How Much Would You Let AI Do for You?

What consumer research reveals about convenience, confidence, and personal AI agents

You hire someone to repair a leaking sink because you want the problem solved without spending your weekend learning plumbing. You agree on the work and the price. If that repair suddenly requires replacing the cabinets, you expect a conversation before the work continues.

Hiring an employee involves a similar judgment. You bring someone in for their capabilities, then define responsibilities, spending authority, and the decisions that need your approval. Delegation helps you accomplish more, even though the outcome is never completely certain.

Personal AI agents bring that familiar calculation into customer service. How much time could an assistant save me? What am I comfortable letting it decide? What happens if it gets something wrong?

I think the most useful way to understand this is to focus on the balance between convenience and confidence. Avaya's Customer Experience on Their Terms offers a timely view of where consumers draw those boundaries, and how enterprises can make delegation more appealing.

Why is AI Delegation a Question of Convenience and Confidence?

Saving time feels worthwhile when people can trust what happens while they are not watching.

Any service we purchase involves some uncertainty. We manage it through clear expectations, relevant expertise, limits on authority, and a way to fix things. Personal AI needs an equally understandable agreement about the work it may do.

The analogy has limits. An AI can misunderstand an instruction and carry that mistake through multiple connected actions. That makes the scope of its authority especially important.

OpenAI's introduction of ChatGPT dots on September 29, 2026, makes this question especially timely. Dots are designed to take on ongoing responsibilities, work between conversations, and return results for review. Users define what their dot can do independently, and it reaches out when a decision needs their judgment. That makes clear boundaries essential to an assistant that keeps working on a customer's behalf.

What are Consumers Ready to Hand Over to Personal AI?

The strongest interest is in familiar tasks that take time and effort.

Asked which company interactions they would be most likely to give a personal AI assistant, 48% selected checking the status of an order or request, and 39% selected waiting on hold or navigating customer-service systems. Respondents could choose up to two options.

But willingness is far from universal. In a separate scenario, a personal assistant contacts a company's AI using only authorized account information and pre-approved actions; about 40% were comfortable, and 42% were uncomfortable. These are preferences under specified conditions, rather than adoption rates.

PwC's 2025 Customer Experience Survey similarly found that customers were more willing to use AI to track orders than to make payments. Although the questions differ, both studies point to the importance of examining specific tasks.

Why Might a Customer Authorize a Refund but Hesitate Over a Purchase?

The consequences of the action matter as much as the technology performing it.

Elsewhere in Avaya's survey, respondents considered AI assistants more broadly. 41% were comfortable pre-authorizing an assistant to initiate a refund when they qualified; 8% selected making a purchase under a spending limit they set.  They could select all applicable actions, with no further approval at the moment of execution.

A purchase can create a new obligation; an eligible refund has a more tightly defined purpose. Those are different decisions about authority.

For product managers, the implication is practical: attach permission to the specific action. A customer who authorizes an assistant to investigate a billing problem has not necessarily authorized it to change their service plan.

Can Clearer Controls Make Delegation More Convenient?

Yes. Well-designed boundaries can let routine work move forward while reserving meaningful decisions for the customer.

When asked which safeguards would most increase confidence in AI acting on their behalf, 59% selected control over which actions it could take, and 52% selected requiring approval for important actions. Respondents could choose up to three safeguards.

Consider an illustrative appliance-repair appointment. A customer might authorize an assistant to move the visit to any available Tuesday morning slot, provided there is no additional charge. The assistant can handle the search and coordination. A new fee or a different day requires another decision.

The customer chooses where to spend their attention. Asking for approval on every status check adds work; silently accepting a new charge exceeds the agreed scope.

This approach aligns with NIST's AI Risk Management Framework, which ties risk management to context, risk tolerance, and defined human oversight. My product-design takeaway is to make those boundaries understandable at the point where customers delegate the work.

How Important is the Ability to Recover from a Mistake?

The recovery path is part of the customer's decision to delegate.

In Avaya's research, 37% said they would allow AI to act on its own only when the action could easily be reversed. For this group, recovery is a condition of independent action.

OpenAI’s practical guide to building agents recommends setting limits on repeated attempts and involving a person when those limits are reached or an action carries significant risk. For customer service, the lesson is to build clear stopping points, approval steps, and human handoffs into the experience from the start.

The human route matters too. After several unsuccessful attempts to resolve an issue with AI, 64% wanted automatic connection to a person with their conversation and information carried over. Help should continue from the work already done.

What Changes When the Customer's Representative is an AI Agent?

The enterprise needs to understand both the customer's goal and the authority granted to the representative.

My colleague Tvrtko Stosic explores the operational implications in The Customer Agent Stampede: many legitimate assistants repeatedly checking progress could create substantial service traffic during a disruption. He presents this as a planning scenario, rather than a measured forecast.

A clear response should tell the representative whether work has started, whether approval is needed, and when an update will be available. That helps it pursue the customer's goal without repeatedly restarting the process.

The customer's permission must remain meaningful as the request moves between systems, AI, and employees. The entire journey must honor that same agreement.

How Can Avaya Infinity Make Delegation More Useful?

Avaya Infinity brings together the context, systems, AI, and people needed to turn an authorized request into completed work.

How Can Enterprises Give Customers More Time Back?

Avaya Infinity connects customer context with workflows across front- and back-office systems. That helps organizations carry a request through the work needed to resolve it, reducing the coordination left to the customer.

How Can Businesses Choose the AI that Fits the Task?

Avaya Infinity's open, model-agnostic approach and support for Model Context Protocol connect AI with enterprise tools and context. Organizations can evolve their AI choices while building on the systems they already use.

How Can Enterprises Keep Control as AI Does More?

Avaya Infinity supports enterprise-defined policies, permissions, and auditability, alongside flexible cloud, hybrid, and on-premises deployment options. This gives enterprises more control over how they introduce AI and apply it to their operating requirements.

What Happens When Human Judgment is Needed?

Avaya Infinity brings AI and human expertise together with shared context, helping an employee continue the customer's journey with an understanding of what has happened.

In the appliance-repair example, the service system reflects appointment changes in the customer's instructions, and the customer receives confirmation. If an exception needs discussion, a prepared person can help. That is a convenience worth delegating.

Where Should CX Leaders Begin?

Start with a frequent task customers would welcome handing over, then make the permission easy to understand.

Identify what the assistant can complete, what needs approval, and how the customer can get help. Judge the experience by whether the need is resolved and how much work the customer still has to do.

The opportunity is to give people more of their day back, with confidence that the choices that matter remain theirs.

Read Customer Experience on Their Terms to explore Tvrtko's perspective on personal agents at scale and learn how Avaya Infinity supports connected customer journeys.

About the research

Avaya's Customer Experience on Their Terms is part of the Avaya Connected Consumer Research Series. The online panel survey was fielded in September 2026 among U.S. adults aged 18 and older. The Avaya statistics here come from questions 12, 15, 19, 21, 23, and 24, each with 558 respondents. The findings describe preferences and expectations, not actual adoption or measured product performance.