a dedicated WFM analyst, a team lead, or the manager themselves depending on team size — work with a standard set of interventions when live performance diverges from plan.

When volume is running above forecast and SLAs are at risk:

Bring agents back from breaks early or delay scheduled breaks if they haven't started yet. This recovers capacity quickly without requiring additional headcount. Move agents from lower-priority queues to the queue under pressure, if skills allow. If the overrun is significant and sustained, consider voluntary overtime for agents finishing their shift. Communicate to the team what's happening and why — agents perform better when they understand the operational context rather than just being told to stay online.

When volume is running below forecast and occupancy is dropping:

Release agents from the queue for training, coaching, or knowledge base work. Schedule any outstanding one-to-ones or team huddles that have been deferred. Use the quiet period proactively rather than leaving agents idle — idle agents disengage and productive use of quiet time builds goodwill.

When an unexpected event generates a spike — a product incident, a system outage, a regulatory announcement:

Activate your incident runbook if one exists. Notify team leads immediately so they can hold agents on the queue. Consider a brief all-hands message to agents explaining the nature of the spike — agents who understand why volume has surged handle the pressure better than those who experience it as unexplained chaos. If the spike is likely to sustain, escalate the staffing decision to the Director level rather than trying to absorb it through intraday adjustments alone.

Intraday reforecasting

The most sophisticated real-time management practice is intraday reforecasting — updating your volume projection for the remainder of the day based on what has actually arrived so far.

The principle is simple: if you've seen 30% more contacts than forecast in the first two hours of the day, the most likely explanation is that today's total volume will be higher than forecast — not that the remaining hours will be quieter to compensate. Intraday reforecasting captures this signal and adjusts staffing decisions for the afternoon based on actual morning data rather than the original forecast.

Most WFM platforms do this automatically. For teams without dedicated WFM tooling, a simple approach is to track the ratio of actual to forecast volume at the end of each hour and apply that ratio to the remaining forecast intervals. If you're running 120% of forecast through hour two, adjust your afternoon requirements to 120% of the original plan and make staffing decisions accordingly.

The adherence-occupancy balance: a practical tension

One tension that real-time managers encounter regularly is the relationship between adherence and occupancy. Strict adherence means agents are always where the schedule says they should be — which is operationally correct but can feel rigid to agents and managers alike. High occupancy means agents are busy, which is efficient but can mask the fact that they have no recovery time.

The practical resolution is to manage adherence as a structural requirement and occupancy as a health indicator. Adherence should be high — above 90% — as a baseline operational discipline. Occupancy should be monitored to ensure it stays within the 75–85% range. If adherence is high but occupancy is consistently above 85%, the signal is understaffing, not individual behaviour. If adherence is low and occupancy is also low, agents are both unavailable and idle when available — a management and process problem that needs direct intervention.

Closing the loop: the continuous improvement cycle

Real-time management handles the immediate. Continuous improvement handles the systemic. The goal of the improvement cycle is to ensure that every week's operational data makes next week's plan more accurate.

The cycle has four stages that repeat on a weekly cadence.

Review actuals versus plan. After each week, compare actual contact volume to forecast volume at the interval level. Identify where the largest gaps were and what caused them. Was it a one-off event, a systematic forecast miss, or a leading indicator you didn't have visibility on? Document the answer.

Measure forecast accuracy. Calculate MAPE for the week. Track it on a rolling basis. If MAPE is improving, the forecasting model is working. If it's stable but consistently high, the model has a systematic bias. If it's volatile — accurate some weeks and badly wrong others — the model is missing a variable that drives volume in some conditions but not others.

Analyse adherence patterns. Review adherence data by team, by time of day, and by individual. Look for patterns: is adherence consistently poor in the first hour of the shift? Is one team's adherence notably lower than others? Are the same agents repeatedly out of adherence? Each pattern points to a different cause and a different intervention.

Update the model. Incorporate what you learned into next week's forecast, shrinkage budget, and schedule. If last week revealed that Monday morning volume is consistently 15% higher than your model predicted, adjust the Monday morning forecast upward. If sick leave ran higher than budget, update the shrinkage assumption. If a specific interval consistently sees adherence gaps because of a system slowdown, flag it for a process fix rather than continuing to absorb it through staffing buffers.

Forecast error: diagnosing what went wrong

When your forecast misses significantly, the diagnostic question is not just how much it missed by but why. Different causes require different fixes.

Volume ran higher than forecast across the whole day. The trend component of your forecast is understating growth. Review your rolling average — if you're using 8 weeks of history and your operation has grown 20% in the last month, older weeks are dragging the trend down. Shorten your history window or add a growth adjustment.

Volume ran higher than forecast at specific intervals but matched for the day overall. The seasonality component is wrong. Your intraday distribution model doesn't reflect current customer behaviour. Rebuild your interval weights from the most recent 4–6 weeks rather than a longer historical period.

Volume was correct in aggregate but the mix of contact types was different. Your AHT assumption broke down. More complex contacts than usual, or a product issue driving a different query type, changed the effective handle time. Review AHT by contact type and update your segmented model.

Volume was unpredictably high due to a one-off event. A genuine outlier — a product incident, a regulatory announcement, a media story. Exclude it from your baseline forecast to avoid it distorting future predictions, but document it as a playbook case for the next time a similar event occurs.

Building a WFM performance dashboard

The continuous improvement cycle requires data that is visible, consistent, and actionable. A WFM performance dashboard — whether built in your WFM platform, your BI tool, or even a well-structured spreadsheet — should surface the following metrics on a weekly basis:

Forecast accuracy: MAPE for the week, trended over the past 8 weeks. The trend matters as much as the current number.

SLA attainment by severity tier: Are you hitting your targets? Which tiers are underperforming and during which intervals?

Adherence rate: By team and by individual. Trending over the past 4 weeks.

Occupancy: Average for the week and by interval. Flag any intervals where occupancy exceeded 85% consistently.

Shrinkage actuals versus budget: Did actual unavailability match your budget? Which categories overran?

Intraday interventions: How many times did real-time management need to intervene? What triggered each intervention? This is a leading indicator of forecast accuracy — frequent interventions signal a forecast that's systematically missing demand shape.

From reactive to proactive: the maturity curve

WFM maturity in a CS operation progresses through recognisable stages. Understanding where your operation sits on this curve helps you prioritise investment.

At the earliest stage, staffing decisions are made on gut feel and historical headcount. There is no formal forecasting, schedules are built on preference rather than demand, and SLA performance is managed reactively — by throwing people at queues when they back up.

At the next stage, basic forecasting exists — usually a simple moving average — and schedules are built from it. Shrinkage is accounted for, though often using an industry average rather than measured data. Adherence is tracked but not systematically managed.

At a more advanced stage, the operation uses decomposition forecasting with leading indicators, runs Erlang C at the interval level, maintains a measured shrinkage budget, and has a real-time management function with clear playbooks for common scenarios. Forecast accuracy is tracked and improving.

At full maturity, the WFM function is genuinely predictive. It anticipates demand shifts before they arrive, updates the staffing model continuously based on live data, and feeds operational intelligence back into hiring plans, product decisions, and SLA commitments. The organisation treats WFM as a strategic input rather than a scheduling administrative task.

Most operations sit somewhere between the second and third stage. Moving from reactive to proactive doesn't require a large team or expensive tooling — it requires disciplined application of the methods covered in this series, applied consistently over time.