AI 每分鐘都在作出決策，企業治理卻仍按季度進行

An AI-assisted system can influence hundreds of operational decisions before a steering committee has scheduled its next meeting. Governance built around quarterly reviews was never designed for that speed, and this mismatch is where many enterprise AI programmes begin to struggle.

We spend enormous amounts of time debating which models to deploy. Yet Gartner predicts that through 2026, organisations will abandon 60% of AI projects unsupported by AI-ready data. Most enterprise AI failures have very little to do with the model itself. They stem from something more fundamental: organisations do not have a trusted, connected understanding of their own business.

The Hidden Cost of “AI-Ready” Data

Most enterprises underestimate the effort required to establish ownership, governance, trust and context around their data. Few invest in the cross-functional expertise and operating model needed to build governance frameworks that deliver the transparency, explainability and accountability required to take AI beyond pilots.

There is a simple test I urge customers to take. Can the organisation clearly explain who owns the data, how it relates to business processes, what controls govern its use, and who remains accountable for AI-assisted decisions? If not, their AI initiatives are unlikely to succeed, regardless of how much they invest in platforms, infrastructure or models.

The first step in any enterprise AI journey should be to invest in governance and a practical operating model that establishes clear ownership, accountability, and decision rights. What I am seeing is that most organisations do this in reverse, putting models and infrastructure in place before the foundations required to operate AI at scale.

But AI-assisted decisions rely on more than individual data points, they also depend on understanding how information relates to business processes, policies, ownership and operational context. When those relationships are fragmented, organisations should not be surprised when AI recommendations become inconsistent.

When AI applications experience near misses, failures or conflicting recommendations, the questions that matter are organisational rather than technical. Who owns the data? Who owns the decision? Who can override the AI? How are exceptions escalated? Without answers, enterprises risk creating automation without accountability.

Trust Is Built Before Deployment

Trust in AI is established long before any model goes live. Effective governance requires collaboration across technology, business functions, risk, legal, compliance and security because AI systems operate across organisational boundaries while accountability often remains siloed. Many governance challenges attributed to AI are symptoms of older problems with data ownership and decision-making authority that the technology has simply exposed.

Most organisations already have governance through security policies, compliance frameworks and approval processes. What they often lack is a connected view of how those pieces fit together.

Trust develops when stakeholders understand the rationale behind recommendations, have confidence in the underlying data and know that appropriate controls and escalation mechanisms are in place. None of that can be retrofitted after an incident.

When Quarterly Governance Meets Minute-to-Minute Decisions

Traditional governance models rely on periodic reviews, steering committees and retrospective reporting. AI-assisted systems generate recommendations continuously and at scale. Existing frameworks are not wrong, but they are too slow and disconnected from operational decision-making. A quarterly review cycle cannot govern decisions being made every hour of every day.

Where I see the most success is with organisations that embed governance directly into operational workflows through policy enforcement, approval workflows, monitoring, auditability, and human oversight. They have committed to moving from periodic to continuous governance, so they can see in real time which data is being used, which decisions are being influenced, who is accountable, and when human intervention is required.

Continuous governance is more than monitoring AI outputs. It means understanding what information AI is using, whether it remains valid and whether decisions still align with business policies and objectives.

AI Will Disagree With Itself

Large organisations already manage competing priorities every day. Security teams want to reduce risk, operations teams want stability, business teams want speed, and finance teams want cost efficiency. Now enter AI systems, which are trained on different datasets and objectives and will naturally reflect those priorities.

A security AI may recommend immediate remediation of a critical vulnerability, while the operations team delays the change because the affected system supports a critical business service. Both positions can be valid. Resolving these conflicts requires clear decision rights, defined risk ownership and escalation paths.

AI should provide the data, context and recommendations, but accountability for balancing business, operational and risk considerations must remain with the organisation.

Handling Scepticism When Inconsistencies Surface

When AI produces an inconsistent recommendation, it is obvious that the business will be sceptical of it in the future. Trust is rebuilt when stakeholders understand why a recommendation was made, what data was used and who remains accountable for the outcome.

AI often exposes data quality issues, ownership gaps and process inconsistencies that already existed within the organisation. Treating these moments as opportunities to improve governance and operating procedures is far more productive than defending the technology. Over time, trust is earned through transparency, consistency and demonstrable business outcomes.

The Trusted Information Foundation

The conversation around AI readiness tends to focus on models, infrastructure and tooling. In practice, the organisations seeing the greatest success are investing in something more fundamental: a trusted information foundation. It provides the business context that explains how information connects to processes, policies, ownership and operational decisions, giving AI a stronger basis for recommendations that organisations can understand, govern and trust.

Even the most advanced AI model cannot compensate for weak information governance, unclear ownership or poor business context. What organisations need to do is build the context, governance, institutional memory and decision intelligence needed to use AI responsibly and effectively at scale.

That work is slower and less visible than deploying a new model, but it ultimately determines whether the model delivers lasting business value.