Most organisations are asking the wrong first question. Instead of "what AI should we build?", the more important question is: "what kind of AI transformation do we actually want?"

There are two fundamentally different paths. The first is evolutionary, you implement AI to improve and augment your existing processes, then revisit how those processes work once you've seen what AI can do. The second is transformational, AI supersedes your current ways of working from the start, replacing rather than enhancing them.

Neither is wrong, but conflating them leads to misaligned expectations and wasted investment. Getting this answered early sets the direction for everything that follows.

Before any AI is built, you need a clear picture of what's actually worth automating. This means auditing your business processes and operations through a service design lens, mapping what can be automated, what genuinely benefits from AI, and what doesn't need AI at all. Not everything should be AI-powered, and forcing it where it doesn't fit creates complexity without value.

The output of this phase is a set of service blueprints: clear maps of your processes, which parts are candidates for AI, the potential ROI of each, and a set of benefit themes that act as a north star for the transformation ahead.

Once you know where you're going, you need to define how you'll know you're getting there. Success measures should be set before any building starts, so goals are unambiguous and progress is measurable against something real, not a feeling.

Alongside this, horizon scanning is a valuable exercise: mapping what's emerging in the AI landscape, what can be leveraged as an opportunity, and what risks or shifts might need to be navigated along the way.

With goals and process maps in place, you can create a prioritised opportunity map, a ranked view of the most valuable problems to solve, aligned to both business objectives and real user pain points. This is built by combining business goals, success measures, and the user needs surfaced through service design.

The inputs come from internal interviews, workshops, and where appropriate, AI tools to help synthesise and prioritise. The result is a clear, defensible view of where to focus, not just what's technically possible, but what's actually worth doing.