Everyone wants to know whether AI costs less than a human. Far fewer ask what it’s worth.
Charlie Mitchell’s recent CX Foundation article, “The Economic Risks of AI in Customer Support,” raises questions every customer experience leader should be asking. AI pricing is shifting, consumption models make costs harder to predict, and implementation costs raise the overall expense of agentic AI solutions. Together, these mean the cost of an AI agent is pliable, and whether it’s cheaper than a human depends on the model used, the work it does and how the solution is designed.
Too often, the conversation stops at the head-to-head cost comparison between AI and human labor. Cost matters, but it’s only part of the question. We should also be asking what combination of AI and humans generates the greatest value from the work.
Regardless of which costs less, AI or humans, the value generated by your AI investments must outpace costs over time. And the architecture behind your AI has to do two things: keep costs under your control and create value. That’s where I think we’re still stopping short on the economic analysis of AI.
Getting beyond the cost comparison
There are really three business cases hiding inside most AI discussions.
Substitution: Can AI perform this work for less than a person can? This is where most AI value conversations begin and end.
Productivity: Can AI and a person together do the work better than the person alone, with higher accuracy, faster resolution, and fewer repeat contacts?
Enterprise value: What happens when the capacity created by the first two is deliberately reinvested in work that protects revenue, reduces risk, improves loyalty, or creates growth?
I believe the real competitive advantage comes from mastering the third. The article points the same way, urging contact centers to move from cost center to revenue center. That is exactly the right destination.
Start with the work, not the automation rate
At Genesys, we’ve been exploring the full AI value equation through the AI Value Intelligence Model, or AVI. Instead of asking how much work AI can automate, we start with the work itself and ask what role AI should play.
Routine interactions are simple enough for AI to handle with significant independence. Contextual interactions are ones where AI does a great deal of the work, but the customer situation still benefits from human involvement. Consequential interactions keep a human accountable, because the financial, regulatory, relationship, or trust stakes are too high to reduce to an automation exercise.
Case in point: Value intelligence for an insurance provider
We recently applied AVI to a large insurance contact center handling more than four million voice interactions a year. It looked like a major automation opportunity, and it was, but not the kind a traditional business case would suggest.
Only about 18 percent of volume was Routine. More than 80 percent was Contextual. Just a tiny percentage was Consequential and kept under strong human ownership. That balance meant that the biggest opportunity didn’t lie in simply removing humans from interactions. In fact, approximately 56 percent of the capacity opportunity came from making the people who remained involved more effective, while roughly 44 percent came from automation.
Consider a roadside assistance scenario. A stranded customer isn’t always clear as they explain where they are, what happened, what the vehicle is doing, and what they need. AI can make sense of that context and move the interaction forward. But it’s the human who knows what to do next. AI doesn’t need to replace the employee. It needs to remove the friction around them.
Now compare that with a complaint, a liability issue, or a regulated financial transaction. AI can still provide tremendous value. It can collect information, summarize history, surface knowledge, document actions, and make the person handling the interaction far more effective. But the final accountability should remain human. That is not a failure of automation. That is good operating model design.
This is where pure cost comparisons become dangerous. If we judge every interaction by how many labor minutes we can remove, we’ll probably end up optimizing the wrong thing. A thirty-minute call looks attractive to automate. But what if it’s a cancellation where a skilled employee can preserve a valuable relationship, or a complaint with regulatory exposure? Eliminating human time there can destroy value.
The lesson here is clear: Don’t just right-size the technology. Right-size the role of the humans.
From hours saved to value created
Our study of the insurance provider identified roughly 190,000 hours of annual capacity in its base scenario. The easy business case would convert those hours into a labor savings figure. We asked instead what the organization could do with those hours if the humans were freed of the burden.
Could some go to customers with delayed claims before those delays become complaints? Could experienced employees focus on cancellation and retention moments? Those questions produced redeployment concepts focused on protecting relationships and creating growth, not only reducing labor.
To be clear, freed capacity is not savings on its own. Hours become value only when they are deliberately redeployed and measured against outcomes like retention, complaint rates, or revenue per interaction. That discipline is what separates the enterprise value case from a spreadsheet exercise.
Orchestration puts the full value equation to work
Clarity on direct costs is crucial, but it’s just part of understanding the economics of AI. A fuller assessment also accounts for how AI and humans create value, both separately and together. AVI helps us identify what role AI and humans should play in each interaction. Orchestration is how that decision gets made in real time – and how it delivers both lower costs and higher value simultaneously.
Too many AI deployments just point a large language model at every interaction in a particular use case and expect it to handle the whole workflow. But that approach ties your cost structure to a single, variable line item. Every simple confirmation consumes the same resource as your hardest conversation, and as usage or model prices change, your whole cost base moves with them.
That is a risk of architecture, not of AI itself. Most customer work does not need a frontier model. Some is best handled by simpler automation that’s fast, cheap and predictable. Some needs a virtual agent that can reason and act across systems. And some needs a person, supported by AI, because judgment and trust are the point. The question is not which to choose, but how to decide, step by step, which belongs where.
That’s the job of an orchestration platform. It coordinates human employees, agentic virtual agents and other automation tools as one system, with outcomes as the North Star: not containment rates or handle time, but cost and revenue per outcome. A claim resolved. A cancellation saved. It can match each task to the right model based on capability, cost, latency and risk, and decide in real time whether a step is automated or handed to a person with AI assistance.
Our AVI work with the insurance provider shows why this matters. Only 18 percent of the volume was Routine enough for AI to handle independently. For the 80 percent that was Contextual, the best answer was usually a person working with AI. An orchestration platform routes accordingly, controlling costs and creating value in each decision. It reserves expensive AI reasoning for work where it earns a return, uses simpler automation where it’s more cost-effective and sends complex conversations that require human judgment to an employee. Working within guardrails your team sets, the system decides, continuously, where people and AI should each spend their time. That’s how freed capacity gets redeployed and how hours saved become value created.
Understand the costs, but focus on the value
Yes, understand token and implementation costs. Yes, compare AI with human labor, because when used for the right work, AI can deliver savings. Yes, challenge whether advanced AI is necessary for the work.
And then keep going. Ask how your platform decides whether a flow, an agent, or a person handles each task, and whether it can change that decision as costs or conditions change. Ask whether you measure cost and revenue per outcome, not just cost per contact. Ask what capacity was created, where human judgment matters most, and where newly available capacity can create revenue that did not exist before.
The answer to the questions about AI costs raised in the article isn’t to hesitate. It’s to make sure value exceeds costs, on an architecture that keeps the cost curve in your hands and creates new value at the same time.
The companies that get the most from AI won’t necessarily be the ones that automate the most work or use the latest frontier models. They’ll be the ones most deliberate about what technology should do, what people should do, and where the combination, orchestrated around outcomes, creates value neither could alone.
