A friend who runs a kedai makan in PJ messaged me at 9pm. He wanted to know, in his words, "kenapa weekday slow, weekend okay, tapi cashflow tetap pening." He had three months of receipts in a POS export sitting on his laptop. Just an Excel file. Untouched.
Four hours later, he had a dashboard open on his phone.
This is what those four hours looked like.
Hour 1: Just talk
No tools. No code. Just a 20-minute WhatsApp call.
What I wanted to know:
- What decision are you trying to make this week?
- What number do you check every night, if any?
- Which 3 items do you wish you understood better?
His answer was honest: he didn't check any number every night. He flipped through receipts at the end of each shift, did rough math in his head, and decided next week's order on Saturday night feel. The 3 items he was curious about: nasi lemak (was it actually the bestseller, or did it just feel that way?), iced milo (margin felt low), and the new beef rendang (was it worth the prep effort?).
Three concrete questions. That's what the dashboard needed to answer. Everything else was noise.
Hour 2: Clean the data
The POS export had 4,800 rows. About a quarter of them were duplicates, refunds, or fat-fingered entries. "Nasi lemak ayam" appeared as 6 different spellings. "Milo ais" and "iced milo" were the same item in two languages.
I opened the file in Python, used pandas to:
- Standardise item names with a simple lookup table
- Drop refunds and zero-RM rows
- Flag suspicious entries (10x normal price, negative quantities)
The result: a clean Google Sheet with 3,700 rows, each item tagged to one of 12 categories.
Hour 3: Build the dashboard
This was the fastest hour. With the Sheet ready, Looker Studio is mostly drag-and-drop.
What I built:
- Today view: total sales, top 5 items, hourly trend
- This week view: revenue vs last week, top movers, bottom 5 items by margin
- Item explorer: pick any item, see its last 90 days
I picked Looker Studio because:
- It's free
- It works on phones natively
- The owner can share a link with his wife, and that's that
Hour 4: Walkthrough
This is the hour most people skip. Don't.
I recorded a 4-minute Loom showing exactly how to read the three views, what to look at first, and what to ignore on a normal day. I sent the link with a short WhatsApp message: "Tengok dulu, kalau ada apa-apa tak clear, just balas."
He watched it twice. Asked one follow-up question. Three days later he messaged: "Bro, beef rendang tu memang tak worth it. Cut already."
What this actually teaches
The dashboard wasn't impressive. It was 3 pages with maybe 15 charts total. The real value was the 20-minute call at the start, which made sure the dashboard answered questions he actually had.
A lot of "data projects" fail because they answer questions nobody was asking. Build the smallest thing that answers the actual question. Ship it. Refine after.
If you have data sitting in a folder somewhere doing nothing, that's usually the starting point. The hard part isn't the dashboard. It's figuring out which three questions are worth answering.
If you have a POS export or Shopee data lying around and want to see what's inside it, drop me a message. Free 20-minute chat, no commitment.