The Human Contact Gap: What the Research Says About Customer-Facing AI
· 5 min read
By Conscientia Labs
Customer-facing AI is everywhere: support chat, voice agents, assistants that answer on a company's behalf. Customers want what it promises, which is fast, always-on help. Yet independent research keeps finding the same gap. AI that sounds polished often fails at the moment that matters most: understanding the person in front of it and actually resolving their problem.
We call this the human contact gap. This article collects the consumer studies, enterprise surveys and industry analysis that measure it.
These studies document a business problem that Connie is designed to address. They did not evaluate Connie, and their authors do not endorse any product, including ours.
Friendly is not the same as understood
The Future of Customer Service: Measuring and Improving Agent Performance Across AI and Human Channels
Qualtrics · June 18, 2026 · 7,001 consumers
AI service interactions scored 4.41 out of 5 on friendliness, tied for the highest of any channel, but only 4.08 on understanding. Understanding was the dimension most closely tied to actually resolving the issue: when an issue went unresolved, understanding ratings fell from 4.44 to 2.78.
Why it matters: Conversational polish does not translate into contextual understanding or good outcomes. Sounding helpful and being helpful are measured differently.
AI-Powered Customer Service Fails at Four Times the Rate of Other Tasks
Qualtrics · October 7, 2025 · More than 20,000 consumers
Nearly one in five consumers who used AI for customer service saw no benefit from it, a failure rate almost four times higher than for AI use in general. The study also found concerns about personalisation, privacy and control over personal data: 53% named misuse of personal data as their top concern, and only 39% trust companies with their data.
Why it matters: Better context, transparency and customer control are not nice-to-haves. They decide whether AI service works at all.
People still want a human within reach
Gartner Survey Finds 87% of Customers Say Companies Using GenAI for Customer Service Must Provide Access to a Human Agent
Gartner · August 4, 2026 · 3,566 customers
Half of the customers surveyed said their interactions are easier when companies use generative AI. At the same time, 87% said it is essential for those companies to offer an option to reach a human agent.
Why it matters: Convenience does not replace choice. AI-assisted experiences need to keep a person reachable and keep the customer in charge of the path they take.
Read the Gartner press release
Gartner Survey Finds Only 27% of Customers Would Try a Chatbot Again After a Negative Experience
Gartner · September 2, 2026 · 3,566 customers
A bad chatbot experience sharply reduces customers' willingness to use one again: only 27% would try again. Gartner recommends reliable issue resolution and accessible human support.
Why it matters: A poor AI interaction does not just fail once. It damages long-term adoption and trust.
Read the Gartner press release
The resolution problem
Consumers Prefer "Always-On" AI Customer Service, But Only When It Can Resolve Their Issue
Ada, research conducted by NewtonX · March 24, 2026 · 2,000 consumers and 500 enterprise decision-makers
Only 24% of consumers reported that AI alone fully resolved their issue. Yet 59% said they prefer instant, around-the-clock AI support to waiting for a human, but only when the technology can actually resolve their issue.
Why it matters: Demand for AI service is real, and so is the gap between that demand and what AI delivers today.
Governance, reliability and return on investment
The AI Production Paradox
Sinch · May 13, 2026 · 2,527 enterprise decision-makers · Vendor-sponsored survey
74% of the enterprises surveyed said they had rolled back or shut down an AI customer communications agent after deployment because of a governance failure. At the same time, 98% reported increasing their investment in AI communications in 2026.
This is a survey of self-reported deployment experience, commissioned by a vendor. It describes what respondents went through, and should not be read as a general failure rate for AI.
Why it matters: Organisations are investing more while discovering that dependable behaviour after launch is the hard part.
Customer Service Leads on AI Adoption but ROI Remains Elusive
Gartner · July 13, 2026 · Enterprise research (subscription)
Customer service organisations are among the fastest adopters of AI, but in Gartner's summary AI is increasing productivity without reducing costs, so financial returns remain hard to demonstrate.
Why it matters: The measure of AI service should be outcomes, not how much activity has been automated.
View the Gartner research summary
Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027
Gartner · June 25, 2025 · Industry forecast
Gartner forecasts that more than 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls.
This is a prediction, not a measured failure rate.
Why it matters: It shows why enterprise AI needs a stronger emphasis on reliability, governance and measurable value.
Read the Gartner press release
AI Agent Success Requires Reliability Before Autonomy
Gartner · October 2, 2026 · Industry analysis
Gartner argues that enterprise AI agents need clear accountability boundaries, and should earn greater autonomy only through demonstrated reliability.
Why it matters: Autonomy has to be earned, and people have to stay in authority while it is.
What this means for how we build Connie
The research describes AI that is friendly but often misunderstands, quick but often fails to resolve, and widely adopted but hard to trust once deployed. Customers respond by wanting a human within reach, and organisations respond by rolling back what does not behave.
Connie is built for that moment of human contact:
- Understanding over polish. Connie works from the context of the conversation and the knowledge it has been given, rather than optimising for a pleasant tone.
- Human authority by design. People decide what Connie does on their behalf, and stay in charge of the outcome.
- Behaviour you can rely on. Clear boundaries on what Connie will and will not do come first; more autonomy comes after.
None of these studies evaluated Connie. Together they set the bar that customer-facing AI has to clear, and the bar we measure our own work against.