Twelve months into a regional bank's AI lending program, the chief risk officer made an observation that stopped the quarterly governance review cold: every escalation in the past year had come from the same three people. One underwriter in commercial credit, one compliance officer in fair lending, and one BSA analyst who had made it a personal mission to document every AI discrepancy she found. The rest of the institution was either quietly routing around the AI tools, passively accepting AI outputs without the verification the workflow required, or waiting for someone else to raise a concern. The CRO's conclusion was not that the program had failed. It was that it had a champion problem: three unofficial advocates carrying a workload that needed to be distributed across every function that touched an AI-assisted credit decision. (The scenario above is a composite drawn from observed governance patterns; the institution and individuals are not real.)

Building AI champions across lending and risk is the organizational design problem that sits underneath every other change management challenge. Governance policies, training programs, and monitoring dashboards all depend on the same thing: people in every relevant function who understand what the AI is doing, care whether it is working as designed, and have both the knowledge and the organizational standing to say something when it is not. That distributed professional responsibility is what makes an AI program robust rather than fragile. The bank that builds it survives its first regulatory examination. The bank that does not survives on luck.

What an AI Champion Actually Does

The term "champion" is used loosely in technology change management, often meaning something like "enthusiastic early adopter who helps sell the new tool." In the context of AI lending governance, the definition needs to be more precise, because the job is not sales. It is stewardship.

An AI champion in a lending and risk context does five specific things. First, they serve as the local subject matter expert on the AI tools used in their function: they understand the inputs, the outputs, the known limitations, and the verification requirements well enough to answer their colleagues' questions without escalating every inquiry to model risk or the vendor. Second, they actively monitor the quality of AI outputs in their area and flag anomalies through the escalation path rather than working around them. Third, they translate between their function's domain language and the governance language used by model risk, compliance, and fair-lending staff, so that a credit-team concern about an AI output pattern gets framed in a way that the governance team can investigate rather than dismiss. Fourth, they provide feedback to the model-risk committee on workflow friction, training gaps, and capability requests that would improve the program. Fifth, they model the verification behaviors that the training requires, so that their colleagues see what "doing it right" looks like in daily practice.

The champion's role is not to be the AI enthusiast who overrides every skeptical colleague. It is to be the bridge between the people who operate the AI and the people who govern it. The champion who understands credit risk and can speak to model risk is more valuable than the champion who only speaks one language.

An AI champion in lending is not a change agent selling a tool; they are a governance steward who understands what the model does, flags what it gets wrong, and holds the accountability line when the workflow is under pressure.

The Champion Coalition by Function

A well-designed champion coalition spans every function that materially interacts with AI tools in the lending and risk workflow. The coalition is not a committee; it is a distributed network of individuals with domain expertise, governance awareness, and a shared understanding of the AI assists, you decide contract. The functions that need at least one champion are the following.

Credit and underwriting

The credit champion is the most operationally critical node in the network. This person works in the underwriting team, uses the AI pre-scoring tool daily, and has enough technical curiosity to understand what drives the model's outputs rather than just accepting or rejecting them. They are typically a senior underwriter with a track record of good independent credit judgment, because the credibility to challenge an AI output depends on demonstrated competency in the underlying skill. The credit champion does not need to be a data scientist; they need to be an excellent underwriter who has learned enough about the model to be a useful interface between the front line and model risk.

The credit champion's specific monitoring responsibility is file-level AI output quality: income extraction accuracy, pre-score consistency with the file data, and adverse-action reason code grounding. They review a sample of AI-assisted files every month against the bank's verification checklist and report anomalies to the model-risk committee. When they find a pattern (the model is consistently extracting self-employment income from the wrong tax return line, for example), they document it with file-level examples and escalate through the model-risk channel, not through informal complaint. The documentation discipline is what converts a file-level observation into an institutional improvement.

Compliance and fair lending

The compliance champion needs to understand the AI tools well enough to examine them from a regulatory perspective. In practice, this means understanding what OCC Bulletin 2026-13 (the April 2026 interagency model-risk guidance that superseded OCC 2011-12 and pulled AI and generative AI under model-risk, fair-lending, third-party, and board-governance expectations) requires of the model-risk program, how to read a disparate-impact analysis and connect it to the bank's obligations under ECOA (the Equal Credit Opportunity Act, which prohibits discrimination in any aspect of a credit transaction) and Regulation B (Reg B, the implementing regulation governing adverse action notices and nondiscrimination), and what the examiner will look for when they review the model-risk file.

The fair-lending champion has an even more specific role. They are responsible for running or supervising the periodic disparate-impact analyses on AI-assisted credit decisions, interpreting the results, documenting any search for a less-discriminatory alternative when a disparity is found, and reporting the findings to the governance committee. This person needs to understand both fair-lending law and enough about the AI model's design to connect a statistical disparity to its potential source. They are not the statistician who runs the regression; they are the professional who interprets the result in regulatory terms and drives the remediation conversation.

Compliance and fair-lending champions also serve a critical function in the escalation chain. When a front-line employee raises a concern about an AI output pattern that might have fair-lending implications, it is the fair-lending champion who decides whether the concern warrants a formal investigation, a model adjustment, or a finding in the monitoring log. The credibility of the escalation path depends on this champion being both accessible (easy to reach, responsive, non-punitive) and competent (able to assess the concern and act on it appropriately).

BSA/AML and fraud

The BSA/AML champion operates in a different AI landscape from the credit champion: the alert volumes are higher, the false-positive rate is dramatically higher (roughly 90 to 95 percent of transaction monitoring alerts in the industry are false positives), and the governance obligations are different because the output is not a credit decision but a decision about whether to file a Suspicious Activity Report (SAR), a regulatory document submitted to FinCEN, the Financial Crimes Enforcement Network.

The BSA/AML champion's monitoring responsibility is alert-triage quality: are the AI's priority rankings producing better analyst outcomes than the prior system? Are the false-positive rates improving? More importantly, is the system catching the true positives that human review needs to investigate? The SAR filing obligation belongs to the human analyst, not the AI, and the BSA/AML champion is responsible for confirming that the AI triage tool is not obscuring genuine suspicious activity in a pile of correctly-deprioritized noise.

The BSA/AML champion also needs to understand the intersection between BSA/AML monitoring and the credit origination workflow. When AI-assisted BSA/AML transaction monitoring flags a prospective borrower, the flag should inform the origination decision through a defined process, not an ad hoc conversation between departments. Building that process, and making sure it is applied consistently, is partly a BSA/AML champion responsibility and partly a cross-functional coordination challenge that the champion coalition is designed to solve.

Loan origination and officer management

Loan officer champions are less about governance monitoring and more about pipeline quality and workflow adoption. They serve as the practical guides who help their colleagues navigate the AI-assisted workflow effectively, identify when AI flags require escalation versus self-service resolution, and communicate upstream when a workflow design is producing friction that slows origination without improving quality.

The officer champion's most important function is feedback collection. Loan officers interact with the AI-assisted LOS (loan origination system) at the point of initial file submission and at every subsequent stage where the AI-generated output appears. They see the user experience of the tool, and they observe patterns in how borrowers respond to AI-assisted communications. Their feedback on what is working and what is not is data that model risk and the technology team need but often do not systematically collect. The officer champion structures that feedback and brings it to the governance committee in a form that can be acted on.

Operations and technology

The operations and technology champion ensures that the AI tools are configured as the governance program requires: that model versions are logged, that audit trails are generated for every AI-assisted decision, that the LOS enforces the verification checkpoints before files advance, and that any model update is deployed through the change management process rather than as an unreviewed configuration change. This champion is less focused on credit quality and more focused on the institutional controls that protect the governance record.

The technology champion also owns the integration between the AI tools and the bank's model-risk infrastructure: how AI-generated outputs are stored, how long they are retained for examination purposes, and how the model inventory is updated when new tools are deployed or existing tools are modified. Under OCC Bulletin 2026-13, these are not IT decisions; they are governance decisions with IT implications. The technology champion is the bridge between the governance expectation and the technical implementation.

Recruiting and Developing Champions

The champion network does not build itself. The chief lending officer or chief risk officer who wants to build it needs to make a series of deliberate decisions about who to recruit, how to develop them, and how to give them the organizational standing to do the work.

Recruit for curiosity and credibility, not enthusiasm. The most effective champions in banking AI deployments are typically not the staff who were most excited about the technology during the deployment announcement. They are the staff who asked the hardest questions about how the model works and what happens when it gets something wrong. Curiosity about the model's mechanism and credibility in the underlying discipline (credit judgment, compliance expertise, BSA/AML investigation) are the two characteristics that make a champion useful to the governance program. Enthusiasm without one or both of these is a communication resource, not a governance one.

Give champions dedicated time. The champion role cannot be an add-on to a full production workload. If the credit champion is expected to review a monthly sample of AI-assisted files, attend governance committee meetings, participate in model-risk escalations, and respond to colleagues' questions, all while carrying their normal underwriting queue, they will either do the champion work badly or burn out within six months. Institutions that build sustainable champion networks allocate dedicated time to the champion role, typically 15 to 20 percent of the champion's work week for a front-line champion and more for a compliance or BSA/AML champion whose monitoring responsibilities are more intensive.

Develop champions through structured learning, not just experience. The AI for Banking and Lending certification program is explicitly designed to develop the competency that champions need. A credit champion working toward Level 3 (AI-Integrated Practitioner) certification is building the workflow design and governance skills that their champion role requires. A compliance champion working through Level 4 (AI Lending Strategist) is building the OCC Bulletin 2026-13 governance architecture knowledge that makes them useful to the model-risk committee. The certification path is not just individual professional development; it is institutional champion pipeline development when the chief lending officer or chief risk officer sponsors it as such.

Calibrate the champion network to the bank's AI footprint. A community bank running a single vendor-provided AI pre-scoring tool needs a different champion network than a regional bank running multiple AI tools across origination, underwriting, BSA/AML, and servicing. The right number of champions per function depends on the volume of AI-assisted decisions in that function, the complexity of the model, and the regulatory risk associated with errors in that function. Credit and compliance champions are non-negotiable for any institution using AI in credit decisioning. BSA/AML champions are essential for any institution using AI in transaction monitoring. Technology champions become more important as the AI footprint grows and the governance infrastructure becomes more complex to maintain.

The Governance Committee as Coalition Infrastructure

The champion coalition is a network of individuals; it needs an institutional home. That home is the AI lending governance committee, which brings together the champion representatives from each function on a regular basis to review model performance, share escalation findings, discuss training needs, and make governance decisions about model changes, vendor issues, and policy updates.

The governance committee's composition should include representation from credit and underwriting, compliance and fair lending, BSA/AML, operations, technology, and legal. The chair is typically the chief risk officer or the chief compliance officer, because the committee's output includes fair-lending determinations and model-risk decisions that need to be owned at that level. Model-risk management, if the institution has a dedicated model-risk function, is a standing participant and may serve as the committee's secretariat.

The committee's standing agenda should cover four areas. First, model performance reporting: how are each of the AI tools performing against their baseline metrics, and are there any anomalies that warrant investigation? Second, escalation log review: what concerns were raised by champions and front-line staff since the last meeting, what investigations were conducted, and what were the outcomes? Third, fair-lending testing results: what did the most recent disparate-impact analysis show, and if a disparity was found, what is the remediation plan and timeline? Fourth, model change and update review: are any model updates pending, and have they been through the change management process including revalidation and disparate-impact testing?

The committee also serves as the institutional memory for AI governance decisions. When the bank decides to approve a vendor's proposed model update, that decision should be documented in the committee minutes with the supporting rationale, the validation findings, and the fair-lending testing results. When the bank decides to require a vendor to provide additional documentation before an update goes live, that decision should be documented with the same rigor. This documentation discipline is what makes the model-risk record under OCC Bulletin 2026-13 more than a static file; it is a living governance record that an examiner can reconstruct to understand how the institution governed its AI program at every material decision point.

Sustaining the Coalition Under Pressure

The champion coalition is most valuable and most fragile at the same moments: when the AI program is under pressure from a regulatory inquiry, a model performance problem, or a business-side push to reduce the friction the governance controls create. The institutions that sustain their coalitions through these moments have built two specific structural protections.

First, they have made champion autonomy explicit in the governance policy. A champion who raises a fair-lending concern to the governance committee should not be in a position where their business-line manager can instruct them to withdraw the concern before the committee acts on it. The governance policy should specify that escalations to the model-risk committee through the champion network are not subject to business-line approval or withdrawal. This protection is both a governance integrity measure and an employment-law risk management measure: a champion who is retaliated against for raising a compliance concern is a whistleblower liability.

Second, they have connected the champion network to the institution's exam-readiness process. When an examination is announced, the champions are not surprised by the examiner's questions because they have been building the evidence base throughout the examination cycle. The credit champion has a monthly file-review log. The fair-lending champion has quarterly disparate-impact testing records. The BSA/AML champion has alert-triage quality metrics. The technology champion has a complete model inventory with current governance status for every AI tool. None of these are assembled for the exam; they exist because the champion role requires them to exist continuously. The examination becomes a demonstration of a functioning governance program rather than a reconstruction of one.

The third-party risk dimension also bears mention. OCC Bulletin 2026-13 requires that the regulated institution govern vendor-provided AI tools as rigorously as internally built ones. The technology and compliance champions are the primary owners of vendor oversight: contract compliance, performance monitoring, incident escalation, and the due-diligence review that precedes any material vendor contract renewal. When a vendor releases a model update without the notification required by the contract, the technology champion catches it. When a vendor's model-validation documentation does not meet the bank's minimum standards, the compliance champion escalates before the model goes live, not after the examination finds the gap. The champion network is what makes third-party model risk management an operational practice rather than a procurement exercise.

The Coalition as Long-Term Institutional Capability

The champion coalition built for the first AI lending deployment becomes the foundation for every subsequent deployment. The credit champion who developed expertise in the pre-scoring model is the right person to evaluate the next generation of income-verification AI. The fair-lending champion who ran the first disparate-impact testing program can extend it to cover a new automated valuation model. The BSA/AML champion who built the alert-triage monitoring framework can apply the same methodology to a new fraud-detection tool.

The 38 percent of mortgage lenders using AI or machine learning in 2024, up from 15 percent in 2023, are not slowing down their AI adoption. The institutions that build champion coalitions now are building institutional capacity that compounds as the AI footprint grows. The institutions that rely on a few unofficial advocates and hope the governance gaps do not surface in an examination are building exposure that also compounds as the footprint grows, because each new AI tool without a competent human champion is an additional governance gap.

The chief lending officer who builds a sustainable champion network is doing something more important than managing a technology deployment. They are building the human infrastructure that allows the bank to operate at the intersection of AI's efficiency potential and the regulatory framework that governs it. That infrastructure does not show up in the AI vendor's performance metrics or the board's efficiency dashboard. But it shows up when the OCC examiner walks in, asks to speak with the staff who operate the models, and finds a group of professionals who understand exactly what the models do, exactly what the governance requirements are, and exactly how to demonstrate that the institution is meeting them.

Key Takeaways

• An AI champion in lending governance is not an enthusiast selling a tool; they are a steward who understands what the model does, monitors its outputs, escalates anomalies through the formal governance channel, and models the verification behaviors the workflow requires.
• The coalition needs at least one champion in each materially affected function: credit and underwriting, compliance and fair lending, BSA/AML, loan origination and officer management, and operations and technology.
• Recruit for curiosity about how the model works and credibility in the underlying domain, not for enthusiasm about AI; the most effective champions are typically those who asked the hardest questions during the deployment announcement.
• Give champions dedicated time (typically 15 to 20 percent of a front-line champion's week), develop them through the structured certification curriculum (Level 3 for credit champions, Level 4 for compliance champions), and calibrate the network size to the bank's AI footprint.
• The governance committee is the institutional home of the coalition, bringing champion representatives together regularly to review model performance, escalation logs, fair-lending testing results, and model change proposals; its minutes are the evidence base for OCC Bulletin 2026-13 examination readiness.
• Protect champion autonomy in the governance policy so that escalations are not subject to business-line withdrawal, and connect the champion network to the exam-readiness process so that the governance documentation exists continuously rather than being assembled on examination announcement.
• The champion coalition built for the first AI deployment becomes the capacity base for every subsequent one; institutions that build it now compound their governance advantage as the AI footprint grows.
• Third-party model risk under OCC Bulletin 2026-13 requires that vendor-provided AI tools receive the same governance rigor as internally built ones; the technology and compliance champions are the operational owners of that vendor oversight, making their roles critical even for banks that do not build AI internally.