Customer service was the first business function AI chatbots disrupted at scale. By late 2023 every vendor was promising "AI agents that resolve 90% of tickets." Three years on, the reality is more nuanced — and in some ways more interesting — than the hype suggested.
The thing that actually worked
AI chatbots turned out to be excellent at the unglamorous middle layer of customer service: the tickets that are too complex for a static FAQ but too repetitive for a human to enjoy. Password resets with a twist, order-status queries with edge cases, refund requests within policy, simple product questions that require reading two paragraphs of documentation. This is now genuinely automated at most well-run companies, with resolution rates in the seventy-to-eighty percent range for these categories.
The thing that did not work
"Replace the entire support team" was always optimistic and remains so. The hardest tickets — the ones where someone is upset, the policy is ambiguous, or the customer's expectation is genuinely off — still need a human. And the better the AI gets at the easy tickets, the harder the average remaining ticket becomes. Support agents in 2026 spend their day on the difficult cases, which is rewarding work but also exhausting.
The unexpected win — internal knowledge
One of the biggest gains turned out not to be customer-facing at all. AI chatbots trained on internal documentation now serve as a kind of "always-on senior agent" that human agents can ask in real time. New hires get up to speed in weeks rather than months because the chatbot can answer "have we ever dealt with X before, and what did we do?" instantly.
What good 2026 implementations look like
The companies doing this well share a few habits. They route AI-handled tickets through a human-readable transcript review, so quality stays measurable. They make the handoff to a human seamless and never adversarial — no "are you sure you do not want to keep talking to me" loops. They publish what the AI can and cannot do, so customers' expectations match reality.
The metric that matters
Resolution rate is the number every vendor sells on. The number that actually matters is repeat-contact rate — how often the same customer comes back about the same issue within a week. If repeat-contact is rising, the AI is "resolving" tickets that were not actually resolved. This is the failure mode to watch.
