How AI-driven AHT spikes are changing workforce management

Before AI was in the picture, a contact centre queue was a mix of easy and hard interactions, with agents handling whatever came through. That mix is slowly disappearing. Now, self-service and chatbots catch most of the easy interactions before they ever reach a human. What lands with an agent is already the harder, more complex end of the queue.

This shift shows up in the data too. Service calls are now 92% longer than they were in 2004, according to ContactBabel's latest research. Agents are left handling a queue of increasingly complex calls that take longer to resolve and demand more empathy.

In a recent interview, we asked Dan Smitley, WFM expert, founder of 2:Three Consulting, and one of ICMI's Top 25 Thought Leaders, how this shift impacts agents and what it means for how contact centres forecast, schedule, pay, and plan in the future. Here's what he said:

1. Does the addition of self-service and AI mean forecasting needs to change from the ground up, or do the fundamentals still hold?

Mature forecasting means tracking what's changed and adjusting for it. Average handling time (AHT) has always fluctuated, and harder calls are just the latest factor driving it higher. According to Dan, AI and self-service are simply new input into an ever-evolving process.

He further shared that some WFM setups are overengineered, and adding solutions that promise new functionality doesn’t always resolve the actual challenge teams face. It's a pattern he’s seen long before AI was in the picture. His advice: Don’t look only at the features. Evaluate a WFM platform based on how approachable it is for your team's maturity level.

2. As agents absorb the harder calls, do traditional occupancy and shrinkage targets still hold?

Dan sees two ways teams can respond. One holds the line: It considers hard calls as a regular part of the job. So the standard shrinkage allowance, commonly 6% for two 15-minute paid breaks, stays exactly where it's always been. The other looks at the person taking the calls and asks what they might actually need to get through a shift of back-to-back, complex interactions.

He also pointed to a generational shift. Younger agents expect a different relationship with work-life balance than previous generations, Dan noted. Using his own kids as an example, he said: they want more time off than he did at their age, and he sees that as a reality WFM teams need to plan around. Holding everyone to occupancy and shrinkage levels set for an easier era, risks burnout and higher attrition.

Contact centres can address this by setting healthy occupancy levels to prevent burnout in employees. 

3. How can contact centres spot agent strain before it turns into attrition?

Dan called the answer deceptively simple: easy to say, harder to actually do. In practice, that means asking agents how the pace feels, beyond whether the occupancy number looks sustainable on paper.

He gave an example: ask someone how a heavy day felt, or how a quiet one did, and use those answers to add context to what the data already shows. One person's answer shouldn't trigger a full overhaul, but the pattern across many conversations should shape how a team reads its own numbers.

That context is the point, according to Dan. Occupancy feeds into customer experience, NPS, attrition, and engagement, alongside the service level number it's usually tied to, and you only get the full picture by tracking all of the relevant KPIs.

4. Should handling complex calls become its own career track and pay level?

Dan's take: a harder task can still get classified as entry-level, and most organisations won't pay more simply because the job got tougher. Markets, especially in the US by his own account, currently reward cost-cutting over wage investment. He noted this depends on the organisation. Some will invest in their people differently, but for most, harder calls probably won't mean higher pay.

5. How do you build a capacity plan when you don't know how much volume AI will absorb?

For long-range planning, Dan's advice is: stay close to the same process, just further out. Make assumptions transparent, and collaborate with finance, marketing, and operations. The key, he said, is being honest about the guesswork involved. It's fine to say "we feel confident in our guesses, but we're making big guesses."

His addition for the AI era: capacity plans should account for the whole pie, the volume AI absorbs included, rather than just what's left over for agents to handle. Self-service saves agent time, but it still carries a real cost to the business: there are token and infrastructure costs behind it, plus a churn risk if it frustrates customers. Weigh all of that against what a human would cost to handle the same query.

Key takeaways on AI's impact on AHT and workforce management

1. AI is absorbing simple service interactions, leaving agents with harder calls that demand more empathy.
2. AI and self-service change the input variables for forecasting, not the core forecasting method.
3. Standard shrinkage assumptions (e.g., 6% for breaks) need revisiting as call complexity rises, to reduce burnout risk.
4. Leaders should pair occupancy data with direct conversations to catch agent strain before it causes attrition, and connect metrics to NPS and engagement.
5. Long-term capacity planning should account for AI costs (e.g., token usage) and the effect of self-service on customer churn, not just headcount.