Guide
How to know if an AI project is worth it (before you spend)
The AI projects that fail usually weren't killed by the technology — they were bad bets before a line of code was written. The good news: you can tell most winners from losers up front, with a few honest questions and a small, cheap first step. Here's how to check before you commit.
By Miguel Alejandro Hayes· Founder, Hayes Projects
The four questions to ask first
Before spending anything, answer four things. What specific, costly problem does this solve? (If you can't name the cost, that's a warning.) How will you measure success — in hours saved, leads captured, errors removed? Who will actually use it, and will they? And what's the smallest version that would already be worth doing?
A project that survives those four questions has a real target, a real metric, a real user, and a small first step. A project that can't answer them clearly isn't ready to fund yet — not because AI won't work, but because you don't yet know what "working" would mean. Getting that clear is most of the battle.
Good bets vs. bad bets
Good bets share a shape: they attack a frequent, expensive, rule-based problem where success is measurable and the first useful version is small. Answering repeat questions 24/7, qualifying inbound leads, keeping systems in sync — you can put a number on the win and ship something narrow fast.
Bad bets tend to be the opposite: vague goals ("use AI to be more innovative"), no way to measure the result, a giant all-or-nothing scope, or a solution chasing a problem no one actually has. If the value only appears after a massive build, you're betting big on an unproven guess. Shrink it until the first useful piece is small — or don't start.
De-risk it before you commit
The single best way to avoid a bad AI project is to not commit to the whole thing at once. Start with a short, paid validation that defines the exact scope, cost, and success metric — so you're buying a plan, not a hope. Then ship the highest-return piece first and measure it. If the numbers show up, that result funds the rest; if they don't, you've spent little and learned a lot.
Insist on ownership too: whatever gets built should be yours — the code, the data, the access — so a project that works doesn't quietly turn into a dependency you can't leave. Do it this way and an AI project stops being a gamble and becomes a series of small, measured decisions.
Frequently asked
How do I know if an AI project will pay off?
+
Check four things before spending: the specific costly problem it solves, how you'll measure success, who will actually use it, and the smallest version that's already worth doing. A project that answers all four clearly is a good bet; one that can't isn't ready to fund yet.
How can I reduce the risk of an AI project?
+
Don't commit to the whole thing at once. Start with a short paid validation that fixes scope, cost, and the success metric, then ship the highest-return piece first and measure it. Let results fund the next step instead of betting everything up front.
What makes AI projects fail?
+
Usually not the technology — it's vague goals, no success metric, an all-or-nothing scope, or a tool chasing a problem no one has. Most failures are bad bets made before any code is written. Clear targets and a small first step prevent most of them.
About the author
Miguel Alejandro Hayes — Founder, Hayes Projects
Economist and essayist turned developer. He founded Hayes Projects, a Miami venture studio and custom software lab, to build software that ships, scales and solves real problems.
Meet the founder→Thinking about an AI project? Book 20 minutes — no pitch — and we’ll pressure-test whether it’s worth it before you spend.
Book 20 min — no pitch →