Our first flagship engagement was a 32-year-old family auto parts and repair shop in rural Thailand. It had zero digital records. Not bad records — none.
Daily sales went into a paper notebook as totals only. The notebook could say ฿2,400 came in today, but never why. Customer debts lived in a second notebook. Prices for thousands of parts lived in the head of the daughter who ran the counter.
Nobody knew the real revenue. The family's best guess was ฿130,000 a month. In 32 years, the business had never seen a profit-and-loss statement.
Ninety days later, it ran on a system we call Auto Shop OS. This is the honest version of how we got there — including the two places it nearly failed.
Discovery in a business with no data
You can't audit records that don't exist. So Mint, our Engagement Manager, started with people instead.
She ran structured interviews, in Thai, with all four family members: the founder, the daughter who runs sales, the son who runs repairs, and the household. She mapped how money, parts, and credit actually moved.
Three findings shaped everything that followed.
First, the constraints were absolute. One shared iPad. No software experience on the team. Everything in Thai. Any system that ignored these would be abandoned within weeks.
Second, the revenue guess could be tested. We pulled 12 months of bank statements and found about ฿500K of QR-payment income — one hard number in a sea of estimates. Cross-referencing it against daily transaction counts gave us an evidence-based revenue range of ฿1.43M–2.0M a year. The guess became a defensible figure before we wrote a line of code.
Third, the product catalog already existed — trapped in Shopee. The shop had years of supplier purchase history sitting in its online shopping account. I wrote a browser script that exported 3,515 order lines in about two minutes. Typing them by hand would have taken weeks. That export became the seed of a catalog that now covers 3,300+ parts.
The build-vs-buy call
We didn't start by building. We started by trying to buy.
We tested a popular Thai POS product, about ฿550 a month, and evaluated it seriously. It's a good product. It broke on this shop's details: its recommended barcode printer could not work with the shop's iPad, and its workflows didn't fit multi-day repairs or informal customer credit.
Free tools failed differently. Google Sheets plus forms works for a small catalog, but gets fragile past 2,000 SKUs: no real record links, painful on mobile, slow formulas.
So we split the decision. Airtable became the data core — a real relational database that still looks like a spreadsheet and works on phones. We built custom only where off-the-shelf failed: label printing, inventory counting, the daily cash close-out, and the AI assistant.
The principle: buy where standard tools fit, build where the business's real shape demands it.
What nearly failed, part one: the interface
Version one of the interface was structured forms — one form per transaction type. Accurate, complete, and wrong.
A busy counter doesn't have time for forms. The most common sale in this shop is a ฿50–200 cash item, and the customer is standing there while you log it. Too many screens, and the paper habit comes back. Once a backlog builds, the system dies.
We rebuilt around one rule: capture must happen in real time, at the counter, in seconds.
The interface went through three generations. Forms first. Then custom mobile apps for the high-friction jobs — counting stock, printing labels, closing out the cash drawer. Finally, a single Thai-language AI assistant became the front door: one chat that handles sales, purchases, repairs, expenses, and stock counts, backed by 22 specialized tools writing clean data into a 14-table backend.
Staff type a normal Thai message — "sold 2 spark plugs, 180 baht, cash" — the same motion as messaging a friend on LINE. Logging a sale takes about ten seconds, while the customer is still counting change.
What nearly failed, part two: the cost
Making the AI affordable was its own engineering problem. In early testing, the assistant cost $4–6 a day to run. At that price, the economics of the whole system fall apart for a small shop.
We set a hard target of $1 a day and worked down to it: prompt caching, trimmed system prompts, consolidated tools, routing simple lookups to cheaper models, and a daily cost cap that stops spending before it runs away.
Today the assistant runs the shop's full daily transaction volume for about a dollar a day. How we did that in detail is a future post.
What adoption actually took
Software that works is not the same as software that gets used. Mint designed the rollout around each person, not around the system.
She prepared a separate "what and why" pitch for each family member, sized to what we were asking of them. The daughter: about 20 minutes of logging a day, in exchange for data that justifies her salary. The son: about 5 minutes a day, in exchange for proof his repair labor was underpriced. The founder: 10 seconds per cash-drawer entry — on paper, the way he's always worked, because we built a form AI reads from a photo rather than asking him to change.
Training happened on-site, in Thai, one workflow at a time, with printed Thai cheat sheets left behind for every task. After go-live, weekly 15-minute calls caught friction early, and we changed the system to fit the people — not the other way around.
Within 30 days of the first training session, every staff member was using the system daily.
Where it landed
The system went live in April 2026. Since then, 100% of daily sales have been captured digitally — more than 90 consecutive days without a gap, and no backlog, because logging happens at the counter in seconds.
The catalog holds 3,300+ parts that previously lived in one person's memory. Total staff time on the system is about 30 minutes a day across three people.
And within the first month, the family sat down with the first profit-and-loss statement in the shop's 32-year history. Real revenue, real costs, real answers — after three decades of guessing.
That system is now Auto Shop OS, which we offer to parts and repair shops across Thailand.