"AI" has become the default answer to every product question, which is a problem, because AI is expensive, unpredictable and occasionally wrong. Ordinary software is cheap, boring and exactly right every time. Most ideas need a bit of the first and a lot of the second.
The short version
AI earns its place when the input is messy and the answer is a judgment. Ordinary software wins when the input is structured and the answer follows a rule.
| AI is the right tool | Ordinary software is the right tool |
|---|---|
| Understanding what a customer's email is asking for | Sending a confirmation when an order is placed |
| Pulling the amount and date out of a photographed receipt | Adding the amounts up |
| Writing a first draft of a reply | Storing, searching and showing replies |
| Grouping a thousand reviews by what people complain about | Counting the stars |
| Deciding which of these looks unusual and should be reviewed | Blocking a payment over the limit |
Look at the right-hand column. That is most of any product: accounts, forms, lists, search, notifications, payments, permissions. It is where most of the work and the reliability live, and none of it needs AI.
The three questions
1. Could a careful person write down the rule?
If you can write the rule in a few sentences, and it holds every time, you do not need AI. "Flag invoices over 5,000 euros" is a rule. "Flag invoices that look suspicious" is judgment.
Be honest at this step. Many things that feel like judgment are actually a rule nobody has bothered to write down. Try writing it. If you get to twenty conditions and they still do not cover the cases, it is judgment.
2. Does the input arrive messy?
Free text, photographs, scans, voice, PDFs from thirty different senders, anything a human typed in a hurry: messy. Dropdowns, checkboxes, numbers from another system: structured.
AI's real superpower is turning the first into the second. If your input already arrives structured, the expensive part of AI is doing nothing for you.
3. Is a wrong answer survivable?
AI is right most of the time, not all of the time, and it cannot tell you which case you are in. So ask what happens when it is wrong.
- Survivable: a suggested tag, a draft reply, a first grouping, a search ranking. A person sees it before it matters.
- Not survivable: transferring money, filing taxes, cancelling an order, sending something to a customer unreviewed, anything a regulator would ask about.
If the answer is "not survivable", you have two options: keep a person in the approval path, or use a rule instead. Both are fine. Pretending the model will be right is not.
What AI costs you that ordinary software does not
- Money per use. Ordinary software costs the same whether ten people or ten thousand use it. Every AI call costs a fraction of a cent to a few cents, forever. At scale it becomes the biggest line in your budget. See what it costs to build and run an app with AI.
- Unpredictability. The same question can get slightly different answers. Anything you promise customers must survive that.
- Review work. Someone has to check the output, at least at the start. That is a person's time, and it is a real cost.
- Privacy questions. Sending customer data to an AI provider is a decision you have to be able to explain, and in Europe, to document.
- Waiting. A database answers in milliseconds. A model takes seconds. That changes how the product feels.
Three ideas, taken apart
"An app that sorts my customer emails"
Understanding what each email is about: AI, clearly. Routing it to the right person, tracking whether it was answered, alerting after two days: ordinary software. Drafting the reply: AI, with a person pressing send. Roughly one AI feature and five ordinary ones.
"An app that books appointments for my salon"
Calendars, availability, reminders, payments, cancellations: all rules, all ordinary software, and all well served by products that already exist. The only plausible AI part is understanding a booking request typed into a chat window, which is a convenience, not the product. Buy something off the shelf and spend the money on marketing.
"An app that tells me which suppliers are overcharging me"
Reading invoices from different suppliers in different formats: AI. Comparing prices, spotting increases, producing the report: arithmetic, exactly the thing computers have been good at since 1950. One AI feature at the front door, ordinary software behind it.
What to do with the answer
If the answer is "no AI", that is good news: your product is cheaper, faster to build and more reliable than you thought, and an existing tool may already do most of it. If the answer is "AI in one place", you now know which place, which is exactly what you need before you brief anyone.
The free check on this site does this analysis for your specific description: which kinds of AI fit, how well, what the risks are, and what to do first. It takes a minute and it does not need your email address. If you are still shaping the idea, start with how to validate an app idea instead, and come back here once you know who it is for.