The AI Play Matrix: Choosing the right path for AI innovation

GenAI has captured organisational imagination like nothing I’ve seen since the early days of the web. The launch of ChatGPT in late 2022 marked a watershed moment, making AI suddenly feel tangible and accessible to everyone, not just data science teams.

Every organisation I work with now has some kind of AI initiative. Board presentations often feature AI strategies. Innovation labs are buzzing with AI experiments. Press releases herald transformative AI journeys.

But here’s the thing: the technology isn’t the barrier anymore.

The models are getting better by the month. The APIs are straightforward. The tooling is increasingly accessible. Cloud providers are throwing AI services at you faster than you can evaluate them.

The barrier is people. How do we think about these new capabilities? How do we approach problems differently when solutions are evolving at this rapid pace? How do we move beyond the obvious use cases everyone else is implementing?

Getting people to think differently about how they work with these technologies is crucial. Because if you’re just using AI to do slightly better versions of what you already do, you’re missing the point.

This all stems from a question I keep getting asked:

We want to be innovative with AI. Can you show us someone who’s done it before?

It used to stump me, but I’ve realised the answer is simple: I can show you people who’ve been innovative with AI before. It’s highly unlikely that they do exactly what you do in exactly your context. Your job is to interpret that inspiration and make it work for you.

Enter The AI Play Matrix.

The Matrix

We need to consider AI in the context of business problems and the solutions it provides. Some business problems are well understood, while many are emerging or unclear. Some AI capabilities are proven and understood, while many are experimental or in the process of evolution.

• The AI Plan quadrant is where most organisations want to live. Known business problems, proven solutions. This is about scaling AI experiments that have proven themselves. It’s incredibly valuable. There also isn’t much in this quadrant at the moment, but it’s the natural mode of operation for most businesses.
• The AI Follow quadrant is where a significant amount of AI activity currently occurs. Proven AI capabilities looking for business problems that may or may not exist. “Let’s add AI to everything!” or “We need some generative AI because our competitors have it.” Easy to dismiss, but it does have value – sometimes you need to follow the herd to maintain credibility while you figure out what works. Also, if your competitors are doing something and you aren’t, then your brand might depend on it.
• The AI Adapt quadrant is where you know you have a business challenge, but aren’t sure which AI approach will solve it. “Our knowledge management is broken, but we’re not sure if it needs RAG, fine-tuning, or something else entirely.” This requires rapid experimentation and iteration.
• The AI Tinker quadrant is pure exploration. Playing with new AI capabilities to see what opportunities emerge. “What happens if we hook up this new vision model to our manufacturing data?” Most organisations find this challenging because there is no easily predetermined ROI.

The bottom half is where actual AI innovation happens. But it’s also where most businesses struggle because that’s not how they’re designed to operate.

The flow

There is a logical progression: ideas from AI Tinkering feed into AI Adaptive processes, which rapidly test concepts through prototyping until they’re ready to scale through planned delivery approaches.

The problem? Most organisations:

• Try to use the same approach for everything (“We are now completely AI-first”)
• Can’t manage transitions between quadrants
• Struggle to identify when different approaches are appropriate

The reality

Anything in the bottom half of the matrix can be genuinely difficult for large organisations. The procurement processes, governance models, investment processes, and risk frameworks are all designed under the assumption of certainty about the solution.

But if you want to do something genuinely innovative with AI rather than just following what everyone else is doing, you need to get comfortable operating in the bottom half.

So, what skills do you need for AI exploration instead of AI engineering? Let’s find out…

The skills of AI exploration

Now we’ve established that different AI approaches require different mindsets and methods, let’s unpack what that means in practice. Let’s examine the skills required for effective AI innovation.

The three mindsets

1. Connect

First, there’s a collective mindset that wants to connect.

• Work openly with AI experiments. Share what you’re learning as you learn it. Not corporate AI theatre or breathless press releases about “transformative AI initiatives,” but honest commentary on what’s actually working (and what isn’t). Document your failures as much as your successes.
• Build diverse AI learning networks. The best AI insights will come from different places. Talk to people from different industries, different roles, and different backgrounds. A retail person figuring out inventory optimisation might have insights for your HR challenge. Learn to leverage this diversity rather than just networking with people who are exactly like you.
• Understand AI team dynamics. AI projects need different team compositions for tinkering, adapting, and planning. A team that excels at implementing proven technology solutions may struggle to determine the best course of action with a new model. AI teams today often don’t understand which mode they’re operating in, let alone how to shift between them.

2. Experiment

Second, there’s an individual mindset that’s willing to experiment.

• Embrace AI uncertainty and impostor feelings. If you’re working with genuinely new AI capabilities, you won’t know what will happen. Get comfortable with that. The organisational pressure to have detailed project plans for emerging AI tech is real, but it’s also pointless. You’ll feel like you don’t know what you’re doing. That’s because you don’t. Nobody does with truly new stuff.
• Prototype and ship AI experiments rapidly. Turn AI ideas into something people can interact with as quickly as possible. Avoid spending months in PowerPoint or conducting endless analysis. Build rough chatbots, connect APIs, use no-code tools. Set tight deadlines. The only way to determine if an AI idea is effective is to test it with real users on real-world problems. The goal is learning, not perfection.
• Learn through AI play. Set aside time to mess around with new models, tools, and approaches without a specific business outcome in mind. Most organisations struggle with this because it feels unproductive. It’s not. It’s how you discover what’s possible. Yes, it feels childlike. Yes, that’s uncomfortable in a work setting. Do it anyway.

3. Build

Finally, there are practical approaches that help you build.

• Act like an artist, not just an analyst. Sometimes your job is to hold up a mirror to your organisation and show them what’s possible with AI that they can’t see for themselves. This feels uncomfortable and “unworklike” because it doesn’t fit traditional business roles.
• Copy and adapt AI patterns. Great AI implementations are rarely completely novel. They usually come from adapting what others have done (sometimes badly). Don’t feel you need to invent everything from scratch. Stand on the shoulders of the AI giants.
• Focus on AI outcomes, not AI hype. Strip out the buzzwords. Stop talking about “transformative AI journeys” and start talking about specific problems you’re trying to solve. If the AI component isn’t crucial to the outcome, perhaps you don’t need AI.

The reality check

Is this a foolproof process for AI innovation success? No.

No such thing exists for AI (despite what the consultants tell you), so don’t fall for it.

But if you want to do something genuinely useful with AI rather than just following what everyone else is doing, these skills will help you operate effectively across all the different approaches you’ll need.

The alternative is staying stuck in the Plan quadrant forever, implementing slightly better versions of what your competitors already have.

Your choice.