Most AI projects don't fail with an error message. They fail quietly.
The software works. The demo goes well. Everyone is polite about it.
Then, around week three, there's one busy afternoon. The counter is full. Someone reaches for the old notebook — "just for today." Tomorrow they reach for it again.
Nobody announces the decision. But the project is over.
I'm the Engagement Manager at Mintari. My job starts where the build ends: getting real people to actually use the thing, every day, until it's a habit. This is what that took at our flagship client — and why the first 30 days decide everything.
The people I was asking to change
The client is a 32-year-old family auto parts and repair shop in rural Thailand.
Three people run it daily. The daughter runs the counter and holds thousands of prices in her memory. The son runs repairs. The founder controls the cash drawer — with a pen, the way he has for 32 years.
None of them had ever used business software. They share one iPad. Everything happens in Thai.
I didn't treat any of that as a problem to fix. I treated it as the requirements. A system that ignored those facts would be abandoned within weeks, no matter how good the software was.
Nobody changes for a demo
Before training anyone, I sat down with each family member separately.
Not one generic pitch for everyone. A separate "what and why" for each person — sized to what we were actually asking of them.
For the daughter: about 20 minutes of logging a day. In exchange, data that shows exactly what she sells — the evidence behind a fair salary.
For the son: about 5 minutes a day on his repair jobs. In exchange, proof of something he already suspected — that his labor was underpriced.
For the founder: 10 seconds per cash-drawer entry. On paper. We didn't ask him to touch a screen at all. He keeps writing by hand, someone photographs the page, and the AI reads his handwriting. We changed the system to fit him, not the other way around.
That's the first lesson. People don't adopt software because it's impressive. They adopt it when they can see what's in it for them, in their own words.
Training that respects how people actually learn
I trained on-site, in Thai, one workflow at a time.
Not a two-hour session covering everything. One task, practiced until it felt normal, before the next one.
And I left printed Thai cheat sheets behind for every task. Not because the team couldn't remember — because at a busy counter, nobody should have to.
The software helped by asking almost nothing. Recording a sale means typing one chat message in Thai, the same way you'd message a friend. It takes about ten seconds.
That speed is not a nice-to-have. It's the whole game. If recording a sale takes longer than the sale itself, a backlog builds. Once there's a backlog, the notebook comes back. And the notebook never leaves twice.
The first 30 days
Going live is not the finish line. It's the most dangerous month of the project.
So we didn't disappear after launch. Every week, I ran a 15-minute check-in call. Just 15 minutes: what was annoying this week, what didn't fit, what took too long.
When something didn't fit how the team actually worked, we changed the software. Not the people.
Small things helped, too. The daily cash close-out shows a streak counter — one more clean day, then another. A small thing. But habits are built from small things.
Within 30 days of the first training session, every staff member was using the system daily. The shop has recorded every sale since April, without a gap.
Staff who had never touched business software now run the whole day through it. Not because they became "computer people." Because the system met them where they were.
Why we don't hand over the keys and leave
Here's what I believe after doing this: adoption is not an event. It's the first stretch of an ongoing job.
Businesses change. New products arrive, prices move, a new person joins the counter. Every change is a small chance for the old habits to come back.
That's why we stay on as the manager — watching how the system is used, fixing friction while it's still small, and adjusting the AI's job as the business grows. The projects that die after 30 days are usually the ones where nobody was assigned to keep them alive.
If you're curious what that looks like in practice, our How We Work page walks through the whole process, from the first call to the part where your team actually uses the thing.