6 min read
You don't need AI (maybe)
Before spending time and money on artificial intelligence, there are five questions to ask yourself. We use them with everyone who writes to us. If the answers are "no", that's absolutely fine.
In short
- You start from the problem, never from the technology.
- Five questions tell you whether AI makes sense now, later or never.
- A "no" isn't a defeat: it's a step forward.
We build software with artificial intelligence. And that's exactly why we can say it calmly: sometimes you don't need AI.
We're not saying it to be contrarian. We're saying it because we often see AI projects that started from the wrong end: not from the problem, but from the technology. "We have to add AI, everyone's using it." And then you end up with an assistant on the website nobody uses, or a subscription left open in a browser tab for three months.
The problem isn't AI. It's the order in which you do things.
- 1Does the problem fit in one sentence?
- 2Does it happen often, always the same way?
- 3Is the data in order?
- 4Can a mistake be handled?
- 5Are simple solutions not enough?
- Let's talk about AI
When someone writes to us, before proposing anything we ask ourselves five questions. Here they are.
1. What's the problem, in one sentence?
It sounds trivial. It isn't.
"I want to use AI in my business" isn't a problem: it's a wish, a legitimate one, but it says nothing about what isn't working.
"We lose two hours a day retyping orders from emails into the management system": that's a problem. It's clear who lives with it, what it costs and how you'd know it's solved.
Try writing it in a single sentence, without the word "AI". If you can, you're already well on your way. If you can't, the first step is understanding what isn't working: often a chat with the person who does that job every day is enough to discover the real problem is different from what you thought.
2. Does it happen often, and always the same way?
AI works best on repetitive work: the same customer questions, the same documents to read, the same steps. A hundred times a day, more or less identical.
If the problem comes up once a month and is different every time, AI risks being a sledgehammer to crack a nut. Maybe a checklist on the wall or a calendar reminder is enough.
A rule we use a lot: if you can explain to a new colleague how it's done in five minutes, and they always do it the same way, it's a good candidate for automation. If explaining it takes half a day and lots of "it depends", maybe not.
3. Is the data there, and is it in order?
To work, AI needs information. It doesn't invent it (or rather: when it does, there's trouble).
Where is the information needed for that job? If it's in a management system or a well-kept spreadsheet, perfect. If it's scattered across five email inboxes, three notebooks and the owner's memory, you need to put things in order before thinking about AI.
And here's something that always makes us smile: when a company puts its data in order, half the problem often solves itself. The real enemy wasn't the lack of artificial intelligence, but the fact that nobody could ever find anything.
4. What happens if it gets it wrong?
AI makes mistakes. Like people, but differently: sometimes it's absolutely sure of something wrong, and says it in such a convinced tone that you fall for it.
So ask yourself: if it gets it wrong, how much does the mistake cost?
If it drafts an email you then reread, a mistake costs thirty seconds. If it sends a quote with the wrong price on its own, it can cost you dearly, in money and embarrassment.
That doesn't mean nothing can be done. It means the more a mistake costs, the more you need a person in the middle checking before things go out. And it has to be designed that way from the start, not bolted on later.
5. Would a simple solution already work?
It's the question that loses us the most jobs. And we're happy about it.
Sometimes the problem is solved by a well-made form that collects the right information straight away instead of chasing it by email. By a rule in the management system. By a shared spreadsheet instead of ten scattered files.
They don't make headlines. But they work, everyone understands them and they don't need any AI.
If the simple solution works, it's the right one. Always.
What to do with the answers
If you have a clear problem that repeats, with data in order, manageable mistakes and simple solutions that are no longer enough, artificial intelligence can make a big difference. And it's worth talking about seriously.
If the answers are more "no" than "yes", it isn't a defeat. There's something else to fix first: clarifying the problem, putting the data in order, deciding who checks what. And it's still a step forward.
In practice, for your business
- Before deciding, have a chat with the person who does that job every day.
- Try explaining it to a new colleague: if five minutes is enough, it's a good candidate for automation.
- Where a mistake is costly, put a person in the middle from the start, not later.
- If an answer is "no", fix that first: it's still a step forward.
Our detour
When someone writes to us, we start exactly here. Even when they arrive convinced they want an assistant, a chatbot or "something with AI".
Sometimes, in the end, the answer is "no AI for now". We say it without hesitation. We'd rather lose a job than sell you a useless one: a client you told the truth to comes back, one you sold smoke to doesn't.
If you'd like to go through this reasoning together, tell us about the problem in a couple of lines. No buzzwords, no need to know anything about technology.
Got more questions? Read the answers to the most common questions