Common contact centre WFM mistakes
Most WFM failures are not dramatic — they are quiet, systematic errors that look perfectly reasonable: a daily average that hides a peak, an offered-contact forecast inflated by IVR deflection, a 20% chatbot deflection assumed to cut 20% of agents. Each produces a confident, wrong number. Here are the most common, and how to avoid them.
The common thread: nearly every WFM mistake is a variant of working with a number that looks reasonable but hides the variation — or the cost — that actually matters. They rarely announce themselves; they show up later as a mysterious SL miss, an unexplained budget overrun, or an attrition spike. Each one below is paired with the principle that avoids it and the guide that covers it in depth.
Planning on averages that hide the peak
Why it happens
A daily or weekly average is simpler to work with and looks reasonable. But staffing is driven by each interval — an average can look fine while the morning peak is badly short and the afternoon is overstaffed.
The fix
Plan and measure at interval level (typically 30 minutes). Use the average only for long-range capacity, never for scheduling or coverage assessment.
Forecast granularity →Forecasting from offered instead of handled contacts
Why it happens
The ACD's headline 'offered' figure is the obvious number to grab. But in a high-IVR environment it includes contacts that self-served and never reached an agent — forecasting from it overstates the staffing requirement.
The fix
Forecast the staffing model from handled (agent-reaching) contacts. Use offered volume only to understand total demand, and watch for IVR changes that shift the deflection rate.
Telephony fundamentals for WFM →Assuming deflection cuts headcount proportionally
Why it happens
If a chatbot deflects 20% of contacts, cutting 20% of agents seems obvious. But the bot takes the simplest contacts, so the residual reaching agents is harder and slower — AHT rises and the real saving is smaller.
The fix
Model deflection as a change to both volume AND residual AHT. A 20% deflection may justify only ~10% fewer agents. Hold cuts behind proven true-containment.
AI & chatbot deflection WFM →Running occupancy too high to save cost
Why it happens
High occupancy looks efficient and saves headcount on paper. But sustained 90%+ occupancy drives AHT creep, absence, and attrition — costs that are deferred and attributed elsewhere, so they never appear on the WFM cost line.
The fix
Target 80–85% occupancy for voice. Build the business case showing the deferred costs of running lean exceed the headcount saved.
Occupancy & burnout →Using the wrong model for the channel
Why it happens
Erlang C is the famous one, so it gets applied everywhere — including to email, case work, and outbound, where it does not belong. Each produces a wrong requirement.
The fix
Match the model to the work: Erlang C for live voice, concurrency for chat, backlog/flow for email and case work, throughput for outbound. Never one model for all channels.
Staffing models guide →Borrowing benchmark assumptions instead of measuring
Why it happens
Plugging in '35% shrinkage' or '6 minute AHT' because it is typical is quicker than measuring. But your real figures may differ by several points — and a few points of shrinkage is several agents.
The fix
Build shrinkage, AHT, and the arrival profile from your own data. Use benchmarks only to sanity-check, never as model inputs.
Shrinkage benchmarks →Reading cost-per-contact in isolation
Why it happens
Cost-per-contact is a tidy headline metric. But it divides cost by volume, not by service — so cutting agents always 'improves' it, even as the service level collapses.
The fix
Always read cost-per-contact alongside service level. A falling cost with falling SL is service erosion, not efficiency.
Scenario planning →Confusing schedule adherence with schedule effectiveness
Why it happens
When SL misses, the reflex is to blame agents for not following the schedule. But the schedule itself may be wrong — built on a bad forecast, wrong shift starts, or under-estimated shrinkage.
The fix
Diagnose before acting: if adherence is high but SL still misses, the schedule is the problem, not the agents. Measure both separately.
Schedule effectiveness →Starting peak recruitment too late
Why it happens
Recruitment starts when volume is visibly rising. But the lead time from decision to a productive agent is 8–14 weeks — by the time the peak is visible, it is too late to be ready for it.
The fix
Work backwards from the peak across the full recruit-train-ramp chain. Begin recruitment months ahead, not when demand starts climbing.
Peak season ramp →Treating long-term sick as ordinary absence
Why it happens
LTS gets absorbed into the daily shrinkage figure like any absence. But it occupies a headcount slot indefinitely with zero capacity return — it is a headcount event, not an absence event.
The fix
Reclassify LTS after ~4 weeks as a discrete headcount-model entry, and make an explicit backfill decision against the buffer.
Long-term sick management →WFM mistakes questions
What is the most common workforce management mistake in contact centres?
Planning at the wrong level of aggregation — using a daily or weekly average where interval-level detail is needed. Staffing is driven by what happens each half-hour, not by the daily total. A day that looks adequately staffed on average can be badly short at the morning peak and overstaffed in the afternoon — and because queuing is non-linear, the understaffed peak causes an SL collapse the comfortable average hides. Almost every other common WFM mistake is the same theme: working with a number that looks reasonable but hides the variation that matters — forecasting from offered (IVR-inflated) instead of handled contacts; assuming a 20% chatbot deflection cuts 20% of agents; running occupancy too high to save cost; reading cost-per-contact without service level. WFM mistakes are rarely dramatic — they are quiet, systematic, and produce confidently wrong numbers.