Does your product really need AI? Five questions to ask first

Almost every business is being told it needs AI. Competitors are announcing features, investors are asking about it, and customers are starting to expect it. So it's tempting to start with "we should add AI" and work out the reason later.

That's backwards. AI is a tool, and like any tool, it's only worth paying for if it fixes something. We build AI features for a living, and we still think some of them shouldn't be built. Here's how to tell the difference.

1. What problem are you solving, in one sentence?

"Add AI to our product" isn't a problem. "Our support team spends half its day answering the same questions" is.

Try to finish this sentence: "Today, ____ is slow, expensive or frustrating for ____." If you can name the task and the person, you have a real starting point. If you can't, more planning will help you more than more technology.

2. What does that problem cost you today?

Put a rough number on it: hours each week, customers lost, sales missed. It doesn't have to be exact.

This number does two jobs. It tells you how much a fix is worth, so you know what's sensible to spend. It also gives you something to measure afterwards. Without it, you'll have a new feature and no way of knowing whether it paid off.

3. Would something simpler work?

This is the question vendors tend to skip. Many "AI problems" go away with a better search box, a clearer form, an automated report or a few fixed rules. Those options are cheaper to build, easier to fix and behave the same way every time.

If you can write the logic down as clear steps, you probably don't need AI. It earns its place when the work involves messy language or judgement: reading varied documents, understanding free-text questions, or summarising long records.

4. What happens when it's wrong?

Sooner or later, AI will make a mistake, and it can sound confident while doing it. So the real question is what a mistake would cost.

• Low risk: a draft reply that a person reads before sending. A bad draft wastes a few seconds.
• Higher risk: anything that moves money, makes promises to customers, or touches legal, medical or personal matters. Keep a person in charge here, and show them the source of each answer so they can check it.

If nobody can check the output and a single mistake would be serious, don't automate that step yet.

5. Is your information ready, and are you allowed to use it?

AI can only work from what you give it. If your policies are out of date, your records are scattered, or the answers live in one person's head, sort that out first. The AI will repeat the same mess, only faster.

Check the rules too. Before customer data goes to any outside service, you need to know what that service does with it and whether your contracts and privacy obligations allow it. That's a much easier conversation to have before you build.

What a good first project looks like

If you got through all five, look for a project that is:

• Narrow: one task for one group of people.
• Frequent: it happens often enough for small savings to add up.
• Low risk: a person can spot and correct mistakes.
• Measurable: you know the number you're trying to move.

Here's an example from Zen365. SMB owners often can't quickly answer simple questions, like how much they made this month or who still owes them. So its AI assistant answers from the company's own numbers. The problem is clear, the people are clear, and the owner can check the answer against the figures.

Remember that launch isn't the end of the cost, either. AI features have running costs for each use, and someone has to watch their quality and keep them up to date. Include that in your numbers.

What if the answer is no?

That's a good result. You've saved money and found out what to fix first, whether that's a simpler feature, cleaner information, or a clearer picture of the problem. When AI does make sense later, you'll be ready for it, and you'll know exactly what you want it to do.