Most AI prioritization advice starts with a scoring matrix: plot every idea by impact and feasibility, rank the quadrants, pick the winner. That's a reasonable exercise, but it has a quiet flaw — it assumes you already have a clean, well-articulated list of candidate use cases to score in the first place. In practice, most lists are a mix of borrowed ideas from other companies, things a vendor pitched, and a handful of guesses about what "AI could help with." Scoring guesses just produces a ranked list of guesses.
There's a simpler starting point that skips the whiteboard exercise: let the complaints pick the project for you.
Start with what people already complain about
Every team has a short list of tasks that come up in one-on-ones, Slack threads, and end-of-week venting — the thing that's "such a pain," the report that "always takes forever," the process everyone agrees is broken but nobody's fixed. These complaints are a better signal than a brainstorm, because they describe a real, recurring, currently-manual task rather than a hypothetical one. Nobody complains about a task that doesn't actually happen.
Before generating a fresh list of AI ideas, spend a week just collecting these complaints as they come up naturally. You'll likely end up with a shorter, more concrete list than any brainstorming session would produce, and every item on it is already proven to be a real problem — not a theoretical one.
Test each complaint with a simple question
Once you have a handful of recurring complaints, run each one through a single filter: could you write down the steps someone takes to do this task, well enough that a new hire could follow them without asking a single question?
If yes, you're looking at a strong AI candidate. A task with clear, describable steps is usually a task with structure underneath it — structure that a model or workflow can be built around. If the answer is no — if doing the task well depends on judgment calls, tribal knowledge, or "you just kind of know when something looks off" — that task isn't ready for automation yet, no matter how painful it is. It needs to be better understood by humans first, or it needs a human kept firmly in the loop rather than replaced.
This single question does a lot of the work a formal scoring framework tries to do, without requiring anyone to argue about numeric ratings. It separates "painful and well-understood" from "painful and still vague," which is usually the real dividing line between a project that ships and one that stalls.
Favor the complaint that unlocks the next one
Among your well-understood, genuinely painful candidates, give extra weight to the one that would also clean up something structural — a messy spreadsheet that becomes a proper source of truth, a manual handoff between two tools that becomes a real integration, a scattered set of documents that get organized into something searchable. These side effects matter because your first AI project is also laying groundwork for your second and third. A use case that leaves behind cleaner data or a clearer process makes everything after it easier and cheaper to build. A use case that solves its narrow problem and nothing else is fine, but it doesn't compound.
Don't let the loudest complaint win by default
One caution: the most frequently repeated complaint isn't automatically the right first project, especially if it touches something customer-facing or high-stakes. A noisy but low-risk internal annoyance — a weekly report that takes two hours to assemble by hand — can be a better first project than a louder complaint about something customers see directly, simply because the cost of an early mistake is lower while your team is still learning what good AI output looks like and how closely to check it.
The real benefit of starting here
Working from real complaints instead of a brainstormed list does something a scoring matrix can't: it guarantees that whoever benefits from the fix already wants it. You won't need to convince anyone that the problem is real or sell them on why it's worth fixing — they've been telling you for months. That built-in buy-in is worth more than a perfect prioritization score, because it means the people who'll actually use what you build are already rooting for it to work.
If you've got a running list of things your team complains about and aren't sure which one is actually ready for AI, that's a quick and useful conversation to have. Book 30 minutes or reach out through our contact page — no pressure, just a second opinion on where to start.