AI Is Moving Fast. What Matters for CX Now?

A customer experience statistic can be accurate and still leave today's most important question unanswered.

A survey about whether people like chatbots tells us something useful. But does it tell us whether they would let an AI change an appointment, spend money, or negotiate with a company on their behalf?

Those decisions ask much more of the customer, and of the business serving them.

Fortunately, we have a fresh starting point. Avaya has just conducted new consumer research, fielded in September 2026 and published in Customer Experience on Their Terms. It explores what people expect as AI takes a more active role in service.

For October, the opportunity is straightforward: focus on useful action with clear authority, a workable recovery path, and evidence that the customer's need was met.

Why Does CX Research Need a More Frequent Refresh?

Because AI capabilities and the choices presented to customers are changing quickly, a survey's timeliness matters as much as its publication date.

Stanford's 2026 AI Index findings document rapid gains in AI capabilities alongside persistent weaknesses, including reasoning and coordination, and show that what is combined matters: what is technically possible can advance faster than our understanding of where it works reliably.

This is why I believe AI has shortened the practical shelf life of some CX research, especially findings about AI capability and willingness to delegate. Revisit a conclusion when the technology, task, or level of authority changes.

Freshness alone does not make a study useful. Look at its fieldwork dates, sample, question wording, and the behavior it actually measures. Keep older evidence that still applies, then test the assumptions most exposed to change.

Avaya's new research gives us a timely opportunity to do that. Several findings help turn a broad conversation about AI into specific product and service decisions.

How Much Freedom Should AI Have?

Start with the action the customer is authorizing. In Avaya's survey, 41% were comfortable pre-authorizing an AI assistant to initiate a refund when they qualified. Just 8% selected making a purchase under a spending limit they set.  

Both involve money. One returns it; the other commits it. That difference reminds us how much a broad label like “financial task” can hide.

PwC's 2025 Customer Experience Survey provides relevant context: consumers were more willing to use AI to track an order than to make a payment. The studies ask different questions, so they cannot establish a change over time. They do reinforce the value of examining specific tasks.

For a product roadmap, define the permission alongside the capability. What may AI do? Under what conditions? When should I ask again?

A refund assistant, for example, needs eligibility rules and an authorized model. The customer actually gets something, and the customer is usually asked for agreement.

Could an Undo Option Make AI More Useful?

For some customers, the ability to recover changes what they are willing to delegate. In Avaya's research, 37% said they would let AI act on its own only when the change could be easily reversed.

That puts recovery on the product agenda. Before enabling an automated account update, consider how the customer could correct an error, how the original information would be restored, and who would help if the change affected another system.

Some actions cannot be fully undone; require that, make that clear, and require appropriate approval or human review before they happen.

The opportunity is practical: a credible way to correct a mistake may help customers feel comfortable authorizing useful work. The survey captures that preference; the next step is to test how a particular recovery experience performs.

What Happens When the Customer Brings Their Own AI?

The business needs to understand who the assistant represents and what it can do. Customer experience may increasingly involve an AI acting for the customer, alongside AI working for the company.

Accenture's June 2026 Consumer Pulse research, covering 25,590 people in 16 countries, found that 74% would delegate routine tasks such as complaint resolution to an AI agent that followed their instructions. Only 9% were open to fully autonomous purchases.

Those are expressions of willingness, not a forecast of how many customers will reach out through an agent. They do give service leaders a reason to prepare.

Imagine an assistant requesting a subscription change for its owner. The company needs to verify that authority, understand its limits, provide accurate options, and obtain any required confirmation. A person should be able to step in without losing the relevant context.

The customer still owns the decision. An assistant changes how the request reaches the business and how the business should verify and fulfill it.

What Should CX Leaders Measure This October?

Measure the customer's completed outcome and the work it took to reach it. Avaya found that 47% of survey respondents said their most recent service issue was resolved during the first interaction. Another 31% said their issue was resolved, but it took more than one interaction.

That gives improvement a concrete destination: more requests completed successfully with fewer unnecessary steps.

McKinsey's August 2026 State of AI survey adds a useful business perspective. Eighty percent of respondents reported improved personal productivity from AI, while 37% reported a positive contribution to their organization's earnings. Measure individualists and taxes. Measure individual gains separately.

For CX, I would apply the same discipline. Alongside response speed, track:

• Completion: Did the customer receive the promised result, and how long did it take from the first request?
• Repeat work: Did they have to contact the company again or supply the same information twice?
• Exceptions and recovery: Did the system recognize when it needed help, and could an incorrect action be corrected?
• Cost per resolved need: What did the full journey cost, including AI usage, employee work, and follow-up?

These are proposed measures, rather than additional survey findings. They help determine whether a new capability improves service in practice.

How Does This Change the Technology Agenda?

CX technology should make it easier to adapt while preserving context and control. A new model or assistant becomes useful when it can work with the information, permissions, and business processes needed to complete a request.

That is a relevant role for Avaya Infinity. Avaya Infinity connects customer context, AI, enterprise systems, workflows, and human expertise. It supports multiple AI models and applies enterprise policies and permissions across the experience, giving organizations flexibility as their AI choices evolve.

For a customer, the benefit should be tangible: an authorized action that gets completed, or a knowledgeable person who can continue the work when help is needed.

This October, choose one frequent request, such as an eligible refund. Document the current outcome and the authority customers are comfortable granting. Test a bounded improvement, including its exception and recovery paths. Compare the results and listen to the people using it.

I recommend reviewing the assumptions monthly, and after a meaningful change to the model or workflow. Keep core questions consistent so you can distinguish a real shift from a differently worded survey.

AI's pace makes a more valuable opportunity for CX leaders to use that evidence to decide what to improve next, then check whether the improvement helped.

Read Avaya's Customer Experience on Their Terms for full findings on AI action, customer control, continuity, and human experience.

Avaya’s online panel survey was fielded in September 2026 among U.S. adults aged 18 and older. The resolution question had 577 respondents; the findings on pre-authorized actions and reversibility each had 558. The research offers directional insight into reported experiences and preferences. It does not evaluate Avaya Infinity or establish changes over time. The external studies provide context from different samples and questions.