Where should a business start with AI?
A practical framework for identifying high-value opportunities.
Most AI initiatives don't fail because the technology doesn't work. They fail because they start from the technology instead of the problem — a team picks a tool, then goes looking for somewhere to use it. The businesses that get real value tend to do the opposite: they start with a specific, named bottleneck, and only then ask what technology might solve it.
Start with friction, not features
The best starting points are usually hiding in plain sight: the process everyone complains about, the report that takes a day to assemble by hand, the question customer support answers fifty times a week. If a task is repetitive, manual, and involves pulling information from more than one place, it's a strong candidate — regardless of whether "AI" was ever mentioned in the same sentence as the complaint.
A simple framework
Four questions are usually enough to sort a long list of ideas into a short one: What should be automated outright, because it's rule-based and repetitive? What should be augmented, giving people better information or faster analysis rather than removing them from the loop? What should stay entirely human, because judgment, relationships or accountability matter more than speed? And what new opportunity does this open up that wasn't possible before?
Prioritise by impact and feasibility
Not every good idea is a good first project. The strongest early candidates sit at the intersection of real business impact and low integration complexity — something that touches revenue, cost or customer experience, but doesn't require rebuilding five systems to reach it. Save the harder, higher-effort bets for once there's a track record of shipping smaller ones successfully.
Common mistakes
The same few mistakes show up repeatedly: buying a tool before the problem is clearly defined, treating integration with existing systems as an afterthought, and launching something with no clear owner responsible for whether it actually works six months later. Avoiding these matters more than choosing the "right" AI model.