Why CX needs a defined role in AI governance

Customer-facing AI can damage trust before the business recognizes a failure

A business can fail a customer before its AI governance process, the way it directs and oversees AI use, recognizes a problem. Consider an AI agent that incorrectly confirms a claim has been approved or offers a price the company cannot honor. An elderly customer, a disabled customer or someone in distress might be unable to leave an automated exchange or reach a person. In these illustrative situations, customers act on wrong information, cannot get help and lose confidence in a business they trusted.

For customer experience, or CX, leaders, the concern is how easily these consequences can escape the measures used to judge a deployment. Accuracy, compliance and efficiency each describe part of the system’s performance, while the customer experiences their combined effect. Years of trust can be damaged before anyone inside the business identifies a failure. So the business needs to assign responsibility for the customer consequences its functional measures leave out.

Organizational preparation shapes AI outcomes

That responsibility begins before an AI agent answers a customer, because its behavior depends on earlier organizational choices. These include the information available to the system, its objectives and the people involved in approving its use. Research from Gartner and McKinsey addresses different parts of this preparation: project viability, returns and reported adverse outcomes. Both firms sell research or advisory services that help organizations make these decisions, giving them a commercial interest in demand for guidance on AI adoption.

Gartner has projected that, through 2026, organizations “will abandon 60% of AI projects unsupported by AI-ready data”. The forecast applies specifically to projects whose data is insufficiently prepared for AI use. For leaders deciding whether to deploy, data readiness is therefore a condition for keeping the project viable. Even a capable model depends on information suitable for the task, so teams need to examine whether incomplete or biased data will undermine its use.

Preparing information also means examining how people will use it. McKinsey’s research links stronger AI returns to redesigning data and workflows, the sequences of work and decisions needed to complete tasks, before deployment. Its finding points to the process surrounding the model, including how information and responsibility move through the organization. Changing the model while preserving that process can leave the conditions that made the task difficult in place.

Assessing value also means accounting for adverse outcomes already being reported. The same McKinsey report found that “51% of organizations using AI reported at least one negative AI incident”. Almost one-third of respondents reported problems involving inaccurate outputs. The first figure concerns organizations reporting an incident, while the second concerns respondents reporting a particular kind of problem; Gartner’s forecast addresses a different population and outcome again.

These measures answer different questions, but each gives leaders a reason to examine the organization surrounding the model. Biased or incomplete data can affect answers, while limited AI literacy, the team’s understanding of how to use and assess AI, can prevent people from recognizing problems. An absent stakeholder may leave a customer consequence unexamined. Unclear objectives and missing guardrails, meaning rules or limits on system behavior, can also leave teams without a shared basis for deciding which results are acceptable.

These weaknesses can persist after launch unless someone remains responsible for checking performance. Consider a system that has operated for six months without anyone verifying whether it still performs as intended. Leaving that continuing responsibility unassigned is a failure, because a launch decision cannot establish fitness for use indefinitely. When an incorrect answer, adverse decision or inaccessible service reaches a customer, its causes may therefore span data preparation, team understanding and oversight.

Accuracy, compliance and efficiency can coexist with customer failure

Even when teams address these preparation problems, their chosen measures can leave customer consequences unexamined. A data team may judge a model “accurate” against its chosen measure, while CX leaders still need to determine whether its use is “good for the customer”. The first judgment concerns performance on a defined task. The second concerns what that performance means for the person receiving the result.

The difference becomes clear when a function meets its requirement but the customer still struggles. A disclosure can meet legal requirements while remaining difficult to understand, leaving the business with a question about comprehension after legal has completed its compliance assessment. Similarly, removing access to a person may reduce service costs while weakening trust when customers cannot resolve an issue through automation. Operations can identify the saving, while CX examines whether the resulting customer experience is acceptable.

These separate assessments leave a gap when the business treats them as a complete deployment decision. Data concentrates on accuracy, legal on compliance and operations on efficiency, with each team meeting its own objective. Yet customers experience the combined result. Unless someone examines that result across those responsibilities, functional success and customer failure can occur together.

Pressure to launch makes the gap harder to address because teams can be rewarded for different outcomes. Chad Sanderson, CEO of Gable.ai, described this conflict in Secoda’s report, State of Data Governance in 2025: “The incentive of one group is to slow down and be careful and do the right thing, and the incentive of the other group is to ship as quickly as possible.” Gable.ai and Secoda both have commercial interests in demand for data governance, so Sanderson’s observation comes from within that market. His point concerns the competing rewards for caution and delivery.

When delivery takes priority, teams may set aside customer-impact checks, and a function-by-function approval process can leave CX outside the discussion altogether. AI’s speed and scale make that misalignment more consequential because customers encounter its results through the deployed service. Decision-makers may see the customer’s experience only through complaints or incidents after the business has committed to its approach.

CX needs explicit responsibility for customer impact

Addressing this gap requires shared AI governance, with clear responsibilities across the organization. Data, legal, security, risk, product and operations each have responsibilities within it. Giving the whole function to CX would place a cross-functional problem in another single department. CX lacks the mandate for that broader responsibility, so its role needs a clear scope.

Within that shared arrangement, CX can take responsibility for representing customer impact. Customer experience crosses product, operations, marketing, pricing and service, giving CX a view of how decisions across those functions combine. CX can use that view to explain how a functional measure translates into a consequence for the customer. For customers facing the claim, pricing or access failures described earlier, accountability needs to exist before an incident report is written.

To make that accountability useful, the functions need a way to examine consequences together. A working group involving data, risk, legal, product, operations and CX can coordinate deployment decisions and assign questions that cross departmental boundaries. The organization may call this an AI Center of Excellence, meaning a group that coordinates AI expertise and decisions. Establishing this collaboration requires neither a platform purchase nor a new bureaucracy.

The group’s value depends on making responsibilities explicit. Each function retains its duties, while CX represents customer impact and helps colleagues examine it when making decisions. This gives CX a substantive role while keeping responsibility for data quality, compliance and other specialist work with the relevant teams. Participation matters when it brings an unresolved customer consequence to someone who can act on it.

CX can establish its contribution before formal governance begins

CX leaders can prepare for that role before receiving a formal invitation to a governance group. The first step is to build a working relationship with whoever leads AI, whether that is the chief technology officer, or CTO, or the chief AI officer, or CAIO. Understanding that leader’s assessment of AI risk helps CX connect customer concerns to decisions already being made. Building the relationship early gives CX a chance to influence those decisions before they become urgent.

With that relationship in place, the second step is to bring existing evidence of customer impact into the conversation. Patterns showing where AI helps customers deserve attention alongside patterns showing confusion or declining trust. Together, they give colleagues concrete consequences to consider when assessing benefits and risks. CX also gains a specific issue to raise and a basis for explaining why it matters.

To keep that evidence connected to decisions, the third step is to make customer impact a recurring part of AI discussions. Asking “how does this affect the customer?” should prompt evidence, expose an unresolved consequence or identify an unassigned responsibility. Repeating the question establishes an expectation, but an effective check also needs someone to examine the answer. When the discussion reveals a gap, the working group needs to assign responsibility for assessing that customer consequence before deployment.

Key highlights

• Assign responsibility for customer outcomes: Customer-facing AI can damage trust before internal measures register a failure. Deployment owners need an explicit check for incorrect promises, inaccessible service and customers unable to reach a person.
• Prepare the organization before deployment: AI sponsors need to assess data readiness, workflows, staff understanding and system limits before launch. Assign ongoing performance checks so responsibility continues as the system operates.
• Assess the combined customer experience: Accuracy, legal compliance and cost savings can coexist with customer failure. CX needs to examine how those results combine, including whether customers understand disclosures and can resolve problems through the service.
• Give CX a defined governance role: A cross-functional AI working group needs CX to represent customer impact, with data, legal, risk, product and operations retaining their specialist responsibilities. Assign unresolved customer-impact questions to someone who can act before deployment.
• Bring customer evidence into AI decisions early: CX leaders can build relationships with the CTO or AI leader before formal governance begins. Use evidence of customer benefits, confusion and declining trust to make customer impact a recurring part of deployment decisions.