DECISION LAB 01 · CONTACT CENTER & WORKFORCE
What Does AI Actually Save in a Contact Center?
Automation rate is not workforce savings.
THE BUSINESS QUESTION
Automation rate is not workforce savings.
A common AI business case begins with a seemingly straightforward assumption:
If AI automates 20% of customer contacts, staffing should fall by roughly 20%.
In practice, the relationship is rarely that simple.
A contact center is a queueing system. Staffing depends not only on total workload, but also on when that workload arrives, the service level the organization promises, the amount of protective capacity required to handle random demand, and whether reduced interval requirements can actually be translated into feasible employee shifts.
The chain is closer to:
Contacts → Workload → Capacity → Schedule → Workforce
AI can affect every step differently.
START WITH THE SIMPLEST USEFUL BASELINE
Erlang C and service-protection capacity
Suppose a contact center receives 100 calls in a 30-minute interval.
Average handle time is 6 minutes.
The service target is 80% answered within 20 seconds.
Shrinkage is 25%.
Using a standard Erlang C capacity model, the operation carries:
20.0 workload-equivalent agents
but requires approximately:
25 active agents
to protect the service target.
After shrinkage, that becomes:
34 scheduled positions.
The difference between workload and active staffing is not necessarily waste.
It is service-protection capacity.
Random arrivals mean that staffing exactly to average workload would create queues and missed service targets.
20.0
workload-equivalent agents
25
active agents for the SLA
34
scheduled positions after shrinkage
SCALE EFFECT
Why utilization depends on operation size
Now imagine exactly the same operating model at different sizes.
Keep:
the same AHT,
the same service target,
the same demand pattern,
but change total volume.
At one-quarter of the original scale, the model may support the service target at only about 62.5% occupancy.
At ten times the original scale, the same SLA may be achievable at approximately 94.8% occupancy.
Why?
Larger pooled operations require more protective capacity in absolute terms, but less protective capacity relative to workload.
This is a fundamental scale effect in service operations.
It also means that there is no universal “good utilization target” for a contact center.
An 85% occupancy target may be conservative for a very large operation and impossible for a small one.
NOW INTRODUCE AI
Automation → workload → capacity
Consider a simple AI scenario:
20% contact deflection
and
10% reduction in average handle time
for the contacts that still reach an agent.
The naïve interpretation might be:
20% automation should create roughly 20% workforce savings.
The baseline model tells a different story.
In our illustrative scenario:
20.0% of contacts disappear
but workload falls by:
28.0%
because AI also reduces handle time.
That lower workload then changes the Erlang C staffing requirement.
At the interval level, scheduled positions fall from:
34 to 26.
That is approximately:
23.5% lower scheduled capacity for that interval.
Already we have three different numbers:
20.0% contacts automated
→ 28.0% workload reduction
→ 23.5% interval staffing reduction
None is inherently the “AI savings number.”
They describe different parts of the operating system.
WHAT IF AI REMOVES THE EASY WORK?
The remaining calls may be more complex
AI often does not automate a random sample of calls. It may remove relatively simple interactions first: password resets, status questions, routine transactions, and basic information requests. The calls remaining for human agents may therefore be more complex.
70% × 125% = 87.5%
of original workload when 30% of contacts are deflected but remaining handle time rises by 25%.
So 30% contact automation can produce only 12.5% workload reduction—and roughly the same baseline schedule reduction. That is a very different business case.
TRY THE MODEL
Explore how the result changes when you adjust volume, handle time, service targets, shrinkage, AI deflection, residual call complexity, intraday demand, and shift length. The objective is not a universal staffing answer—it is making the tradeoffs visible.
Contact Center Capacity Planner
Start with a transparent Erlang C baseline, then explore scale effects, AI-driven workload changes, intraday staffing, and a transparent baseline shift schedule using the same capacity engine.
INTRADAY DEMAND CREATES ANOTHER LAYER
When savings occur matters
Contact-center demand is not constant throughout the day.
A productivity improvement during a low-volume interval may reduce workload without eliminating a single shift.
The same improvement during the daily peak may reduce the number of people required across several hours.
So the next step is to run the capacity model interval by interval.
For each 30-minute period we calculate:
workload-equivalent agents
active agents required for SLA
and
scheduled positions after shrinkage.
The result is an intraday staffing curve.
This lets us ask a more useful question:
Where does AI reduce capacity requirements during the day?
not merely:
How much workload did AI remove?
INTERVAL SAVINGS ARE STILL NOT FTE SAVINGS
Interval savings are not FTE savings
In our illustrative intraday profile, the current operation requires approximately:
301.5 scheduled capacity-hours
across the day.
Under the AI scenario, this falls to approximately:
226.5 capacity-hours.
That is a reduction of roughly:
24.9%.
But employees do not work in 30-minute fragments.
They work shifts.
With a simple baseline of fixed 8-hour shifts, the current requirement becomes approximately:
384 paid schedule hours
versus:
288 hours under the AI scenario.
With the profile repeated five days per week and 40 paid hours representing one FTE, the baseline schedule corresponds to approximately:
48 FTE currently
and
36 FTE under the AI scenario.
So under these particular assumptions:
20% contact deflection
→ 28% workload reduction
→ 24.9% interval capacity-hour reduction
→ 25% baseline-schedule FTE reduction
The numbers happen to be fairly close in this example.
They do not have to be.
WHERE THE BASELINE BREAKS
The next question—not the starting assumption
The model deliberately begins simply.
It assumes a pooled skill group, Poisson arrivals, an Erlang C queue, no abandonment, and steady-state behavior within each interval.
Real operations may also involve multiple skills and routing rules, abandonment, chat and asynchronous channels, different call types, variable service times, schedule preferences, labor agreements, part-time employees, lunches and breaks, outsourcing, cross-training, and uncertain forecasts.
Those complications matter only if they materially change the decision.
That is the next question—not the starting assumption.
MANAGEMENT IMPLICATIONs
Four questions executives should ask
When evaluating an AI contact-center business case, do not ask only:
What percentage of calls will AI automate?
Ask four separate questions:
How many contacts disappear?
How much workload disappears?
How much capacity is no longer required to maintain the SLA?
How much of that capacity can actually be removed from the workforce schedule?
The answer at each step may be different.
And that difference can materially change the economics of the AI investment.
Understand deeply. Solve simply.
Have a contact-center planning problem that goes beyond the baseline?
Discuss Your Planning Problem