Blog

The Skyrocketing Costs of Agentic AI: What Are We Doing Wrong?

Written by USAN Team | Nov 16, 2026, 2:19:56 PM

Have you seen the headlines about AI costs lately? Many are highlighting a problem that many organizations are experiencing after deploying AI voice agents: unpredictable costs.

For years, businesses were sold on the idea that AI agents would make operations cheaper, faster, and easier to manage. But now, many of those same organizations are seeing a reality that looks very different from what they expected. In fact, one of Goldman Sachs' recent research reports highlighted just how tricky the economics of AI in contact centers can get. Its analysis found that running a real-time voice agent could actually cost more than a human agent.

So, how did we get here? The answer comes down to the difference between what vendors advertise and what it actually costs to run an AI-powered conversation with a real customer.

Having AI handle a real-time customer interaction can be much more complicated than it sounds. It can involve speech-to-text, text-to-speech, speech-to-speech, one or more language models, reasoning, retrieving information from enterprise knowledge bases, API calls, orchestration, monitoring, retries, and potentially a transfer to a human agent. So many moving parts, right?

This is exactly what many have failed to account for. All these costs aren't necessarily reflected in a simple AI model price, or a demo built around a narrow set of customer intents. Us humans are much more complicated than that. We take unexpected paths, integrations fail, edge cases show up, and some interactions that were supposed to be automated end up involving human agents anyway.

This whole journey brings us to one important question: What does it actually cost to successfully resolve a customer interaction?

Back to Basics: How Are You Calculating Agentic AI Costs?

At this point, it's important to understand that not all AI solutions are priced the same way. Some organizations use component-based pricing or outcome-based pricing. Others bundle much of the consumption into a simple unit, such as minutes, while some offer an all-in consumption price.

We're not here to tell you there's one perfect pricing model that will guarantee predictable costs. The model that makes the most sense for your organization may depend on your needs, your usage, and how much cost uncertainty you're comfortable taking on. But a pricing model is only one part of the calculation.

The actual cost of running an AI agent can also change depending on how it handles a conversation. How long does the interaction last? How many systems does the agent need to access? Does it need to retry an action, retrieve additional information, or transfer the customer to a human? These details can all influence what it takes to complete an interaction, and they aren't always obvious when looking at a price per minute or a model's advertised cost.

This is why human agents and AI agents aren't interchangeable widgets. Comparing their costs can quickly become a complicated exercise because they operate and handle interactions differently and bring unique capabilities to the customer experience.

Just because an AI agent can handle a conversation at a lower cost doesn't necessarily mean it has resolved the customer's issue. If the customer still needs to be transferred to a human agent, for example, you need to account for the cost of both.

So, with so many different AI pricing models and no straightforward way to compare AI costs directly with human labor, how can contact centers better understand, predict, and manage their Agentic Voice AI costs?

Explore the New Economics of Agentic Voice AI

To help answer these questions, we've partnered with Opus Research on our new whitepaper, The New Economics of Agentic Voice AI: The True Cost Per Resolution.

In this research-backed paper, we take a closer look at what really drives the cost of Agentic Voice AI and what organizations should consider before deploying and scaling these solutions.

Here's what you'll learn:

  • Why traditional AI pricing models can fall short when it comes to understanding the full cost of an interaction.
  • Why two seemingly identical conversations can have vastly different costs, depending on what happens behind the scenes.
  • How predictable pricing supports technology changes as AI solutions evolve.
  • What managed services change about ongoing responsibilities and the work required to keep AI solutions running.
  • How monitoring and tuning affect resolution costs and the overall performance of your AI agents.
  • Five questions to ask before scaling Agentic Voice AI to make more informed decisions about your investment.

Get your copy and explore the new economics of Agentic Voice AI.