When Customers Send AI Assistants, Who Finishes the Job?

Imagine asking your AI assistant why an order hasn't arrived.

It contacts the retailer and discovers that the item is delayed. A substitute is available, but it costs more. Canceling is another option. So is waiting.

Your assistant has found the answer. Your problem still needs a decision.

That small moment reveals a much larger customer experience challenge: a simple request can reach the edge of an assistant's authority before it reaches a useful conclusion.

In my recent Avaya Insights guide, When AI agents act on behalf of customers: How customer service must prepare, I explored how businesses can use AI assistants to serve customers. Avaya's new Customer Experience on Their Terms research adds a valuable perspective: what people would actually want those assistants to do.

What Do Customers Want to Delegate to Personal AI?

Routine follow-up is a promising place to start. When asked which tasks they would be most likely to delegate to a personal AI assistant, 48% of respondents selected checking the status of an order or request. Another popular choice, at 39%, was waiting on hold or navigating customer service systems. Just 9% selected canceling or changing a service.

Respondents could choose up to two tasks so that these groups can overlap. The results describe stated preferences, not current usage.

My reading is that many customers see value in delegating the administrative effort around a decision. That gives businesses a practical starting point: make it easier for an authorized assistant to obtain a useful answer, while preserving the customer's choices about what happens next.

There is also a reason to keep choice visible. In a separate question, 40% were comfortable with their personal AI contacting a company's AI, while 42% were uncomfortable. The scenario explicitly limited the assistant to authorized account information and actions approved in advance.

Interest in particular tasks and comfort with direct AI-to-AI service are different questions. Both deserve attention when designing the experience.

Where Does the Customer Journey Begin when AI Gets Involved?

It may begin in a tool the customer chose before the business ever hears from them. That changes where service teams need to look for customer effort.

A Gartner survey published in July 2026 of 3,566 B2B and B2C customers found that customers were about three times more likely to use third-party generative AI than company-provided chatbots to resolve service issues.

That finding does not mean those tools contacted companies independently or changed accounts. It does suggest that a business's own chatbot gives it only a partial view of how customers seek help.

For CX leaders, the useful question is what an authorized representative would need to continue the journey: accurate information, a workable way to request service, and a clear next step when the answer requires a customer decision.

How Can a Business Know What an AI Assistant May Do?

It needs to establish both who the assistant represents and the scope of the customer's permission. A recognizable agent still needs authority for the particular action it requests.

That distinction is becoming part of the infrastructure for agent commerce. Visa's Trusted Agent Protocol, introduced in October 2025, helps merchants recognize trusted agents and their commercial intent. Recognition helps a business assess who is approaching it. The business must still determine which actions are permitted.

Google's April 2026 update to Agent Payments Protocol describes support for agent payments under instructions authorized by the user in advance. The application is payments, but the design principle is relevant to service: the permitted action should be explicit.

Return to the delayed order. Permission to retrieve its status should let the assistant complete that inquiry. Replacing it at a higher price may require new approval. A well-designed experience makes that boundary clear and lets the customer decide without restarting the conversation.

What Counts as Resolution When an AI Assistant is Involved?

Resolution means the authorized request is complete and the customer understands the result. A fluent exchange between two systems is only one part of that work.

If the request is simply to check delivery status, an accurate update may complete it. If the customer subsequently approves a replacement, completion requires more: the order must change, any payment adjustment must be handled, and the promised delivery information must be confirmed.

Businesses should also plan for partial completion. If the order changes but the payment adjustment fails, who takes responsibility for the unfinished work? What will the customer or their assistant be told? How will an employee pick up the request?

These questions belong in the original service design. Otherwise, a system may appear successful while leaving the customer to discover and repair what went wrong.

I would evaluate a delegated journey by whether it produced the agreed result, how much follow-up the customer had to do, and whether exceptions reached someone able to resolve them.

How Should Enterprises Prepare Their Service Systems?

Connect the incoming request to the context, business rules, systems, and people needed to fulfill it. An additional way to contact the company becomes valuable when it can carry useful work through the organization.

The wider foundations are still developing. NIST's AI Agent Standards Initiative, launched in 2026, includes work on interoperability, agent identity, and authentication. This is ongoing standards work. Enterprises should expect interfaces and practices to evolve as they learn which approaches work.

That makes flexibility useful, alongside clear responsibility for each step of a service request.

This is where Avaya Infinity fits into the discussion. Infinity connects customer context, AI, enterprise systems, workflows, and human expertise under enterprise policies and permissions. Those capabilities can help a business carry an authorized request through its own processes and preserve context when a person needs to step in.

The organization still needs to establish the external assistant's authority for the requested action. Connecting systems and establishing customer permission are related parts of the design, each with its own work to do.

Where Should CX Leaders Start?

Choose one frequent task customers would like to delegate, and test the entire journey. Order status is a strong candidate because it attracted the most interest among the listed tasks.

Work through an ordinary request, a delayed order, and a case where an account change is needed. Check that the assistant receives the information it is allowed to see, that a new decision reaches the customer, and that an employee can continue with the relevant context.

Ask customers whether the experience saved them work and left them confident about the outcome. Keep direct human service available for those who prefer it or need it.

There is a relationship opportunity here, too. When an assistant handles routine follow-up, a person can focus on the choice that matters most to them. The business earns trust by making that choice easier and following through.

In our delayed-order example, the customer should be able to say, “I know my options, I chose what happens next, and it has been taken care of.” That is a useful ambition for the next chapter of customer service.

Explore the consumer findings in Customer Experience on Their Terms, then use my enterprise guide to AI agents acting for customers to examine what those preferences mean for your service design.

About the research

Avaya fielded its online panel survey in September 2026 among U.S. adults aged 18 and older. The personal AI findings cited here each had 558 respondents. They provide directional insight into stated preferences and hypothetical comfort, rather than observed adoption or a forecast.