How trust and governance make or break AI investments

Trust and governance are crucial factors that determine the success or failure of AI investments. People won't rely on AI systems unless they comprehend and trust the information and decisions it produces. To ensure AI delivers meaningful results, it requires a robust foundation. This foundation consists of two key elements: a singular, trustworthy source of truth and a well-defined governance framework.

The reliable source guarantees that, in case of data conflicts, one authoritative record prevails, identifies asset ownership, and ensures compliance. Simultaneously, governance entails predetermined decisions regarding the necessity of human oversight before automated actions proceed. In simpler terms, this foundation calls for dependable data and a human presence to oversee the process.

While it may seem counterintuitive to prioritize governance in AI development, the truth is that organizations are increasingly embracing AI across various workflows. Nevertheless, only a small fraction of surveyed companies reported not using AI at all. Similar to relationships in the professional world, effective AI automation relies on clear understanding, which fosters trust.

AI excels at processing information and suggesting actions when it can tap into a reliable system of record, but it falls short of verifying what exists within an organization's environment, who owns each asset, and whether any of that information has changed recently or would pass an audit. Therefore, maintaining accurate records and assigning accountability is essential.

Creating this record may seem like a daunting task for many organizations, but it is more manageable than one might initially think. Start by conducting a discovery process that results in an authoritative inventory of all existing assets, attaching ownership and lifecycle information to each one. Keeping a change history is crucial, as auditors typically inquire about how changes were approved.

Once the record is established, decide on a method to resolve discrepancies when two systems disagree, and document this resolution. While some teams may be attempting to address these issues post-hoc, it is far more effective to address them beforehand. In a recent survey of 1,500 IT professionals, 68% reported experiencing AI-generated hallucinations with potential operational consequences.

Half of the respondents said a team member identified these issues before they caused problems, while 16% admitted that hallucinations led to issues. This catch rate often serves as the sole control framework for many companies. However, relying on individual alertness to catch errors is not an effective strategy for running a business.

Instead, categorize actions into two distinct groups before deployment. The first category includes routine, reversible tasks that can be executed autonomously, such as restarting a service or applying a standard patch. The second category encompasses actions that modify system-wide settings or involve live incidents, which should undergo human review.

Document each group within the workflow rather than leaving it to individual judgment. Surprisingly, despite 85% of IT professionals in the survey indicating that every AI agent and workflow should have a named owner in their organization, only 42% believe ownership is clear in practice. Even among those with policies in place, just 24% of employees stated that they consistently follow these policies.

While some may view this as tedious governance, it is essential to distinguish between governance as a concept and the execution of governance. It is not uncommon for IT professionals to cite governance, security, or compliance as the primary hindrance to AI deployment, with skills shortages and technology limitations trailing behind.

However, the issue lies not in governance itself but rather in how it is implemented. Governance functions as a means to establish expectations and manage accountability within a working relationship, eliminating uncertainty and providing clarity. By setting clear rules regarding data, ownership, and human review, organizations can expedite approvals and move forward more swiftly.

Implementing this approach may appear oversimplified—focusing on trustworthy data and maintaining a human presence in the loop—but it offers a wealth of benefits that can be realized more rapidly than anticipated. IT professionals who utilize AI at the most advanced level reported saving an average of six hours per week, twice as much as their peers at the earliest stage.

Moreover, half of these professionals reported that AI allows them to concentrate on more complex, strategic work. With trustworthy data and a human in the loop, organizations can confidently expand their AI programs, knowing they are building on solid ground. This approach not only streamlines operations but also enables the development of comprehensive, end-to-end automation systems that improve efficiency without unnecessary complications.

When considering resource allocation for the upcoming quarter, it is advisable to prioritize testing AI within one familiar workflow rather than hastily adopting a bolted-on solution. By reconciling the data presented by your systems, identifying the responsible individuals for approvals, and repeating this process a few times, you will establish trust in your records and position yourself for streamlined operations and confident decision-making.

As others continue to bolt on solutions reactively, you will be establishing a strong foundation for fewer headaches, improved efficiency, and more assured outcomes. Melissa Puls serves as the chief marketing officer.