A common question we hear from support teams is how to push back against automation goals that might risk harming the user experience. Deploying AI is part of their mandates, but the most important question is how to do it without letting quality slip?
Where the Deployment Enthusiasm Comes From
Teams are excited to capture the productivity and efficiency gains AI can bring to the table. The AI labs have done a great job of building generally intelligent models trained on massive datasets, configurable for a wide range of knowledge tasks and marketing them extensively.
However, it is important to be mindful of user expectations and likelihood that interacting with these models exclusively as representatives for your businesses can lead to major frustration if the experiences are poorly designed. People outside the AI bubble are building up a negative sentiment against AI primarily because much of the messaging since its inception has been fear-driven: the "job apocalypse," AI taking everyone's jobs, AI outperforming the majority of knowledge workers.
The Overemphasis on Resolution Rates
That framing has led to an overemphasis on maximizing AI resolution rates, pushing teams to overdeploy AI across all conversation types without discernment. Using an AI agent to help your customers is great in theory, especially outside of regular staffed hours but it’s important to ensure that the model can genuinely help the user because if not, then that is a poor interaction.
Interaction quality must be gauged and treated as a metric in its own right, and it must not degrade when AI agents are deployed. Not every team takes this into account. Most start with good intentions and do in fact want to help their customers; many simply want to reduce costs. The teams with CX leaders at the helm tend to be the more mindful ones.
The ones that aren’t should nonetheless ensure that their agents are properly configured in such a way that they do not create frustrating experiences for customers, because no one needs more meaningless interactions.
There are a few ways to ensure that is the case.
Two Simple Checks That Protect the Experience
Before engagement. Before the conversation gets going, the agentic AI solution should understand the customer's intent and verify that it has the requisite knowledge to meaningfully respond.
Always an exit. There should always be an exit or escalation path. If all you offer is AI support, or you use a product with no live handoff to a human agent, that's fine — the simplest exit path is instructing the customer to reach out via an alternative channel or file a ticket, if your agent can't create one automatically in the platform you use.
Our Philosophy: Meaningful Resolutions Only
We're not fans of lazy attempts at resolution. AI responses should only be attempted when they're meaningful to reduce the negative sentiment and maintain the quality of the interaction. L0 deflection, where users get a knowledge-base answer by default, is helpful but relying solely on knowledge retrieval produces poor outcomes. The right approach is to determine the customer's intent, check whether it falls within what the agent is configured to resolve, and then ask whether we can meaningfully resolve it — and only proceed when the answer is yes.