Most contact centers now have two workforces. One has names, badges and shift bids. The other runs on software and answers thousands of customers a day. But in most workforce management plans, only the first one is treated as a workforce at all.
The usual approach goes like this. Forecast total demand, subtract what the bot is expected to contain, and staff people for whatever is left. AI shows up as a single deflection percentage, usually set once during the business case and rarely revisited. That was good enough when bots handled password resets. It isn’t good enough now that AI agents handle billing disputes, order changes and the first stage of complex service requests.
When AI handles a large share of your contacts, it stops being a discount on demand. It becomes supply, and supply has to be planned.
Why the Deflection Percentage Fails
A flat deflection rate assumes AI performs the same way every hour of every day. It doesn’t. Containment changes with contact type, time of day, customer group and what’s happening in the business. A product launch, a billing error or a weather event can bring in contacts the AI was never trained for. Containment drops right when volume spikes, and the overflow lands on a human team staffed on the assumption the AI would absorb it.
The opposite problem also happens. When AI performs better than planned, centers end up overstaffed and send agents home early or pile on idle time. Either way, the plan was wrong because one input was treated as fixed when it’s actually a moving workforce.
Treat AI Like a Team in the Plan
Planning human and AI capacity together starts with giving AI the same basic attributes we give any staff group.
- Capacity by work type: What can the AI handle on its own, what can it start and hand off, and what should it never touch? Each work type needs its own containment rate, not a single blended figure.
- Performance that changes over time: Track containment and handoff rates by interval, the same way you track handle time and occupancy.
- Cost per contact: AI left unmanaged can be your most expensive resource. Usage pricing, platform costs and the cost of handoffs that fail all add up. Comparing the real cost of an AI-resolved contact with a human-resolved one lets planners make actual trade-offs.
- Failure modes: Every team has a plan for absences. AI needs one too. What happens if containment drops by half for two hours?
Plan for the Handoff
The most under planned part of hybrid service is the handoff. When AI passes a customer to a person, that contact is often longer, more emotional and more complex than the average. The customer may already be frustrated, and the agent has to pick up the context quickly.
Handoff volume should be forecast as its own work type, with its own handle time and skill requirements. Staffing it as ordinary inbound volume underestimates both how long it takes and how hard it is on agents.
What Changes for the WFM Team
Bringing AI into the capacity plan changes the role of the workforce planner. Forecasting now includes predicting how the AI will perform, not just how many customers will contact you. Intraday management includes watching AI containment as closely as service level. Long-range planning becomes a question of mix: how much of next year’s demand should people handle, how much should AI handle, and what happens to cost, quality and employee experience under each option.
It also means planners need a seat at the table when AI changes are made. A new prompt, a policy update or an expanded use case can change human staffing needs overnight. If WFM finds out after the fact, the plan is already wrong.
The Employee Experience Angle
When the AI side of the plan is wrong, agents pay for it in surprise overtime, canceled time off and long stretches of difficult contacts with no relief. Planning both workforces together protects schedule stability, which is still one of the strongest drivers of agent retention.
Start Small
You don’t need new technology to begin. Pick your highest-volume AI-handled work type. Track containment by interval for a month. Add handoffs to the forecast as a separate work type. Then compare your staffing plan with what actually happened.
You may find that you’ve been planning for one workforce when in fact you have two.