Brief
The premise for chatbots and voicebots powered by generative artificial intelligence is straightforward: faster service at lower cost. If a bot can solve issues quickly, it should create happier customers and reduce the load on expensive human agents. Reality, however, is proving different than this promise, and it’s not because the bots aren’t capable.
In banking, for instance, people reach out to customer service for both simple transactions (ordering checks, sending a wire) and for more complex issues (a fraudulent charge, an unexpected fee, a rejected payment).
Bots handle the simple stuff well. For example, NPS Prism® data shows that US banking and telecommunications consumers give fairly high scores for simple digital episodes—generally averaging between 40 and 50 on a scale of 0 to 100. Many companies have successfully migrated customers to apps and websites for resolving simple transactions. But for more complex digital episodes, scores fall dramatically, sometimes near or below 0.
What accounts for customer unhappiness with companies’ handling of complex digital episodes? Complex requests are tricky. Even some simple requests fail to resolve online because of a login problem, a confusing interface, or other technical shortcomings.
I want to speak with a person!
Customers’ expectations exacerbate the problem. When a customer disputing a fee reaches a human agent and the agent declines to reverse the fee, the customer then asks for a supervisor, who might also decline. It’s frustrating for the customer, who might get angry and yell a bit. But eventually they move on.
Imagine the same situation with a bot that declines to reverse the fee. Most people will not accept that decision. They feel they have to escalate to a human, if only to have someone to argue with. A no from a machine feels less legitimate than a no from a person, even if the outcome is the same.
This is the sticking point for many complex use cases. It’s not that the bot is worse than a person, but rather that the customer won’t accept a bad outcome from a bot.
A hybrid bot-human approach
To address this natural inclination among customers, companies should adjust their method of contact triage. When someone calls or chats, they can first communicate with a bot to explain the issue. For a straightforward, low-stakes issue, let the bots handle it. For a complex or high-stakes issue, immediately transfer to a human agent. That human will have an AI copilot to help them get up to speed on the issue and resolve it quickly.
This hybrid approach doesn’t just protect customer satisfaction and loyalty; it also makes the most effective use of AI. Human agents can lean on AI copilots to surface information faster, draft responses, and handle routine parts of a workflow. The customer speaks with a real person when needed, and that agent is much more efficient with the support of AI.
Over time, the balance of calls that require a human agent will likely shift, because customers could very well adjust their expectations. If experience shows that a human agent does not deliver a better outcome, people may begrudgingly come to accept the bot’s answer and move on. That adjustment could take many years, though, and in the meantime, companies risk alienating customers by forcing them into unsatisfying interactions.
The upshot today: Companies without robust online self-service options can dramatically lower costs and improve satisfaction by using AI-supported bots. But that doesn’t mean cutting customer service agents. Forcing bots into situations where customers won’t accept them will only foster costly discontent. Instead, the fruitful path to effective customer service starts by understanding each customer episode, then mapping the pain points and matching the channel to the need.
NPS Prism®
Our cloud-based customer experience benchmarking service provides actionable insights and analysis that guide your creation of game-changing customer experiences.
First published in September 2025