Before You Buy AI: The Three Questions That Decide Whether It Will Work

Most businesses that come to us asking for AI don't need AI yet.

That isn't a sales tactic — it's the opposite of one. We say it because the projects that fail almost never fail on the technology. They fail because nobody asked three basic questions first. Here they are, in the order that matters.

1. What is the task, described as a verb?

"We want AI" is not a task. "We want to stop manually copying order details from WhatsApp into a spreadsheet" is a task. It has a verb, a start, and an end.

If you cannot describe what you want in one sentence containing an action, the project has no edge — and a project with no edge expands until it collapses. Write the sentence down. If it takes you more than one sentence, you have more than one project.

2. Who does it now, and how long does it take them?

You need a number. Not a feeling — a number.

If Ama spends ninety minutes a day reconciling payments, that is roughly thirty hours a month. Now you know what the problem is worth, and you know whether a solution costing X is sane. You also know how you'll prove it worked: the same measurement, taken again in sixty days.

Businesses that skip this step can never tell whether their AI investment paid off, because they never knew what they were paying to fix. Then the tool quietly falls out of use and everyone concludes "AI didn't work for us."

3. Is the input consistent enough to automate?

This is the one that kills projects, and almost nobody checks it.

Automation needs predictable inputs. If your invoices arrive in eleven different formats, half of them as photographs taken at an angle, no model will save you until that's tightened up. If your customer records live in three systems that disagree about the same person's phone number, an AI layer on top will confidently give you wrong answers faster than before.

Fix the inputs first. Sometimes that fix alone solves the problem and you never need the AI at all — which is a perfectly good outcome, and one we've delivered more than once.

The uncomfortable pattern

Work through those three questions honestly and one of two things happens.

Either you end up with a tight, measurable, genuinely automatable task — in which case the build is straightforward and the return is easy to prove. Or you discover the real problem is a broken process, unclear ownership, or messy data. In which case, fixing that is cheaper, faster, and worth more than any tool you were about to buy.

Both are wins. Only one of them costs a lot of money.

Where we come in

At DeMario Glow we run this diagnosis before we quote — because we'd rather build one system that works than three that get abandoned. Sometimes the answer is a custom AI build. Sometimes it's a two-week process fix and an honest invoice for less than you expected.

If you want to run the three questions against your own operation, send us a message. It's a conversation, not a pitch.

DeMario Glow — Adenta, Accra. Building AI-powered systems for businesses in Ghana and worldwide.

https://demario.pro⁠