Scaling a contact centre

A contact centre planning to grow asks a reasonable-sounding question: we are handling four hundred calls a day with twelve agents, so what do we need for eight hundred?

The intuitive answer is twenty-four. The correct answer is fewer, and understanding why changes how the whole growth is planned — because the same effect explains why a small team feels permanently under pressure while a larger one handling proportionally more feels comfortable.

Why staffing is not linear

Calls do not arrive evenly. They arrive randomly, which means clusters and gaps. A team must be staffed for the clusters, and in the gaps that staffing is idle.

The larger the team, the less this hurts. With three agents, two simultaneous calls plus one more means a queue. With thirty agents, the random clustering averages out across a much larger pool, so the same proportional surge is absorbed without anyone waiting.

This is the pooling effect, and it is why doubling volume requires substantially less than double the staff. It is formalised in the Erlang C model, which is the basis of essentially all contact centre workforce planning, and the specific numbers depend on your call volume, average handling time and target service level.

Occupancy: the trade-off nobody states

Occupancy is the proportion of an agent's logged-in time spent handling calls. Low occupancy looks like waste and every operation under cost pressure tries to raise it.

The trade-off is that queue time rises non-linearly as occupancy approaches full utilisation. Moving from moderate to high occupancy costs some waiting; moving from high to very high costs disproportionately more, because there is no slack left to absorb the random clustering described above.

Two consequences follow, and the second is the expensive one.

• A service level target and an occupancy target are in direct tension. Both can be stated; both cannot be maximised. Which one yields should be an explicit decision rather than an emergent one.
• Sustained very high occupancy is a staff retention problem. Back-to-back calls with no recovery time produces burnout, and attrition in a contact centre is expensive — recruitment, training, and a period of lower productivity while a replacement learns.

That second point is why cost-driven occupancy targets frequently fail to save money. The saving is visible in the staffing model and the cost appears in the attrition rate, where nobody attributes it back.

Shrinkage: the number that makes forecasts wrong

Agents are not available for the whole of their paid time. Breaks, training, meetings, coaching, system problems, holiday and sickness all remove availability, and the total is large.

Forecasts built on headcount rather than on available agent hours are wrong by whatever that total is, consistently, in the same direction. This is the most common reason a centre that is staffed on paper misses its service level in practice.

1. 1

   Measure your own shrinkage rather than using a benchmark

   It varies enormously by operation, and an industry figure applied to your centre is a guess wearing a number.

2. 2

   Split it into planned and unplanned

   Training and holiday are schedulable. Sickness and system outages are not. They need different treatment in the model.

3. 3

   Forecast in available agent hours

   Not headcount, not full-time equivalents. The unit that matters is hours actually on the queue.

4. 4

   Track forecast against actual, per interval

   A forecast nobody reconciles never improves, and reconciling at the daily level hides the intervals where the failures happen.

What breaks first on the infrastructure side

Growth exposes infrastructure constraints in a fairly predictable order, and knowing the order lets a centre address them before they bind rather than during a hiring push.

The last row is not infrastructure and is included deliberately, because it binds sooner than most of the technical constraints and is almost never planned for. Growing from twelve agents to thirty without adding supervision produces a quality problem that gets attributed to hiring standards.

Carrier capacity heading the list is worth acting on directly. Channel capacity is purchased, and a centre that grows headcount without increasing it discovers the limit at the busiest moment of the busiest day, which is exactly when the failure is most visible to customers.

Where the real gains usually are

Before adding agents, three changes routinely deliver more than a proportional headcount increase, and all three are configuration or process rather than capacity.

• Pool the queues. Separate small queues per team or product each suffer the small-team inefficiency independently. One pooled queue with skills-based routing captures the pooling effect and needs fewer agents for the same service level.
• Reduce handling time by removing work from the call. Screen pop, click-to-call and write-back remove seconds of manual lookup and typing from every single call, and at volume that compounds into a headcount equivalent.
• Remove the calls that should not be calls. A meaningful share of contact volume in most centres is status enquiries that a notification or a self-service view would have answered. Those are the cheapest calls to eliminate because the customer prefers it too.

The third is the largest in most operations and the least often examined, because contact volume is treated as demand to be met rather than as a signal about something upstream. A recurring enquiry type is usually evidence of a process or communication gap, and fixing that is cheaper than staffing it forever.