Most AI Ideas Are Not Worth Doing Yet — Here's How to Tell Which Ones Are

The List Is Never the Problem

Sit any leadership team down and ask where AI could help, and within ten minutes you'll have a long list — automate this report, draft that email, summarise these tickets, add a smart feature to the product. Generating AI ideas has never been the bottleneck for software companies. Deciding which ones are actually worth doing, in what order, with what resourcing, is where almost every company gets stuck, and it's why so many AI initiatives are simultaneously started and none are finished.

The instinct is usually to chase whichever idea is most exciting or most visible — often the customer-facing product idea, because it photographs well in a board deck. That instinct is exactly backwards. Without a structured way to compare ideas against each other on consistent criteria, the loudest idea wins, not the best one, and the company ends up with several half-built AI initiatives instead of one that's actually delivering value.

Three Criteria, Not One

A use case prioritisation matrix scores every AI idea against three things: value, feasibility, and risk. Value asks how much this actually matters if it works — time saved, cost reduced, revenue influenced, customer experience improved — stated in terms specific enough to be falsifiable, not 'this would help a lot.' Feasibility asks how realistic this is to actually build or implement with current data, tooling, and skills, not with the tooling you wish you had.

Risk asks what goes wrong if the AI output is wrong, and how bad that is — a mistake in an internal draft email is a different order of risk than a mistake in a customer-facing financial calculation or a piece of code shipped to production. Scoring all three, rather than just value, is what prevents companies from chasing high-value ideas that are actually unrealistic this quarter, or low-risk ideas that are barely worth the effort.

Why High-Value Isn't Automatically High-Priority

The instinctive ranking is almost always by value alone, which is why so many companies commit to ambitious, high-value AI product features before they've built any operational muscle — the exact premature productisation pattern that pillar-sequencing is designed to prevent. A prioritisation matrix forces a more honest comparison: a modest, high-feasibility, low-risk internal automation might be lower value on paper than a flagship product feature, but it's achievable this quarter, builds real capability, and creates almost no downside if it underperforms.

The strongest early AI portfolios are usually a mix — a few quick, low-risk wins that build momentum and organisational trust, alongside one or two higher-value bets that are given proper time and resourcing because they cleared the feasibility and risk bar honestly, not because they were the most interesting idea in the room.

Making the Matrix a Living Decision Tool

A prioritisation exercise done once and filed away doesn't help much six months later when the list has changed and half the assumptions about feasibility have shifted because tooling improved. The matrix works best as something revisited every quarter alongside the broader AI roadmap conversation — re-scoring existing ideas, adding new ones, and retiring ideas that turned out to be lower value than they looked.

Use the AI Use Case Prioritisation Matrix to score every AI idea currently in play or under consideration, and let the scoring — not enthusiasm or visibility — determine what gets resourced next.