How we decide what AI is worth building.

Every company we talk to has a list.

Sometimes it's a real document — “AI Opportunities,” sitting in somebody's drive. More often it's a loose pile of ideas that surface in meetings: a chatbot for the website, an agent for the inbox, something — anything — with the word copilot in it.

Here's the uncomfortable math: most of that list deserves a NO-GO.

Not because AI doesn't work. It works. The list deserves a NO-GO because most of it was built backwards — it starts with what AI can do and goes looking for a home, instead of starting with the work and asking what actually deserves help.

The demo problem

The most expensive AI project isn't the one that fails fast. It's the one that almost works.

It demos beautifully. Everyone in the room nods. Then it meets a Tuesday. The real data is messier than the demo data. The edge cases show up on day three. The person who championed it gets pulled into a proposal deadline, and six weeks later the thing is a tab nobody opens.

Demos are easy. Tuesdays are hard.

So before we build anything — for a client or for ourselves — we run the idea through three gates. It takes about five minutes per idea. A healthy list loses half its items at this stage, and that's the system working.

Gate 1 — Real work

Does it attach to a workflow someone on your team runs every single week?

Frequency beats wow. An unglamorous workflow that runs fifty times a year will beat a spectacular one that runs twice — every time — because repetition is where the hours hide and where the learning compounds.

The ideas that pass this gate sound boring on purpose: the compliance matrix you build for every RFP. The account research your reps do before every first call. The variance questions finance answers every close. Weekly work, real stakes, known shape.

The ideas that fail it sound exciting: the all-knowing assistant, the automation for a task that happens twice a year, the tool for a workflow that doesn't exist yet. That last one matters most — AI doesn't create discipline, it amplifies it. If nobody runs the workflow today, AI won't summon it into existence.

Gate 2 — Owned

Can your people run it, fix it, and extend it without us?

There's a graveyard in most companies: automations that a vendor or a long-gone employee built, that still technically run, that nobody understands, that everyone is afraid to touch. That's not a capability. That's a dependency with a login.

This is why we build with your team in the room instead of disappearing for six weeks and coming back with a black box. The deliverable isn't just the workflow — it's the person on your payroll who can change it when the business changes. And the business will change.

If an idea can only live as something we maintain forever, it fails this gate — even if it works. Especially if it works, because that's the dependency you'll feel the most.

Gate 3 — Measured

Can we point to hours saved or work won — not vibes?

Pick the number before you build. Hours per RFP response. First calls booked per week. Days to close the books. If you can't name the number in advance, you won't be able to defend the spend later — and “everyone seems to like it” does not survive a budget review.

The number doesn't have to be big. It has to be real, and it has to be visible to someone who signs things.

What a GO actually looks like

Run enough ideas through the gates and a pattern shows up. The survivors are boring, weekly, owned, and measured.

AI earns its keep on the work everyone agrees is necessary and nobody loves. That's the honest pitch for what it is right now: not a replacement for judgment, but a relentless first-pass machine for the work that eats your team's week. The first draft of the response. The first pass at the research. The first read of the numbers. Your people stay where the judgment lives — at the end, deciding.

A NO-GO is a gift

Killing an idea at the scorecard stage costs you five minutes. Killing it after a quarter costs you the quarter — plus the budget, plus something harder to win back: the team's belief that the next idea is worth their effort. Every stalled AI project makes the second one harder to sell internally.

So the discipline isn't generating ideas. Ideas are free. The discipline is killing them while they're still cheap to kill, so the two or three that pass get the focus they need to survive contact with a Tuesday.

That's the whole system. Three gates, five minutes per idea, no hype in either direction.

Score your own list

The fastest way to use the gates is out loud, with someone who has run them a hundred times. Book a call, bring your list — or bring nothing but questions — and we'll run it through the gates with you. You'll likely find your first GO, and a few merciful NO-GOs, before the call is over.

Can't find a time that works, or have a question? Email Chris.

— Chris Winters · Moriva AI · Houston, TX