Skip to content

Content Library

Back to Event Q&As

Event Q&A November 3, 2026

Enterprise AI Success Depends on Trust Governance and Clear Accountability

Enterprise AI Success Depends on Trust Governance and Clear Accountability

Wei Ma, Managing Director, Head of Risk Modelling Centre of Excellence, SMBC

As organizations move from AI experimentation to enterprise-wide deployment, governance, accountability, and trust are becoming critical success factors. In this Q&A, Wei Ma shares insights on scaling AI responsibly, highlighting the importance of secure architectures, clearly defined ownership, continuous monitoring, and system-level governance. The discussion explores how organizations can move beyond pilots and build the foundations needed for sustainable enterprise AI adoption.
  • Examines the governance and security challenges that emerge when scaling AI beyond pilot programs.
  • Explores how organizations can define ownership across business, technology, risk, and compliance teams.
  • Discusses the benefits of reusable AI platforms and governance frameworks.
  • Highlights the shift from model-centric governance to system-level oversight.
  • Explores the growing importance of continuous assurance, explainability, and traceability.
  • Shares practical advice for organizations looking to scale AI adoption successfully.

display-middle-ads-here

1. What has been the biggest governance or security challenge you've encountered when moving from AI pilots to enterprise-wide deployment, and how did you address it?

The biggest challenge has not been the AI model itself. It has been establishing trust around data access, output reliability, and accountability once AI moves beyond a small pilot group and starts interacting with production information and business processes.

In a pilot, teams can often rely on limited datasets, manual reviews, and subject matter experts to identify issues before outputs are used. At enterprise scale, those controls no longer scale. The key risk becomes ensuring that the right people can access the right information, that sensitive data is protected, and that AI-generated outputs are explainable and auditable.

Our response has been to treat governance and security as design requirements rather than post-implementation controls. That includes:

• Role-based access and least-privilege entitlements.

• Data classification and masking of sensitive information.

• Retrieval and prompt-response logging for traceability.

• Source-grounded outputs rather than unconstrained generation.

• Formal human-in-the-loop review for higher-risk use cases.

• Ongoing monitoring of output quality, usage patterns, and model behavior.

The lesson we learned is simple: pilot success is usually driven by speed and experimentation, while enterprise success is driven by trust and control. Organizations that build governance into the architecture from day one scale much faster than those that try to retrofit controls later.


2. How have you structured ownership and accountability for AI across business, technology, risk, and compliance teams to avoid duplication or gaps in decision-making?

 One of the most important lessons has been recognizing that AI ownership cannot sit entirely within either technology or the business.

We have found it useful to separate accountability into three distinct layers:

• Business and AI development teams own what the AI does. They are accountable for the use case, prompts, retrieval strategy, analytical methodology, expected outcomes, and ongoing performance of the AI capability.

• Technology owns how the AI runs. They are responsible for infrastructure, hosting, security, identity management, observability, integration, and operational resilience.

• Risk, governance, compliance, and independent validation provide challenge and oversight. They establish standards, stage gates, testing requirements, approval processes, independent review, and ongoing monitoring expectations.

 What has worked particularly well is defining decision rights across the full lifecycle, from intake and risk assessment through design, build, validation, release, and monitoring. Without that clarity, many pilots stall because teams debate ownership rather than delivering value.

An emerging industry trend is the adoption of AI lifecycle governance models that mirror model risk management disciplines, with clearer accountability matrices and documented approval gates for higher-risk AI applications.

 

3. Looking back at your AI scaling journey, what operating model decision delivered the greatest impact on adoption and value realization?

 The most impactful decision was shifting from individual AI projects to a reusable AI capability model.

Many organizations run successful pilots, but each one becomes a custom implementation with separate architecture, controls, prompts, testing approaches, and governance reviews. That does not scale.

Instead, we focused on creating reusable components that can support multiple use cases, including:

• Retrieval-augmented generation capabilities.

• Human approval workflows.

• Monitoring and feedback loops.

• Audit trail and traceability mechanisms.

• Risk classification and exception management patterns.

This approach transforms AI from a collection of disconnected pilots into a governed enterprise capability. Once common patterns are established, the effort required to launch new use cases drops significantly while governance consistency improves.

The operating model shift is really from "building AI applications" to "building AI platforms and reusable governance frameworks."

 

4. As regulations and AI capabilities continue to evolve, what changes do you anticipate organizations will need to make to their governance and risk management frameworks over the next 2–3 years? 

I believe three major changes are coming.

First, governance will move from model-centric to system-centric. 

Traditional model governance was designed for relatively static analytical models. GenAI systems increasingly incorporate foundation models, retrieval layers, agents, tools, workflows, external data sources, and human approvals. Governance frameworks will need to assess the entire AI system rather than only the underlying model.

Second, continuous assurance will become more important than pre-deployment approval.

Many organizations still focus heavily on upfront validation. That remains important, but the real challenge is ongoing monitoring of output quality, drift, usage behavior, emerging risks, and changing external models. Governance frameworks will increasingly require continuous evidence that controls remain effective after deployment.

Third, explainability and traceability requirements will increase.

Regulators are increasingly interested in how decisions are made, what information influenced outputs, and whether organizations can demonstrate accountability. Retrieval logs, source attribution, audit trails, evaluation records, and human review evidence will become standard governance artifacts rather than optional documentation.

Overall, I expect governance frameworks to become more dynamic, lifecycle-oriented, and evidence-driven than they are today.

 

5. What differentiates organizations that will successfully scale AI across the enterprise from those that remain stuck in the pilot phase, and what should leaders be doing today to prepare?

The organizations that succeed are not necessarily the ones with the most advanced models. They are the ones that build trust alongside innovation.

 In my experience, the differentiator is whether an organization can bridge the gap between pilot value and enterprise trust. The technology challenge is often easier than the operating model challenge.

Successful organizations typically share several characteristics:

• They prioritize use cases based on both business value and control readiness.

• They establish clear ownership and decision rights.

• They invest early in secure AI architecture and data governance.

• They implement robust testing, validation, and monitoring disciplines.

• They create reusable AI platforms and governance patterns.

• They invest heavily in training, adoption, and change management.

Organizations that remain stuck in pilot mode often focus primarily on model experimentation and underestimate the importance of governance, accountability, and operational readiness.

For leaders preparing today, my recommendation is to spend as much time designing the operating model as selecting the technology stack. Build governance, validation, data controls, and accountability into the journey from the beginning. AI at enterprise scale is ultimately a trust challenge as much as it is a technology challenge.

As I often summarize it: scaling AI is not about deploying faster at any cost. It is about building the control, ownership, and trust required for sustainable enterprise adoption.

Wei Ma

Managing Director, Head of Risk Modelling Centre of Excellence, SMBC

Wei Ma is Managing Director and Head of Risk Modelling Center of Excellence at SMBC U.S., leading enterprise risk analytics, model development, and AI-enabled risk management initiatives. With more than 20 years of experience in banking and financial risk management, Wei focuses on applying AI and advanced analytics to modernize how financial institutions identify, measure, and manage risk. Before joining SMBC, Wei held senior quantitative and structured finance roles at AIG, Merrill Lynch, and PwC. He also teaches graduate-level courses in risk management and analytics at Rutgers University, helping develop the next generation of risk and AI professionals. Before joining SMBC in 2013, Wei worked for five years at AIG where he managed structured product portfolios composed of CDO, RMBS, and CMBS assets with a notional amount of over $60 billion. He also led the development of structured product risk analytic capabilities at AIG Asset Management and led work streams in stress testing and model risk management initiatives.

Events you may be interested in

View all of our upcoming events

Connect Plus

Unlock more with Connect Plus membership

This content is available to Connect Plus members.