Here's a number that should worry every contact centre manager: 31% of agents say they're likely to quit within six months. That's not a projection or an industry estimate — it's from CX Today's 2026 workforce survey, and it matches what we're hearing from operations directors across the UK.
The cost of replacing a single experienced agent runs between £8,000 and £16,000 when you factor in recruiting, training, and the productivity ramp. For a 100-seat centre with typical turnover, that's half a million pounds walking out the door every year.
So when vendors pitch AI as the solution to everything, the natural question is: what does this actually do to the people who work here?
The honest answer is complicated.
What burns agents out (it's not what you'd guess)
Ask agents why they're thinking of leaving, and the answers cluster around three themes:
Repetitive work that doesn't use their skills. Answering the same "where's my order" question fifty times a day isn't challenging. It's draining. Agents who joined because they enjoy solving problems spend their shifts on queries a moderately competent FAQ page could handle.
Lack of support on difficult calls. When a customer is angry, confused, or dealing with a genuinely complex issue, agents often feel abandoned. Searching through knowledge bases while a customer waits, or putting someone on hold to ask a colleague — it makes you feel incompetent even when you're not.
After-call documentation. The call ends. Then comes the note-taking, the CRM updates, the ticket categorisation. This admin work adds 30-60 seconds per interaction and nobody enjoys it. It's the tax agents pay for doing their actual job.
Notice what's missing from this list: "too many customers." The volume itself isn't the primary complaint. It's the quality of the work and the support available to do it well.
Where AI tools genuinely help
When deployed thoughtfully, AI addresses exactly these pain points:
Removing repetitive queries from the queue. Autonomous voice agents that handle order status, account balance, and appointment scheduling don't just reduce costs — they remove the work that makes agents feel like glorified answering machines. The queries that reach human agents become more varied and interesting by default.
One facilities management company we work with shifted 43% of their inbound calls to AI handling. Their agent satisfaction scores went up, not because of any deliberate engagement initiative, but because the remaining calls actually required human judgment.
Real-time knowledge surfacing. Agent assist tools that pull relevant information during a call eliminate the "please hold while I look that up" moments. The agent isn't hunting through tabs — the answer appears alongside the conversation. It sounds minor until you've watched an agent's posture change when they realise they have what they need.
Automated post-call work. AI-generated call summaries aren't perfect, but they're good enough that agents spend seconds reviewing rather than minutes writing. That 30-60 second saving per call adds up to hours per week. More importantly, it removes the task agents resent most.
Where AI tools make things worse
Not every AI deployment helps. Some actively damage agent experience:
Surveillance dressed as assistance. When AI monitors every word an agent says, scores their performance in real-time, and flags "compliance risks" for supervisor review, agents feel watched rather than supported. There's a meaningful difference between a tool that helps you and a tool that judges you.
The worst implementations use AI to track metrics that don't reflect actual service quality — talk time, script adherence, sentiment scores divorced from context. Agents learn to game these metrics rather than serve customers well.
Automation that creates more work. If your AI handles simple queries but escalates anything slightly ambiguous, agents end up with concentrated difficult calls and no relief. The emotional labour per hour goes up, not down.
Similarly, if AI tools require constant correction or generate summaries that need extensive editing, the productivity gain evaporates. Agents end up cleaning up after the AI rather than being helped by it.
Unclear boundaries and accountability. When an autonomous agent makes a mistake, who's responsible? If the answer is "the human agent who should have caught it," you've created anxiety without providing control. Clear escalation paths and reasonable expectations matter more than the technology's theoretical capability.
The job changes. That's not automatically bad.
Let's be direct about what AI means for contact centre work: fewer people will be needed to handle the same volume. That's the whole point of automation. Pretending otherwise insults everyone's intelligence.
The question is what happens to the roles that remain. If AI removes routine work and leaves agents handling complex queries, complaints, and relationship-building, the job becomes more skilled — and should be compensated accordingly.
The contact centres getting this transition right are:
Redefining the agent role. Not "person who answers phones" but "person who handles situations that require judgment, empathy, and problem-solving." That's a different job with different requirements and different rewards.
Investing in training for the new work. Complex queries require deeper product knowledge, better de-escalation skills, and more autonomy to resolve issues. Agents need development, not just deployment alongside new tools.
Being honest about headcount changes. Gradual reduction through attrition is more humane than sudden layoffs, but it still needs to be acknowledged. Teams that understand the direction of travel can prepare, upskill, and make informed career choices.
Sharing productivity gains. If AI makes your operation 30% more efficient, some of that should flow to the people still doing the work. Higher base pay for a more demanding role isn't generosity — it's fair exchange.
Practical questions before you deploy
Before adding any AI tool to your contact centre, ask these:
What repetitive work does this remove? If the answer is "none," the tool might be solving the wrong problem. Efficiency gains that don't improve agent experience often create other costs.
How will agents interact with this daily? Watch the actual workflow, not the demo. Where are the friction points? What requires agent correction or oversight? How much time does that take?
Does this monitor or support? There's room for both, but be clear about which you're doing. Tools that help agents should feel helpful. Tools that measure agents should be positioned that way honestly.
What happens when it's wrong? Every AI system makes mistakes. How does the agent recover? Who's accountable? Is the correction process quick enough not to frustrate customers waiting on the line?
How will you measure impact on your people, not just your numbers? Track agent satisfaction alongside handle time. Watch turnover rates alongside resolution rates. If efficiency goes up but your best agents leave, you haven't actually won.
The bottom line
AI tools can significantly improve agent experience — but only if that's actually a design goal. Technology deployed purely for cost reduction, without consideration for the humans using it, tends to accelerate the burnout and turnover it was supposed to address.
The contact centres that will thrive in 2027 and beyond aren't the ones with the most AI. They're the ones that use AI to make agent work more interesting, better supported, and more fairly compensated.
That's not a technology problem. It's a leadership one.
Hostcomm provides AI voice agents and contact centre technology for UK businesses. If you're evaluating how AI affects your team — not just your metrics — we'd welcome the conversation.