What Business and AI Together Look Like: The Value vs. Effort Matrix I Actually Use

Coming up with AI ideas was never the hard part. Deciding which ones deserve to get built is. Every planning conversation has three initiatives that look strategic on a slide and capacity for one. AI made this worse, not better — ideas got cheap, so now everyone in the room has three of them.

Business and AI, without the deck

Most “AI transformation” talk starts from the technology: what can a model do now that it couldn’t two years ago. Wrong starting point. Start from a business outcome that already matters — a cost that’s too high, a decision that’s too slow, a queue that’s backed up — and ask whether AI is actually the right lever. Sometimes it isn’t. A slow decision caused by three people signing off in sequence is a process problem, not an AI problem, and an LLM will happily paper over it instead of fixing it.

Once you’ve filtered to the initiatives where AI genuinely is the right lever, you still have more than you have capacity. That’s the real problem, and a value-vs-effort matrix solves it as well as anything does.

The value vs. effort matrix

Value (vertical) — what happens if this works: revenue protected, cost removed, risk reduced, hours given back. Score it in whatever terms the budget holder already uses. Nobody funding this cares that it uses an LLM.

Effort (horizontal) — what it actually costs: does the data already exist somewhere usable, is this a simple classification call or an agent making multi-step decisions, how many systems does it touch, and how much work is it to get people downstream to trust the output.

Four quadrants fall out of that:

Quick wins — high value, low effort. Ship these first, always. Auto-drafting a summary a human still checks, or tagging and routing that used to be manual. Low glamour, real return, and a fast enough feedback loop that people learn what “good” looks like early.

Big bets — high value, high effort. Worth doing, but they need real discipline: a staged rollout, a rollback plan, and a sponsor who understands it won’t ship in one sprint.

Fill-ins — low value, low effort. Fine to do when there’s slack, actively wrong to prioritize over anything above. Also the easiest demos to build, which is why they’re overrepresented in every “look what AI can do” deck.

Thankless — low value, high effort. Kill these in the planning conversation, not six months into the build. The giveaway is a pitch that leads with the technology — “we could use an agent for this” — instead of the outcome.

Where this actually earns its keep

Drawing four boxes takes five minutes. The hard part is getting a room of technical and business people to agree on where an initiative actually sits before anyone’s built anything.