Where AI Actually Helps Customer Support — and Where It Doesn't
If you read the headlines, you'd think this argument was already settled. Gartner expects agentic AI to autonomously resolve 80% of common customer service issues by 2029. In the same breath, a Gartner survey of nearly 5,800 customers found that 64% would prefer companies didn't use AI for customer service at all, and 53% said they'd consider switching to a competitor if they knew a company was going to use it.
Both things are true at once. AI is genuinely good at some parts of support and genuinely bad at others, and the companies getting this right aren't the ones going all-in or holding out. They're the ones who figured out where the line sits.
Here's how we see it.
Where AI actually helps
The repetitive questions that never stop coming in. "Where's my order?" "How do I reset my password?" "What are your hours?" A large share of any support queue is a handful of questions asked thousands of different ways. Customers don't want a relationship for these — they want the answer, fast. Zendesk's research found 67% of consumers are ready to hand tasks like order tracking to AI. When AI clears this volume off the queue, it isn't replacing good support. It's freeing your people to spend time where time actually matters.
Behind the scenes, helping your agents work faster. This is the quiet win most customers never see. AI that drafts a reply, summarizes a long ticket history, or surfaces the right knowledge-base article the moment an agent needs it makes a good agent noticeably faster — without ever touching the customer directly. Gartner expects 73% of customer service organizations to have implemented agent-assist tools by the end of 2025, and it's one of the few AI use cases that landed in their top ten most valuable service technologies. There's a reason for that: it keeps a human in charge of the answer while cutting the busywork around it.
Routing, triage, and coverage when no one's awake. AI is good at reading an incoming message, understanding roughly what it's about, and getting it to the right place — or holding the line at 2 a.m. until a person can pick it up in the morning. Used this way, it shortens the path to a resolution instead of standing in front of it.
Self-service that actually finds the answer. Search that understands a messy, real-world question and returns the right help article is a real upgrade over a static FAQ page. Gartner expects self-service and live chat to overtake phone and email as the most valuable service channels by 2027. When customers can solve simple things themselves, everyone's happier.
The pattern here is worth naming: AI is strongest when the question is common, the stakes are low, and the answer already exists somewhere. That covers a lot of support. It just doesn't cover the part that decides whether a customer stays.
Where AI does not help
When the customer is upset, or the stakes are high. This is the clearest line in the data. In a 2025 SurveyMonkey study, 85% of consumers wanted a human for billing and payment disputes, and 78% for data security and privacy issues — versus single digits who wanted AI. When someone is worried about their money, their data, or a mistake that affects them personally, being handed to a bot reads as a company that doesn't want to deal with them. That's the exact moment a relationship is won or lost, and it's the exact moment AI is weakest.
When the problem is genuinely complex. AI is good at questions with a known answer. It struggles the moment a problem has three moving parts, a weird edge case, or a history that doesn't fit the script. Ironically, the better your self-service gets, the more the tickets that reach a person are the hard ones — which raises the bar for the human on the other end, not lowers it.
When it becomes a wall instead of a door. The single biggest fear customers report about AI in service is that it will make it harder to reach an actual human. They've all been trapped in a loop that won't let them out, and they remember it. In that same SurveyMonkey research, 89% of consumers said companies should always offer the option to speak to a person. An AI layer with no obvious way to a human doesn't save you money — it quietly costs you customers.
When being confidently wrong is expensive. AI can give a wrong answer in a completely convincing tone. For "what time do you close," that's a minor annoyance. For a return policy, a warranty question, or anything with legal or financial weight, a fluent wrong answer is worse than no answer — because the customer believes it. These are the interactions that need a person who can be held accountable for what they said.
The honest middle
It's worth being clear-eyed about how often these projects don't pan out. Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027, citing rising costs, unclear return, and weak controls. Gartner has even coined a term — "agent-washing" — for vendors rebranding basic rule-based chatbots as autonomous AI agents. A lot of what gets sold as transformative is a decision tree with better marketing.
None of that means AI doesn't belong in support. It means the tool is real and the hype is oversold, and the gap between those two is where a lot of money gets wasted.
What good actually looks like
The companies getting real value aren't choosing between AI and people. They're drawing the line deliberately: let AI take the repetitive, low-stakes volume and quietly assist the humans behind the scenes — then make sure a real person is easy to reach, and owns every interaction where trust, judgment, or emotion is on the table.
Even Zendesk, whose whole business is this software, frames it this way. Their CEO put it plainly: AI should be "in service to humans" — not a replacement for them. The goal was never to remove people from support. It was to make sure the people you do have are spending their time on the interactions that actually build loyalty, instead of drowning in password resets.
That's the part technology alone doesn't solve. You still need people who are trained, supported, and accountable — and a system around them good enough that AI makes them faster instead of replacing the judgment customers are actually paying for.
That's the part we care about most. Automate the routine, keep real people on the moments that matter, and never make a customer fight to reach one.
Sources: Gartner — Agentic AI to resolve 80% of common service issues by 2029 · Gartner — 64% of customers would prefer companies didn't use AI · Gartner — Self-service and live chat to surpass traditional channels by 2027 · Gartner — Over 40% of agentic AI projects canceled by 2027 · Zendesk — 2025 CX Trends Report · SurveyMonkey — Customer service statistics: humans vs. AI