Finding method in the madness: Practical steps for good AI Governance

Why good AI governance is about creating the conditions for to scale safely, consistently and commercially

When people hear ‘AI Governance’, they tend to fall into one of two camps. They either switch off immediately or start nervously perspiring. And I get it. It’s easy to dismiss governance as a problem for other, less innovative and visionary people, or something to worry about tomorrow. But, as AI tool adoption grows, and we empower systems with more and more proprietary data and access, these positions need to change. For all businesses, but especially enterprise organisations, governance is becoming the conversation they need to have before their AI adoption gets ahead of them.

Recent stats support this concern. 64% of businesses are reporting transformational impacts from AI adoption, yet fewer than one in five have a thorough governance process in place. In other words, businesses are benefiting from the upside of AI without fully managing the potential downside, and over time, this imbalance will become harder to ignore. 

In my experience, the brands getting governance right aren’t focused on achieving perfection from day one. Instead, they’re prioritising a pragmatic, scalable approach that does four things:

• Understands the risks 
• Creates the space for experimentation
• Maintains close control of data, brand and direction
• Builds shared capability and ownership

Understanding the risks 

Every business needs to understand their own level of risk, based on their activities and appetite. But for most companies and their agency partners, the biggest risks revolve around the data (where it is gathered, how it is retained, and how it's used), accuracy, and bias (both implicit and explicit) within the models themselves. If you don’t have a handle on these areas, you are opening yourself up to a boatload of potential challenges.

The first step in managing these types of risks is not gaining complete control but establishing an appropriate level of understanding and visibility. Teams need to know the tools they’re using and the implications of how they operate. In practice, this often means favouring enterprise-grade tools over free alternatives, especially where data protections and contractual clarity are stronger.

Another way to look at it is to think of AI-powered tools as wrappers around LLM providers. You can’t treat an AI tool as a black box, so good governance requires visibility of what sits underneath. Predictable, consistent outputs come from an understanding that the same model can behave differently depending on how it’s used and by whom. Governance is therefore concerned with the implementation, as well as the model itself. 

Create space to experiment

Across all industries, AI is a sometimes-unpredictable technology, subject to constant evolution and change. Amidst this fast-paced background, the one thing developers, agencies and their partners can agree on, is that AI adoption is happening in an environment of uncertainty. No one organisation is in a position to confidently predict what their operating model will look like in five years. And in respect to this context, governance can’t be overly restrictive.

In our own experience, the approach has been ‘controlled experimentation’, with teams encouraged to explore, test capabilities, and understand both strengths and weaknesses of tools by using them in low-risk ways. Most initial outputs are imperfect, with duplication, dead ends, and partial solutions. Much of this experimentation will, in all likelihood, be discarded. However, it’s what we learn in the process that’s valuable. We still have a responsibility to ensure activity is occurring within known boundaries, and with the risks properly considered, but, simply put, ‘failure’ is all part of the journey.

We've taken this approach because there’s currently a gap in the market. Right now, many tools work well for individuals but lack enterprise-level controls and governance features. This will undoubtedly improve, but in the meantime, we still need a layer of internal policy and training to reduce risk without stifling experimentation.

Maintain control of data, brand and direction  

One of the biggest considerations isn’t AI misuse – it’s a dilution of the things that make you unique and interesting. As AI content continues to scale, 94% of businesses are now reporting that ‘preserving brand integrity’ is a primary concern.
By default, AI-generated content tends to be neutral, generic, and lacking personality. ‘Average’ answers and a lack of lived in experience is fundamentally at odds with most brand strategies, which aim for distinctiveness in tone of voice and personality.

Teams can mitigate this by: 

• Providing structured brand guidance and instructions to AI tools to work within your boundaries
• Embedding TOV rules into prompts and workflows and ensuring everyone has access to the right information to deliver consistently
• Using validation tools to assess output quality and ensure standards are met

A lot of what I’ve said above is still tech focused. But as we keep saying, technology alone isn’t the answer. Despite the doomsdayers saying otherwise, the human component in AI adoption remains essential. No content, research, or outputs should be published without a human review, and this responsibility needs to sit with the individual approving the output. AI is a tool, not an author. The work ultimately belongs to the person who signs it off, just as the artist, not the paintbrush, has ownership over a finished portrait.

Build shared capability

Over the next 12-24 months, success will ultimately come down to how businesses manage tools and capability. 

I anticipate the biggest challenge being tool sprawl. Often, individuals choose their own tools, which restrict the ability to train, scale and share capabilities efficiently. Organisations that standardise, while still allowing controlled flexibility, will outperform those that do not. True, this doesn’t lend itself to immediate results, but over time, shared capability, consistent training, and aligned usage will create a stronger foundation.

At the same time, beware of over reliance on any single provider. Today’s pricing is unlikely to hold, and as costs increase, deeply integrated tools could become a commercial risk. The solution here is balance standardising where AI tools build capability while staying flexible, where it protects resilience and innovation. 

Take AI governance seriously with IDHL

At the end of the day, in the current AI gold rush, it’s tempting to follow the old Facebook strategy of ‘move fast and break things,’ but that only works until it doesn’t. You only need to look at the recent issues with AI data validation (Deloitte) or AI chatbots providing incorrect advice to customers (Air Canada) to see that the risks I am talking about are real and potentially very expensive if you end up on the wrong side of the law, or your audience.

Good AI governance is not about slowing adoption down. It is about creating the conditions for AI to scale safely, consistently, and commercially. It keeps you moving in the right direction, enabling safety without restricting innovation. To understand how IDHL can help you leverage AI opportunities without risking your reputation, get in touch with our experts today.