You've decided to adopt AI. Now you need to find someone to build it.
But before you start talking to vendors and comparing quotes, ask yourself these 5 questions. Not someone else — yourself.
Because if you can't answer them, it doesn't matter how talented the people you hire are. It's not going to go well.
Question 1: Can I describe the problem I want to solve in one sentence?
"We want to adopt AI" is not a problem description.
"Our support team spends 4 hours a day answering the same questions over and over, and I want to cut that in half" — that is.
A problem you can state in one sentence is a problem that can actually be solved. If you can't state it clearly, it usually means you haven't thought it through yet.
Test yourself: Write the problem you want to solve on a sticky note. If it doesn't fit, you're not done thinking.
Question 2: How is this problem being handled right now?
A lot of people skip this step and jump straight to "how could AI solve this?" But if you don't understand your current process, how will you know whether the new approach is actually better?
You need to know:
- Who does this? Which people, which department
- How do they do it? What are the steps, what tools do they use
- How long does it take? Hours per day, week, or month
- How well does it work? Error rate, complaint rate, efficiency
Write this down. This is your baseline. After you build the AI system, you compare against it. That's how you know if anything actually improved.
Improvement without a baseline isn't improvement. It's just a feeling.
Question 3: Where is my data, and what does it look like?
AI needs data. Most people understand this in the abstract but haven't actually prepared for it.
You need to verify:
- Where does the data live? Paper? Excel? A system? LINE group chats?
- Is the format consistent? Does everyone fill things in the same way? Are there gaps?
- Is there enough of it? Hundreds of records? Thousands? Tens of thousands?
- Can you actually get it out? Some systems lock data in — exporting it is a real ordeal.
I had a client who told me "we have a lot of data." When I got there, it was spread across 12 different Excel files, every one formatted differently. Some fields were handwritten Chinese text, some were numbers, some were blank.
Just cleaning the data into something AI could work with took a month.
If your data is dirty, the AI output will be dirty too. Garbage in, garbage out.
Question 4: Who will use this system — and do they actually want it?
This is the question most people ignore.
The person who decides to build an AI system is usually a manager or executive. The people who have to use it every day are the frontline staff.
If those employees don't want to use it, don't know how to use it, or roll their eyes and think "great, another new system" — it doesn't matter how well it's built. It won't get used.
You need to do one thing before you start: talk to the people who will actually use it.
Not to tell them "we're building an AI system." Ask them:
- What's the most frustrating part of your job right now?
- If a tool could help you with something, what would you want it to do?
- What's your biggest fear about a new system?
These answers will directly shape how the system gets designed. And when people feel like their input was heard, their buy-in for the new system is dramatically higher.
The best system isn't the one the boss wants. It's the one the employees are willing to use.
Question 5: Can I live with an imperfect version one?
This one is the hardest, but also the most important.
The first version of any AI system will not be perfect. Accuracy might be 70–80%. Features might be bare-bones. The interface might be rough.
A lot of executives see version one and say: "This isn't what I imagined." And they give up.
But the projects that actually succeed have one thing in common: the person in charge can tolerate imperfection.
Because they understand that version one isn't about "finishing" — it's about "starting." Version one lets you collect real usage data. It shows you what needs to change. It's how you get to version five.
It's like opening a restaurant. The food on day one is never as good as it'll be on day 100. But if you close down because day one wasn't perfect, you'll never see day 100.
Ask yourself honestly: if version one only hits 70%, will your reaction be "great, now we know what the other 30% needs" — or will it be "what a waste of money"?
If it's the second one, you might not be ready yet.
Quick Checklist: The 5 Questions
Fill this out before you talk to anyone:
| # | Question | Your Answer |
|---|---|---|
| 1 | What problem am I solving? (One sentence) | |
| 2 | How is this handled now? How long does it take? | |
| 3 | Where is my data? Is it consistent? Is there enough? | |
| 4 | Who will use this? Do they know? What do they think? | |
| 5 | Can I accept an imperfect version one? |
All 5 answered? You're ready. Go find someone to build it.
Fewer than 3? Don't rush. Take a week or two to work through these. It'll save you months of painful backtracking later.
One Last Thing
When you're evaluating vendors, watch for one signal in particular:
If they give you a quote before asking you any of the questions above — be careful.
A good partner will take time to understand your problem before proposing a solution. Like a good doctor who asks where it hurts before writing a prescription.
Diagnose first. Treat second.
I'm Young. I've led over a dozen digital transformation projects across industries — strategy, architecture, development, AI adoption, all of it. My job isn't to sell you technology. It's to help you figure out what you actually need, then build it with you. If you've got a fuzzy idea you want to sharpen, let's talk for an hour.