Gartner Finds 87% Needs Human Access in AI Customer Support

Customers are increasingly willing to use generative AI for service, but they do not accept automation that removes control over when they can reach a person.

August 6, 2026: A Gartner survey of 3,566 B2B and B2C customers found that 87% consider access to a human agent essential when a company uses generative AI for customer service. 

The result does not mean customers reject AI. Half said GenAI made service interactions easier, and 58% of GenAI users had allowed it to complete a task on their behalf. Among B2B respondents, task completion reached 74%.

The message for service leaders is therefore not “avoid AI.” It is “do not use AI as a gatekeeper.”

Key Takeaways

• Eighty-seven percent want an option to contact a human.
• Fifty percent say GenAI can make service interactions easier.
• B2B customers are especially willing to let AI complete tasks.
• Forced AI interactions can reduce future adoption.
• Escalation design is becoming a customer-experience differentiator.
• AI should collect context before transferring the customer.
• As AI service agents gain more authority, businesses will need clearer controls over what actions they can take autonomously. 

What Happened

Gartner conducted the survey in February and March 2026. It found that customers are becoming more comfortable with AI but resist situations where automated systems prevent or delay human assistance.

Gartner advises companies not to make GenAI a mandatory first step for every issue. The preferred model is for AI to identify intent, gather relevant information, and attempt a resolution when confidence is high, while preserving an obvious path to a person.

The research also found that customers were roughly three times more likely to use third-party GenAI services than company-provided chatbots during their most recent service interaction. 

Why This Matters

Many automation business cases are built primarily around deflection: fewer tickets reaching expensive human agents.

That metric can produce the wrong behavior. A chatbot may appear successful because it prevents escalation, even when customers abandon the conversation, repeat themselves, or move to another channel.

For B2B companies, the risk is greater because individual support cases may affect renewals, account expansion, or operational continuity. Preventing a high-value customer from reaching an expert can cost substantially more than the support interaction saves.

How Better AI Customer Support Works

GenAI should function as an intelligent intake and resolution layer.

For a routine billing question, it might authenticate the customer, identify the relevant invoice, explain the charge, and offer an approved action. For a disputed contract term or a production outage, it should quickly transfer the conversation with the customer’s history, intent, and completed troubleshooting steps attached.

That approach uses AI to reduce customer effort and agent workload without making automation the final authority.

How It Compares With Traditional Chatbots

Traditional chatbots were commonly designed around decision trees, frequently asked questions, and keyword matching. Their role was primarily to retrieve information.

Generative systems can interpret open-ended language and initiate actions such as managing subscriptions, submitting documents, or placing orders. Gartner’s data shows customers increasingly expect this type of task completion.

The higher capability also raises the consequences of mistakes. An AI that merely produces an irrelevant answer is frustrating; an AI that changes an account incorrectly creates operational and compliance risk.

What Businesses Should Do

Customer-service leaders should measure successful resolution, repeat contact, escalation quality, and customer effort, not simply chatbot containment.

Every automated journey should include a visible human option. Escalations should preserve the conversation context so customers are not forced to repeat information.

Companies should also define which actions AI may complete autonomously, which require customer confirmation, and which must always involve an employee. High-risk actions need runtime controls, audit logs, and clear reversal procedures.

Limitations

The survey combines B2B and B2C participants, whose service expectations and economic value can differ significantly.

It also measures stated preferences rather than the long-term effect of specific service designs. Companies still need controlled testing across their own customer segments, issue types and channels.

Pranali Shelar