Organizations now trust AI agents to handle more customer interactions than ever, and the efficiency gains are real. But for customers, the experience disappoints, with our latest Consumer Trends research showing AI for customer service has the highest failure rate of any AI application, underperforming on key metrics like convenience, time savings, and usefulness.
In our AI-first world, the customer experience organizations deliver is where they win or lose. To help organizations navigate this shift, we surveyed 7,000+ consumers across seven countries and seven industries about their most recent customer service interactions. From this, we discovered that while AI agents are scoring well on friendliness they’re falling short on understanding — the dimension most tied to actually resolving the issue.
We built a framework to measure this: the Agent Performance Framework, or AP3. It tracks three agent-specific behaviors — friendliness, knowledge, and understanding — across both human and AI channels.
When agents fail to understand the problem customer satisfaction suffers
When issues go unresolved, customer satisfaction drops to 2.83 — a score and experience that puts the entire relationship at risk.
The leading cause of unresolved issues? A lack of understanding. And this doesn’t just apply to your human channels, but your AI agents, too.
When agents don't resolve issues, consumers rate their understanding 37% lower (from 4.44 to 2.78) — a steeper drop than knowledge (34%) or friendliness (20%).
An agent can be friendly and knowledgeable, but without genuine understanding of the issue resolution is unlikely. And unresolved issues are a business problem, driving churn, complaints, and repeat contacts.
“Without context, AI risks causing more friction than it solves. Satisfaction plummets when there’s a lack of understanding that prevents issues from being resolved.”
AI scores highest on friendliness and lowest on understanding
Across every channel and demographic in our study, consumers rate agents as friendlier (4.36) than knowledgeable (4.17) or understanding (4.16). But the gap is widest in AI.
AI-handled interactions tied with social media for the highest friendliness score of any channel we measured, scoring 4.41. But on understanding — the dimension that captures whether the agent truly grasped the customer's issue — AI scored just 4.08.
The 0.33-point gap between AI's friendliness and understanding scores is the largest spread of any behavior we studied.
“Your AI agent might be polite, but our research shows it lacks the context to turn the experience into a meaningful outcome for the customer.”—Leonie Brown, Head of Qualtrics Labs
The urgency here for businesses is that we found the consequences of a poor experience may be more severe for AI than for other channels. This is because the social contract between a customer and an AI agent is fundamentally different from the one between a customer and a human.
Friendliness from a human feels genuine. Friendliness from an AI agent that can't back it up with understanding feels hollow and patronizing, and is more likely to fuel frustration for consumers.
Younger consumers are harder to satisfy on understanding
Understanding shows the greatest variation by age in our data, trending upward with age. In plain terms: Younger consumers — the ones most likely to interact with AI-powered channels — are the hardest to satisfy on the understanding dimension.
They have higher expectations for being heard and lower tolerance for interactions that feel formulaic. As these consumers become the primary customer base, the understanding gap isn't going to close on its own. Uncovering and fixing bad experiences will be increasingly important to ensure this customer base keeps choosing you.
The industries deploying AI fastest aren't scoring best
Scores across industries show that speed of deployment doesn’t win — context does. We found that industries deploying AI agents most aggressively were among those lagging in performance. Telecommunications scored lowest on understanding (4.05), while retail received the lowest overall CSAT (4.07).
At the other end, insurance (excluding healthcare) leads in every dimension, with the highest scores for friendliness (4.55), knowledge (4.42), and understanding (4.40) in our study. Travel and hospitality rank second.
How to improve the experiences your AI and human agents deliver
For experience management leaders evaluating or expanding AI across their customer-facing channels, the research points to four immediate priorities to ensure your investment pays off:
- Enable AI agents with the context of past experiences, preferences, and purchase history so they can provide personalized, helpful solutions at scale.
You might feel like you're listening to your customers and reducing the cost to serve by implementing an AI agent as your first line of defense. But if it lacks understanding, satisfaction — the foundation for customer loyalty and repeat spend — will plummet. - Measure what matters beyond containment. Containment tells you the AI handled the interaction. It doesn't tell you whether it handled it well. Layer in agent effectiveness metrics — particularly understanding and issue resolution — to get the full picture.
- Design for the handoff, not just the deflection. The future of AI in customer service is intelligent collaboration between AI and human agents. Context must carry over, ensuring the customer doesn’t have to start from zero. So make sure the handoff itself is measured as its own moment of truth, too.
- Keep measuring after you deploy. AI agents aren't static — the conversations they handle evolve, customer expectations shift, and performance can degrade in ways that won't show up in a quarterly review. The organizations that iterate fastest will outperform the ones that deployed fastest and thought the job was done.
For more best practice tips, watch our guide to customer experience benchmarking
Leonie Brown is Head of Qualtrics Labs, where she leads research into emerging customer and employee experience methodologies.
Data cited in this article is sourced from the Qualtrics 2026 Agent Effectiveness Benchmark Study (n=7,001 consumers across 7 countries: Australia, Canada, France, Germany, Mexico, United Kingdom, and United States and 7 industries: Financials, Insurance excluding healthcare, Local/national government agencies, Retail, Telecommunications, Travel/hospitality, Utilities; conducted December 2025 – January 2026).
