Agentic AI customer service in 2026: What it really means for BPOs and their clients
At a glance
- Agentic AI customer service takes action for the customer. It can resolve tickets, process refunds, and run multi-step workflows from start to finish.
- The promise is real: 24/7 resolution, lower cost per contact, and headcount cuts at scale.
- Full headcount replacement is not realistic within a year. The next three to five years favour hybrid models, with people supervising AI on complex work.
- So the winning move for BPOs and CX buyers is to lean in early: supervise AI rather than fear it, scope pilots by intent, and design outcome-based pricing before rivals do.
Silicon Valley’s pitch for the post-BPO future is simple. Agentic AI will wipe out the human side of customer service, and operators who do not pivot will go the way of Kodak. Public BPO share prices have already taken the hit.
However, for operators deploying agentic AI customer service, the reality is messier than the slide deck.
In the 589th episode of the Outsource Accelerator Podcast, Simone Bartlett is direct about both the threat and the trap. She is co-founder and CEO of Africa Operations at Hugo, a 4,500-person outsourcing firm across Nigeria, South Africa, and the US. In her view, the technology is unstoppable. The timeline is not.
What is agentic AI customer service?
Agentic AI customer service is AI that takes action for the customer, not just AI that answers questions. It can resolve tickets, trigger refunds, update accounts, and run multi-step jobs end to end.
- It goes beyond chatbots and IVR, because it can navigate workflows and take actions on its own.
- Gartner frames it as a new paradigm where AI systems can act autonomously to complete tasks.
- For BPOs, it shifts the human role from handling every contact to supervising the AI.
For Simone, it is the live conversation of the moment. As she puts it, “In CX, the big talk now is agentic.” This is the next step beyond the rule-based scripts behind earlier customer service outsourcing setups.
The promise and honest trade-off of going agentic
The pitch is compelling: 24/7 resolution, lower cost per contact, and no seat-based scaling pain. In fact, Gartner expects 40% of enterprise applications to include task-specific AI agents by the end of 2026, up from less than 5% in 2025.
Still, the catch is what that does to BPO headcount. Simone does not dress it up:
“It’s difficult to go tell one of your best clients, ‘Have you started thinking about agentic and how we can layer into our way of working?’
Because it will mean headcount reduction. When done well, it will mean headcount reduction.”
As a result, operators willing to have that conversation early shape the work that remains. Those who hide get cut out of it. This is why the debate over whether AI will replace call centre agents matters so much to BPO leaders right now.
Why full replacement is not happening yet in customer service
Despite the projections, end-to-end agentic AI customer service is not a present-day reality at scale. For example, McKinsey’s AI trust research flags a new risk. Firms must now manage systems that not only say the wrong thing but do the wrong thing.
Simone puts it sharper:
“Simple mistakes will shake trust. If we say, ‘Let’s go fully agentic,’ and we don’t have humans, and then something goes wrong, [you’d think], ‘I should have humans just overseeing or checking it.'”
Her analogy makes the point well. “Even if the teenager’s really smart, the 40-year-old has wisdom from having seen it many times.” So her working window, before AI handles routine work unassisted, is three to five years.
The hybrid model for agentic AI customer service
Hugo’s bet on agentic AI customer service is not on the AI itself. Instead, it is on the layer above it.
“The one thing that has never been wiped out in the whole wide world are supervisors. How do we become people that help AI triage?”
In practice, lower-complexity contacts get absorbed by agentic AI. Meanwhile, mid- and high-complexity work moves to humans who supervise, evaluate, and step in. A 2026-realistic split looks like this:
| Customer Service Task | Agentic AI Handles | Human Supervises / Owns |
|---|---|---|
| Password resets, order status, balance enquiries | ✓ | Spot-check QA |
| Refunds, returns, policy lookups | ✓ (within rules) | Escalations + edge cases |
| Multi-system account changes | Partial | Final action + audit |
| Complaints, churn risk, complex billing | Triage only | ✓ |
| VIP / enterprise account handling | Assist | ✓ |
| Model evaluation & guardrail testing | n/a | ✓ |
How to approach agentic AI customer service
Turning the hybrid model into action takes a clear playbook. Four moves separate the operators leaning in from those waiting it out.
1. Don’t hide
The Kodak risk for BPOs is not too much disruption. Rather, it is pretending the change is not coming.
Simone says, “If I don’t show courage in confronting it, the door is open for others that are willing to just come and exist in that paradigm.”
In practice, that means raising agentic AI with clients before they raise it first. So the operator who opens the conversation keeps the relationship.
2. Reframe the question
Move from “where can AI supplement our service?” to “where can our service supplement your AI?” Buyers in 2026 are increasingly agentic-first. Many have already piloted internally, and they want partners who fit around their stack. Lead with that framing, and you become a complement to the buyer’s AI roadmap, not a casualty of it. This is also reshaping how firms think about AI outsourcing as a whole.
3. Pilot on lower-stakes flows first
McKinsey analysis shows scaled agentic adoption is concentrated where errors are cheap, such as IT, software engineering, and knowledge management.
Customer service follows the same logic. So scope agentic AI by intent. First, start with high-volume, low-risk flows like password resets and policy lookups. Next, prove reliability. Then expand the agent’s remit. The same caution applies to wider generative AI rollouts.
4. Rethink the pricing model
As agentic AI eats into billable hours, the per-FTE model erodes. The alternative is outcome-based pricing, which bills per resolution or per outcome delivered. Still, no one has fully solved it yet.
Simone concedes Hugo is still working it out: “If we create value, how can we capture some of it? That’s always a discussion.” As a result, the BPOs that design new commercial models early will set the benchmark.
FAQs
Is agentic AI customer service already replacing call centre jobs?
Some routine contacts, yes. However, replacement claims outpace what has shipped. Gartner’s Hype Cycle places agentic AI at the Peak of Inflated Expectations, with only a minority of organisations having deployed AI agents so far.
How fast will agentic AI customer service mature?
Gartner projects that agentic AI will autonomously resolve about 80% of common customer service issues by 2029. Meanwhile, most operators expect a three-to-five-year transition window where hybrid models dominate.
What roles will survive in BPOs?
Supervisors, escalation handlers, model evaluators, and owners of high-complexity workflows will survive. First to shrink is the lower-complexity tier.
Should BPO clients bring everything back in-house?
Not necessarily. Building, governing, and supervising agentic AI is itself becoming a service you can outsource. So the smart question is who supervises the AI, not where it sits.
How should BPOs measure agentic AI customer service?
Track resolution rate, escalation rate, and cost per outcome rather than seats filled. In addition, watch quality and trust signals closely, because a wrong action costs more than a wrong answer. These customer experience statistics can help you set a baseline.
Independent
