Automated Customer Service: When to Use a Bot and When to Hand Off to a Person

In the last few weeks, automated customer service has gained a bad rap in recent news. In September 2026, a study of 2,009 U.S. consumers by survey was conducted by software company Alchemer. The study found that 89% said AI had hurt their experience with a company in some way. The top complaint, named by 44%, was difficulty reaching a person when they needed one. On September 28, the state of California signed AB 1609. The bill stated the requirement of large businesses to make a good-faith effort to connect a customer to a human within 15 minutes of asking.

None of this says automation is a mistake. Half of the customers in Gartner's survey say their interactions are easier when companies use GenAI. The message is narrower and more useful: automate the simple work, and never make a person hard to reach.

KDCI's Position: Automate Customer Service, But Keep a Person Within Reach

Our view is simple. Automate what a bot can answer with confidence, and always offer live support within a set coverage window that customers can see.

That second half matters as much as the first. A bot that hides the human option protects a dashboard metric and costs you the customer. A bot that answers the easy questions instantly, then hands the hard ones to a person who already has the full conversation. Additionally, it gives customers speed and care at the same time.

The coverage window is a business decision, not a technical one. Some stores need a person available around the clock during a holiday rush. Others are fine with business hours and a clear promise of when someone will reply. What doesn't work is a chat that loops forever with no way out. Extended hours and coverage across time zones is one of the most common reasons companies tell our sales team they're looking for support help.

What Automated Customer Service Does Well, and Where It Breaks

Knowing how to automate customer service plays an integral part of support that runs without a person handling it at the moment. Automations such as chatbots and AI agents, autoreplies, order-tracking lookups, routing rules, and self-service help centers are some examples. For an online store, the strong cases are obvious. "Where's my order?", "What's your return window?", and "Do you ship to Canada?" have one right answer that lives in a system. A bot can give it at 2 a.m. without making anyone wait.

The trouble starts at the edges. Alchemer's study lists the failures customers notice most:

• Can't reach a person: 44% say AI made this harder.
• Answers that miss the question: 37% got a reply that didn't understand what they asked.
• Repeating themselves: 30% had to give information they'd already provided, often after a handoff to a person.
• Slower resolution: 19% say AI made their issue take longer to solve.

Support teams see the same pattern from the inside. Recent online discussion has sparked interest on whether AI has actually helped support, one support lead described celebrating a quarter of strong deflection numbers. Interestingly, it then noticed the same customers came back days later asking for a person. The discussion demonstrated that automation hadn't solved anything, as it had only moved the wait. Another commenter with decades in support reframed the question: how much repetitive work can AI take off the humans?

That gap between "handled" and "solved" is where most automation programs go wrong.

When to Use a Bot and When to Hand Off to a Person

Gartner's Eric Keller put the core rule plainly: "Service leaders should not use GenAI as a mandatory first step for every issue." In practice, the split for an ecommerce store usually looks like this:

These triggers match the ones we’ve set up for clients. They include: a customer asking for a person, signs of frustration, sensitive topics, refund limits, and questions the AI can't answer with confidence. The client approves every rule.

What This Means for Ecommerce Support Teams

Deflection is the wrong scoreboard. A chat that ends without escalation looks like a win, even when the customer emails the next day about the same problem. Alchemer recommends judging automation on customer effort, repeat contacts, and resolution rather than containment alone. Track reopen rate, repeat contacts within a few days, and how often customers have to repeat themselves after a handoff.

The handoff is part of the product. Thirty percent of customers in Alchemer's study had to repeat information. A good handoff passes the whole conversation and a short summary, so the person picks up where the bot left off.

The legal bar is rising. AB 1609 applies to businesses with more than $500 million in annual gross revenue that serve California customers, according to Law Commentary's summary of the law. Most online stores won't meet that threshold. But it puts three expectations in writing. First is a simple way to ask for a person during business hours. Next is a quick connection or a scheduled callback. Lastly, a clear disclosure when a customer is talking to a bot that could be mistaken for a human. Building those habits now costs little, wherever the law ends up.

Customers will tell you what they value. In Alchemer's study, 43% said they'd pay more for a product or service that guarantees access to human support. A visible human option is part of what you're selling.

What to Do Now: How to Automate Customer Service Without Losing the Human Path

1. List your top 10 contact reasons. Pull them from your helpdesk tags or a week of tickets. Mark which ones have a single, system-based answer.
2. Automate only the high-confidence ones first. Order status, tracking, shipping zones, and policy questions are the usual starting point.
3. Write your handoff rules. Use the triggers in the table above, and add a hard rule: any customer who asks for a person gets one.
4. Set and publish your coverage window. Decide when a person is available, say so in the chat, and offer a callback or reply time outside those hours.
5. Pass context, every time. Make sure the person taking oversees the full conversation and a summary before saying hello.
6. Tell customers when they're talking to AI. It sets honest expectations, and some places now require it.
7. Review escalations monthly. Look at why handoffs happen and feed fixes back into your help content so the bot gets better over time.

At KDCI, this is how the human-in-the-loop escalation team works: trained support specialists step in when a customer asks, the AI gets stuck, or a decision needs a person. Once they see the full conversation, customers never have to start over. They also spot-check AI replies, fix knowledge gaps, and report on why escalations happen. If you don't have an AI agent yet, our AI voice and chat agents team builds one with handoffs built in. If we’re asked whether an AI agent can replace a support team, our answer is: "Not fully, and we don't recommend trying."

Where's the line for your store? The answer will shift as the tools improve, and the stores that review their handoffs every month will be the ones who find it first.

Keep a Person Within Reach

At KDCI, we build customer support teams that work in your tools, cover the hours you choose, and take over when your bot is overloaded. If you're planning how automation and people fit together in your support, we're happy to talk it through.