Most conversations about AI and customer experience still start with the model. Which model should we use? How much service can we automate? Should we build agents? How much will it save? How quickly can we deploy it?
Those questions matter, but they skip over something more fundamental.
As AI moves from answering questions to taking actions, companies need to know who the AI is acting for, what it is allowed to see, what it is allowed to change, and who is accountable when something goes wrong.
That puts identity much closer to the center of the AI architecture than most CX teams realize.
For years, identity was treated mainly as a security issue. Employees logged into applications. Customers authenticated themselves. Administrators controlled access to systems and data. The experience implications were usually limited to passwords, multifactor authentication, login friction, and fraud prevention.
AI changes the role identity plays because software is starting to act on behalf of people.
A service agent can now do far more than answer a question. It can change a reservation, issue a credit, update an account, move an appointment, cancel a subscription, or initiate a payment.
Once software starts doing those things, identity becomes part of the decision itself.
Consider a customer whose flight has been canceled. An AI service agent identifies the customer, checks loyalty status, finds alternative flights, rebooks the trip, issues a hotel credit, updates the record, and sends confirmation.
That could be an excellent experience.
It could also create a serious problem if the AI has the wrong customer, accesses information it should not see, issues more compensation than it is authorized to provide, or changes something outside the customer’s request.
The system needs to know who the customer is. It needs to know which information belongs to that customer. It needs to know which data the AI is permitted to use for that interaction. It also needs to know what the AI itself is authorized to do.
That is why companies such as Microsoft, Okta, SailPoint, CyberArk, and others are spending more time on AI and non-human identity. They are responding to a real architectural problem.
AI creates actors that are not employees.
Those actors still need permissions, limits, ownership, and an audit history.
For CX leaders, that creates three identity problems that increasingly overlap. The first is customer identity. Can the company reliably determine who the customer is, which accounts and interactions belong to that person, and which preferences and permissions apply?
The second is employee identity. What can the employee serving that customer see and do?
The third is AI identity. Which agent is acting, who owns it, whose authority it is using, what data it can access, and what actions it can take?
In a growing number of customer interactions, all three will be involved at the same time.
That has major implications for personalization.
Companies have spent decades trying to create a more complete view of the customer. One system knows purchase history. Another knows service interactions. Marketing knows campaign behavior. The mobile app knows recent browsing. The loyalty platform knows status. Billing has a different set of information again.
AI makes that fragmentation harder to live with because the quality of the response depends on connecting the right information to the right person.
Imagine asking your bank’s AI whether you can afford to buy a house next year.
A useful answer could require checking deposits, expenses, savings, debt, investments, and previous conversations about financial goals. That is a significant amount of personal and financial context.
The bank may know all of those things about you. That does not mean every AI system inside the bank should be able to use all of them.
A service agent may need your checking-account balance but have no reason to see information used by a fraud team. A marketing agent may need to know that you own a mortgage but should not necessarily receive detailed income information. A financial planning agent may require broader access.
This is where identity starts shaping the actual quality of personalization.
Good personalization will depend less on giving AI everything the company knows and more on giving it the right information for the specific interaction.
That is a better experience for the customer and a safer operating model for the company.
Customer service makes the issue even more immediate.
The first wave of generative AI in service was relatively contained. AI summarized calls, recommended responses, found knowledge articles, drafted emails, and helped employees work faster. The next wave is different because AI is beginning to complete the work itself.
A customer asks to cancel a subscription, replace a card, move an appointment, waive a fee, change an address, or refund a purchase. The AI can increasingly carry out the request without a person handling each step.
Every one of those actions requires authority.
A service AI should not have unrestricted access simply because the customer service organization has broad access. Its permissions should reflect the customer, the action, the amount of money involved, the risk, and the context.
A company might allow an AI agent to issue a $50 service credit automatically. A customer with a long relationship and a documented failure might qualify for $250. A $2,000 reimbursement may require an employee to approve it.
The customer does not need to see all of that complexity.
They just need the problem resolved correctly.
Identity and authorization become part of the machinery that makes that possible.
This is also where CX and security goals become more closely aligned than they often appear.
Security teams want to prevent unauthorized actions.
CX teams want legitimate customer requests completed with as little friction as possible.
A well-designed identity model can improve both.
Authentication is a good example.
Anyone who has spent time looking at service data knows how much customer effort is created by identity verification. Passwords are forgotten. Security questions are useless. One-time codes fail. Phone numbers change. Customers cannot act on behalf of parents, spouses, or family members without getting stuck in rigid processes.
AI makes some of those risks worse because synthetic voices, forged documents, and deepfakes make traditional identity signals less reliable.
But AI can also make authentication more intelligent.
Companies can assess risk throughout an interaction instead of forcing every customer through the same process. A customer asking for store hours should not need the same level of authentication as a customer moving $50,000. Changing an address may justify stronger verification than checking the status of an order.
That creates room for a more graduated model of trust.
Customers face less unnecessary friction, while higher-risk actions receive stronger controls.
That is a customer experience improvement, not merely a security improvement.
Identity also matters in a less obvious area: analytics.
Imagine an executive asking an internal AI tool why customer retention declined last quarter.
The system searches across multiple data sources and produces an answer in seconds. The explanation sounds clear and confident.
But Finance defines retention one way. Customer Success defines it another. Product has a different definition of an active customer. A calculation changed six months ago. One dashboard was never updated. Ownership shifted after a reorganization.
The problem is not whether the AI can calculate the answer. The problem is whether the company knows which definition is authoritative.
This is one reason I am cautious about companies rapidly building their own analytics applications with tools such as Claude Code or Codex and assuming the hard part is over.
The code is becoming easier to create. The operating discipline behind trustworthy analytics is not.
Metrics still need owners. Definitions still need change control. Data needs lineage. Teams need to know which source is authoritative when two systems disagree.
AI can make weak analytics more dangerous because it produces answers quickly and with confidence. The user may never see the disagreement or ambiguity underneath the response.
Identity does not solve every data-governance problem, but ownership and authority need to be tied to specific people, teams, and systems. Otherwise the company has no reliable way to determine whose definition should prevail.
The same principle applies to AI agents themselves.
Most enterprises have designed identity systems around people. Employees join the company, change roles, move departments, and eventually leave.
AI agents do not behave like that. A large company may eventually operate many more AI agents than employees. Some agents will exist for years. Others may be created for a single task and disappear 20 minutes later. One agent may create or invoke several others.
That creates a very different identity-management problem.
A 50,000-person company could eventually have hundreds of thousands of active or temporary AI identities moving through its systems.
Every consequential agent should have a clear owner, a defined purpose, approved data sources, specific permissions, limits on what it can change or spend, escalation rules, and some form of expiration or review.
If nobody inside the organization can say who owns an AI agent and what it is allowed to do, that agent should not be changing customer records.
This will become particularly important as regulation catches up.
The European Union’s AI Act already includes transparency requirements around certain customer-facing AI interactions. That pushes companies toward clearer disclosure when people are interacting with AI.
The experience implication matters.
For years, some companies tried to make automated service look and sound as human as possible. That approach becomes harder to defend as AI systems become more capable and begin taking consequential actions.
Trust is more likely to come from clarity.
Customers should know when they are interacting with AI. They should have some understanding of what the AI can do. They should have access to a person when the situation requires one.
At the same time, the company needs a record of which AI agent acted, what authority it used, which systems it accessed, and which policies governed the decision.
That combination of transparency and traceability will matter more as AI becomes more autonomous.
There is another change coming that may have an even larger impact on CX. Customers will start bringing their own AI agents into interactions.
Imagine telling your personal AI:
“My cable bill increased again. Find out why and move me to the least expensive plan with at least one gigabit of service. Do not accept a contract longer than 12 months.” Your AI contacts the provider.
The provider needs to know that the agent legitimately represents you.
Your agent needs to prove which actions you authorized.
The provider’s AI needs authority to make an offer.
Your agent compares it against alternatives and completes the change.
You may never open the provider’s app or speak to a service employee.
The same pattern could apply to booking travel, disputing an insurance claim, canceling subscriptions, refinancing a loan, returning a product, or managing investments.
That changes the identity problem again.
A company now has to establish trust with software acting on behalf of a customer.
The airline needs to know that the agent really represents the traveler. The bank needs to know whether the agent can transfer money but cannot close the account. The insurer needs to distinguish a legitimate customer agent from automated fraud. The customer needs a simple way to revoke that authority when they no longer want the agent acting for them.
These are likely to become mainstream customer-experience issues over the next decade.
They will also change digital design.
Companies have spent the last 20 years designing websites and mobile apps for human attention. Buttons, menus, content, personalization, search, and visual design all assume a person is interacting with the company.
A customer’s AI agent may care about very different things.
It may evaluate price, product specifications, service history, contract terms, cancellation policies, delivery performance, and whether it can complete a task directly through an API or other machine-readable interface.
That means companies may eventually need to serve two types of customer interaction.
One is designed for people.
The other is designed for software acting for people.
Identity will connect the two. The customer needs a trustworthy way to delegate authority. The company needs a reliable way to recognize that delegation. Both sides need a record of what was authorized and what occurred.
For CX leaders, this means identity cannot remain a topic that gets handed entirely to the security organization.
Security should own major parts of the technical architecture. Legal, technology, data, product, and operations all have important roles too.
But many of the decisions are also experience decisions.
How much authentication should a customer face for a given action?
What information should an AI service agent be able to see?
How much money can it refund automatically?
When does a human need to approve a decision?
How should customers delegate authority to a family member, employee, or AI agent?
How should the company explain when AI is acting on the customer’s behalf?
Those choices affect customer effort, resolution time, fraud, trust, cost, retention, and revenue.
CX leaders should be involved because the wrong identity model can create a technically secure experience that customers hate, or a frictionless experience that creates unacceptable risk.
There is also a larger strategic point here.
AI models will keep improving. They will also become easier to switch between.
Companies will use different models for different tasks based on accuracy, speed, privacy, cost, and performance. Over time, the model itself may become less important than the context and authority surrounding it.
An AI system needs to know who the customer is.
It needs to know which information is relevant.
It needs permission to use that information.
It needs authority to take an action.
And the organization needs to know who is accountable afterward. That is why I think identity becomes more valuable as AI becomes more capable.
For CX leaders, the effect is direct.
Identity will determine how personalized an experience can become without feeling invasive. It will determine how much service can be automated without creating unacceptable risk. It will influence how much authentication friction customers face. It will affect whether AI-generated analytics can be trusted. And eventually, it will determine whether a company can safely interact with AI agents acting on behalf of its customers.
For years, companies have talked about building a 360-degree view of the customer.
AI makes the next question much more important: once the company has that information, who is allowed to use it, for what purpose, with what authority, and on whose behalf?
That is where identity moves from security infrastructure into the customer experience itself.
