AI is a tool, and a good one. In the contact center and in insurance operations it earns its place when it clears one specific hurdle: the seam between the strategy deck and the operations floor. That seam is where the value is won, and it is the part the demo never shows.
The research shows how much room there is to do this better. MIT's 2025 study of enterprise AI found only about 5 percent of task-specific generative AI tools had reached production, and the ones returning value were built into a real workflow rather than bolted beside it. Gartner expects more than 40 percent of agentic AI projects to be canceled by the end of 2027, mostly the ones launched on hype rather than a clear business case. Deloitte's 2026 insurance outlook found 90 percent of insurance executives see the urgency of pairing people with machines, while 25 percent have started. The intent is there. The value shows up when the operation catches up to the intent.
I have spent most of a 23-year career on the operations side of that seam, running contact center and BPO delivery with the P&L attached, and more recently building and evaluating the commercial insurance tasks that a frontier AI lab tests its models against. From both seats the pattern is the same. A pilot handles the clean cases and demos beautifully. The return lives in the harder minority: the exceptions, the escalations, the edge cases, the attrition, the quality assurance, the customer who is already angry. Get AI working in that part of the operation and it pays for itself, and that is exactly the part the deck leaves out.
One idea does most of the work here, and it is the same in CX as in insurance. Place AI by where a mistake surfaces. Where an error shows up fast and is cheap to catch, an assistant or an agent earns its place easily. Where an error surfaces late and is expensive to unwind, a trained human stays on the decision and AI supports them. I made that case for insurance in AI agents for commercial insurance, and the logic does not stop at insurance. The question, as I put it in the right mix of AI and human labor, is not whether to use AI. It is where AI does the work well, where a person still does it better, and how to design the split by outcome rather than by automation rate.
In the contact center, that line is easy to draw once you have worked a floor, and AI does real work along it. Intent routing, knowledge retrieval with a live agent in the loop, and after-call summarization are places where a mistake shows up in seconds and someone catches it, so the tool adds speed with a safety net under it.
The judgment call is the complex or emotionally charged issue, because there a confident wrong answer does not surface right away. It surfaces later as a complaint or a chargeback. That is an argument for grounding the model in real documents and keeping eyes on it, however, not for leaving the tool on the shelf. The gap shows up in the field: a 2026 survey of 815 enterprise CX leaders, conducted by Ryan Strategic Advisory for TELUS Digital, found roughly 60 percent of enterprises now run AI-assisted agents in their top customer-facing functions, while 32 percent have AI-powered quality assurance behind them. The opportunity is in closing that second gap, so the operation can see how the AI is doing and let it do more. A trained human agent is a trained human agent no matter where they are, and the best results come from pairing the two.
Insurance operations sort the same way. Submission intake and triage are fast-surfacing, and AI does real work there today. Underwriting is the largest prize and the clearest example of assist rather than replace: Accenture and The Institutes found underwriters spend about 70 percent of their time on work that is not underwriting, and 40 percent on pure administration. Give that time back with AI decision support and you have moved the number that matters, with the underwriter still owning the risk decision. Claims, including first notice of loss, and policy servicing follow the same rule. Automate the intake and the routing where errors surface fast, and keep a person on the coverage decision and the denial, where the call surfaces late and carries the most weight.
This is why the seam is an operator's job as much as a strategist's. Deciding where AI goes is not only a technology question. It is a question of where your operation's mistakes show up, how fast, and who absorbs them. You answer it best if you have run the operation and watched those mistakes travel. Cost is never the wage rate alone, and the same logic applies to an AI deployment: the real math includes the rework, the oversight, and the escalation path, alongside the very real savings when the tool is placed well. Count the whole stack and the good deployments still win. You get what you pay for holds here too.
Used where it belongs, AI lowers cost, speeds production, and raises quality. However, all three at once, promised up front, is a vendor-shaped claim, so I grade them separately. Faster is the surest of the three. Better usually comes from giving people more coverage, not from replacing them. Cheaper takes the most discipline, because the savings on the slide only hold if the rework does not quietly reappear downstream.
My read on running it well is simple enough. Start where mistakes surface fast, keep a person on the decisions that surface late, and measure against the operation's real numbers rather than the demo's. In CX that means containment that holds CSAT, and cycle time that holds quality. In insurance it means loss ratio, cycle time, and first-contact resolution. There is no perfect answer for every operation; it is a mix, and the right mix is specific to the client and the situation.
The deployments that make it, in the MIT data and in my experience, are the ones built into the workflow rather than beside it, with the visibility to see how they are doing once they are there. That is the seam. Close it and AI does what a good tool should: it makes the operation measurably better. That is well within reach, and it is worth building toward deliberately rather than in a rush.
If you are weighing where AI fits in your contact center or your insurance operation, it is worth at least having the conversation.
Figures cited above, with the caveat that survey dates and definitions vary by source. Treat them as directional.
Tell us what you are trying to fix, scale, or evaluate. We will give you an honest read on whether we can help.
Contact us