Share
The filter AI creates is also the pressure you haven't prepared for
AI in a contact center does one thing well. It resolves the interactions that don’t require judgment: policy lookups, payment status checks, renewal reminders, password resets. When it works, customers self-serve and agents are freed up. That’s the pitch. The operational reality is what the pitch leaves out.
What remains after AI filters out the routine is everything insurance CX is actually measured on:
- Coverage disputes after a denied auto claim
- Policyholders two weeks into FNOL on a total loss who are confused, scared, and ready to call a lawyer
- Fraud edge cases where a licensed agent has to make a coverage determination that could expose the carrier to bad faith litigation if it goes wrong
These are not easier calls because AI handled the easy ones first. They are harder calls, and they are landing on agents who in many operations were hired and trained for a much simpler environment.
The pressure to deploy AI is coming from the top, and it is legitimate. The problem is that most carriers are treating AI deployment as the destination rather than the beginning of a harder execution problem.
The proficiency trap that AI makes more expensive
There is a structural problem in US insurance contact center operations that predates AI by years: attrition running near 30% annually, sustained across most of the last decade. This is not a post-pandemic anomaly. It is a durable operational condition.
In insurance specifically, where an agent needs months to reach genuine proficiency in claims adjudication, subrogation procedure, multi-state regulatory compliance, and fraud detection, 30% annual attrition means perpetual rebuilding. Carriers are continuously training agents who leave before they generate value. The investment in getting an agent to proficiency walks out the door before it compounds.
AI does not solve this. It reveals it. When AI handles everything routine, the agent population that remains is handling a portfolio of interactions where the cost of proficiency failure is highest. A non-proficient agent handling a billing inquiry is a friction problem. A non-proficient agent handling a coverage determination on a disputed claim is a legal exposure and a retention catastrophe.
The math does not work unless the talent side of the equation receives the same investment as the technology side. Most carriers have not made that pairing.
What the carriers getting this right are actually doing
The architecture that works is not complicated to describe. It is difficult to execute. AI handles self-service and routine inquiry. Non-licensed agents handle transactional support, guided by real-time AI assist tools built on a unified knowledge base. Licensed agents focus on the complex, high-stakes decisions that require human judgment, regulatory expertise, and the kind of empathy that insurance customers in crisis specifically need. Intelligent routing keeps each tier operating at the right level of complexity. The knowledge infrastructure connecting all three tiers is unified, so an agent picking up an AI escalation has full context rather than a cold start.
What separates the carriers executing this well is the discipline applied to the talent layer. Career pathways from non-licensed to licensed roles change the attrition dynamics because agents see a development track, not a ceiling. Training frameworks differentiated by agent tier address the proficiency gap at its source: a billing inquiry agent and a claims adjudication specialist are not doing the same job and treating them as identical in onboarding is precisely how proficiency gaps form. Speed-to-proficiency is measured as a business metric, tracked against downstream customer outcomes, not filed in an HR report.
When a Fortune 500 insurance group brought ResultsCX in to manage non-licensed operations across three brands, this architecture was built correctly from the start. The results: 400% headcount growth in five months, 87% training throughput, a 2% improvement in speed-to-proficiency within the first 90 days, quality targets exceeded in under 60 days, the client’s internal channels outperformed on CSAT within eight months, and cost-to-serve reduced by 39.5%.
That is not an AI story. It is an execution story that AI supported.
The question worth asking before the next AI deployment
If you are investing in AI-powered contact center capabilities, the ROI depend almost entirely on what receives the escalations. The technology performs. The outcomes are determined by the agents, the training, the routing logic, and the career architecture underneath it.
The question is not how much AI to deploy. It is whether what AI hands off to is ready to receive it.
Three actions you can take now
Audit what AI is actually sending to your agents. Map the interaction types AI is currently escalating to human agents. What percentage requires licensed agent judgment? What is the proficiency level of the agents receiving those escalations today? Most carriers do not have a clear answer to this. Getting that answer is the first step in understanding whether the AI investment is compounding or exposing a capability gap.
Separate your agent tiers in training and measurement. If non-licensed and licensed agents are moving through the same onboarding program and evaluated on the same metrics, there is a structural misalignment in the operation. Speed-to-proficiency for a licensed agent handling claims adjudication is a materially different investment than for a billing inquiry agent. Differentiated frameworks are not a training upgrade. They are operational corrections.
Evaluate career architecture before the next deployment cycle. Attrition at the agent tier that receives AI escalations is not an HR problem. It is a technology ROI problem. Carriers that have built licensed-track pathways from non-licensed roles report materially different retention and proficiency trajectories. The investment required to build those pathways is a fraction of the cost of perpetually rebuilding a proficiency layer that keeps walking out the door.
Read our detailed whitepaper (authored in collaboration with ISG), to learn more about how you can transition your CX operations to a preventive and predictive model.