Preview build — not the live site. Feedback welcome.

[email protected]

United Kingdom · South Africa · Australia

Insights · 10 November 2026 · AI Governance | Series | Part 6

From Principles to Practice: NovoFinity's Approach to AI Governance

  • Albert VenterChief Executive Officer, NovoFinity

Over the course of this series, we have explored a fundamental challenge facing organisations adopting AI: how they realise its potential while ensuring that it operates safely, responsibly and as intended. We have looked at the importance of data, the governance of models, the growing role of observability, and the need for effective assurance and clear accountability.

Each of these matters, and the challenge is bringing them together.

AI governance cannot simply be a collection of policies, controls and committees operating independently around the technology. As AI becomes more embedded in business processes, decisions and customer interactions, organisations need a more integrated approach — one that connects data, models, observability, assurance, accountability and control throughout the AI lifecycle.

This is at the heart of NovoFinity’s approach to AI governance, and the final article in our series, following Assurance and Accountability: Proving AI Behaves as Intended.

AI Governance Is a System, Not a Checklist

There is an understandable temptation to approach AI governance by breaking the problem into individual components. Is the data governed? Has the model been validated? Are appropriate safeguards in place? Is the system being monitored? Has someone been made accountable?

All are important questions. But answering each individually does not necessarily mean that the AI system is well governed.

The components are interconnected. Poor-quality or inappropriate data can affect model behaviour. Changes in model behaviour can create new risks. Those risks may only become visible through effective observability. What is observed then needs to trigger appropriate controls, escalation and intervention. And throughout that chain, someone must ultimately remain accountable.

Governance therefore needs to operate as an integrated system rather than a series of independent activities.

Start With the Data

AI governance begins before the model. AI systems depend upon data for training, grounding, context and decision-making. Understanding the provenance, quality, sensitivity and permitted use of that data is therefore fundamental.

Traditional data governance remains important, but AI introduces additional considerations. It is no longer enough to ask whether data is available and technically fit for purpose. Organisations also need to understand whether its use is appropriate within the context in which the AI is operating, and whether changing data could materially alter outcomes.

This creates the first layer of an integrated AI governance capability: understand and govern the data upon which AI depends.

Understand the Models and Systems Using It

The next layer is the AI itself. Organisations need visibility of the models, applications and increasingly agents operating across the enterprise, together with an understanding of their purpose and the business processes they influence.

Not every AI application presents the same level of risk. An internal productivity tool is fundamentally different from an AI system influencing a customer decision, or an autonomous agent capable of taking actions across multiple systems.

Governance therefore needs to be proportionate to the potential impact. The higher the risk and autonomy, the stronger the governance, controls and evidence required.

But governance cannot end when a system is approved for production. That is where observability becomes critical.

Move From Approval to Continuous Visibility

Traditional governance often concentrates on decision points: assessment, approval, deployment and periodic review. AI requires something more dynamic.

Models, data, prompts, integrations and patterns of use can change. Behaviour that was acceptable during testing may not remain so indefinitely. Organisations therefore need visibility into how AI is behaving once it is operating in the real world.

Observability provides that bridge. It allows organisations to move beyond asking whether appropriate controls were designed, towards understanding whether the system continues to behave within agreed boundaries. This shifts governance from something predominantly periodic towards something increasingly continuous.

You cannot effectively govern what you cannot observe.

Connect Observability to Assurance

Visibility alone, however, is not governance. Knowing that something has happened does not tell an organisation whether the appropriate controls operated, whether an issue was addressed, or whether behaviour remains within approved tolerances.

This is where observability and assurance need to connect. Assurance provides confidence that governance is working as intended. Rather than relying solely on policies, control descriptions and periodic reviews, organisations should increasingly be able to demonstrate evidence of how their AI systems are operating and how risks are being managed.

The question moves from whether we have the right controls to demonstrating that those controls are working. That distinction becomes increasingly important as AI systems become more autonomous and more deeply embedded within business operations.

Keep Accountability Human

AI may make decisions, generate recommendations or increasingly execute actions. But accountability cannot be delegated to an algorithm.

Every AI system needs clear ownership. Someone must understand why it exists, the risks associated with it, the boundaries within which it should operate, and what happens when those boundaries are breached. That accountability needs to extend across the lifecycle, not simply exist at the point of approval.

Technology teams, data teams, risk functions and business leaders may all have responsibilities. But responsibility distributed across several functions cannot result in accountability belonging to nobody. As AI autonomy increases, human accountability becomes more important, not less.

One Integrated Governance Capability

Bringing these elements together creates a different view of AI governance:

  • Data establishes what the system relies upon
  • Models and systems determine how AI is applied
  • Observability creates visibility into how it behaves
  • Controls establish the boundaries within which it should operate
  • Assurance provides evidence that those controls are effective
  • Accountability ensures that people remain responsible for the outcomes

These should not operate as separate governance silos. They should operate as one connected capability across the AI lifecycle, with information flowing between them and governance adapting as the risk, behaviour and use of the AI changes.

That is the NovoFinity view of AI governance.

Governance Should Enable Confidence, Not Create Friction

There is an important final point. Good AI governance should not be designed simply to restrict AI. Its purpose is to create the confidence required to use AI responsibly.

If organisations understand the data their AI uses, know where models and agents are operating, can observe their behaviour, maintain appropriate controls, demonstrate that those controls work and establish clear accountability, they are in a much stronger position to move successful AI initiatives beyond experimentation.

Governance then becomes an enabler of adoption rather than an obstacle to it. And perhaps that is the most important shift.

From Governing AI to Trusting AI

Throughout this series, we have argued that governing AI requires more than frameworks and policies. It requires organisations to connect what they intend their AI to do with what it is doing, while maintaining the controls, evidence and accountability necessary to intervene when those two begin to diverge.

For NovoFinity, that is the foundation of responsible AI at scale:

  • Govern the data
  • Understand the models
  • Observe the behaviour
  • Control the risk
  • Assure the outcomes
  • Maintain human accountability

Because ultimately, the objective of AI governance is not simply to demonstrate that AI has been governed. It is to create the confidence to use it.

If your organisation is considering how to strengthen AI governance, or move AI initiatives from experimentation into production with greater confidence, we would welcome the opportunity to share our perspectives and client experiences, and to discuss the challenges you are navigating.

About the authors

Albert VenterChief Executive Officer, NovoFinity

Albert Venter is a Data & AI Governance leader and Chief Executive Officer at NovoFinity, specialising in financial services and regulated industries. He helps organisations turn complex data, AI and regulatory challenges into practical, measurable business value. His expertise spans data and AI governance, strategy, BCBS239, automation, cloud transformation and process optimisation, with experience delivering major initiatives across Australia, Africa and Europe.

Back to Insights

Ready to move AI into production with proof?

Leave your details and we will arrange a conversation.

We reply within one business day. We never sell your details. Privacy notice