Why transparency and explainability are now the currencies of AI success.

Fast forward to today, and the idea of AI tools engaging with us like that seems far less surprising. And the AI demo has turned out to be a landmark moment, but not necessarily in the way everyone expected. Initial amazement at the bot’s ability to mimic human speech has steadily morphed into concerns about the ethical questions it surfaced.

Had the hair salon consented to talking to AI? Or was the staff member unfairly duped into thinking they were interacting with another person? As AI’s prevalence and capabilities have evolved, these issues of transparency and explainability have become supercharged. And for businesses, trust is now key to AI success.

Distinctly related

First, let’s quickly bust a myth: Transparency and explainability aren’t the same thing. Transparency is about disclosure. It means being clear with people about when they’re interacting with AI. Alongside fears about privacy and security, failure to provide this clarity pricks a long-standing human nerve; no one likes the idea of having the wool pulled over their eyes. So, when organizations omit transparency, they open up a raft of reputational and loyalty risks.

On the flip side, openness about AI can be a relationship builder. One business I know of has set clear ethical rules around when its call center uses virtual agents and when it doesn’t. For example, when someone calls about death benefits, they’re always connected to a human. This signals to customers that the company is taking their concerns about appropriate AI use seriously.

Explainability goes deeper. It’s about being able to articulate how and why an AI system reached a certain conclusion. Large language models are certainly improving here and now also routinely cite information sources, but this remains imperfect and needs to go across the full AI spectrum.

As a business, can you explain how your model forecasts rising costs during supply chain disruptions? Does it recommend products based only on past behavior? Why exactly is it suggesting a new market expansion? Without this reasoning, leaders, customers and employees are left in the dark, and the outcomes can’t be fully relied on.

Reinvent the wheel — of governance

C-suites and boards are starting to press harder on transparency and explainability too. Many are asking their AI leaders detailed questions about governance frameworks in the same way they do with technical capabilities.

For most, the answers lie in establishing a new, multi-lever “wheel of governance.” This goes beyond just transparency and explainability to include other key trust drivers, like sustainability, value realization, bias, security and an AI ethics compass mapped to corporate values. These levers can then be adjusted according to an organization’s stakeholder network and operating landscape.

The wheel’s effectiveness relies on balance. Charge ahead recklessly on AI innovation, and firms risk falling short of stakeholders’ and/or regulators’ expectations. Yet become paralyzed by governance, and they may end up drowning in bureaucracy and missing opportunities to progress. The sweet spot lies in between: robust oversight that still moves at the speed of innovation.

Principles in action

It also requires a bias for action. Many organizations treat governance as a theoretical exercise that looks neat on a framework slide but doesn’t actually shape daily decisions. But the reality is it needs to be practical, starting with prioritizing resources and the parts of the business they should be focusing on reimagining with AI.

If you’ve got a backlog of 300 AI use cases, not all deserve the same scrutiny. Instead, leaders should zero in on programs where the stakes — and dollars — are highest. If an AI solution can impact growth, it should rise to the very top of the governance agenda, including being allocated the right team. Experiments that boost internal productivity by half a percentage can wait.

Boards and C-suites can add discipline here too by assigning ownership of specific governance pillars to senior executives and tying that directly to performance goals. When leaders know their compensation depends on whether the AI program under their remit is trustworthy, governance stops being a vague “shared responsibility” and becomes a concrete action.

A question of intelligence

Of course, AI transformation is never a one-and-done; it’s a journey with many milestones. We’re now at a stage where AI can take on entire processes, but humans remain essential to executing them — overseeing outcomes, questioning conclusions and applying judgment.

This balance will shift as the technology, workforce and regulatory landscape evolve. Will AI agents disrupt functions like finance? In short, yes. But are we heading to a future where banks are staffed entirely by bots? Not likely. The practical way point is using AI for heavy lifting, while we act as guardians and decision-makers.

There will be nuances too. An AI tool that lets a retailer use customer data to make decisions carries different reputational threats than a healthcare provider doing the same thing. Likewise, organizations cannot let themselves be held back by the fragmented US policy environment. It’s far better to prepare now for areas most likely to be regulated, such as facial recognition and those that intersect with existing data privacy policies.

Whatever your industry, transparency and explainability are now about more than compliance checkboxes — or booking haircuts. They’re competitive advantages. Customers today are savvy. If you try to obscure your AI usage, they’ll feel misled. If you can’t explain why your AI reached a certain decision or outcome, they’ll question its credibility.

In both cases, the impact on your bottom line may be severe. However, if you’re forthright, practical and proactive, you’ll avoid this risk in the short term, and you’ll earn people’s trust in the long term too. In a world where growth is complex and loyalty is fleeting, that may just be AI’s most valuable currency of all.

This article was originally published on FastCompany.com.