Personalizing the Plate: How We Reimagined Digital Ordering with AI

Let's go back to 2018.

Digital ordering was everywhere. QR codes. Kiosks. Table-side payments. Every vendor promised speed. But honestly? Most of it felt exactly the same.

I was the product owner at a tiny 16-person startup. We had big dreams and even bigger caffeine habits. We knew we couldn't just digitize the menu. We had to make it feel human again.

The Real Problem: Decision Fatigue

We hit the ground running. We visited outlets, watched diners, and bugged the frontline staff with endless questions.

And we kept seeing the same thing: indecision.

Has that ever happened to you? You pull up a 50-item menu and just freeze. Diners kept falling back on the safe stuff — fried rice, Coke, chicken wings. Safe, but boring.

Meanwhile, the high-margin seasonal dishes? The hidden gems? Nobody even noticed them.

That tiny friction became our obsession.

The Breakthrough: Help, Don't Sell

We didn't want to replace choice. We just wanted to support it.

What if we could bring a Netflix-style recommendation engine to food?

Bold idea. But we had the transaction data. We just had to make it useful.

The "Oh Crap" Moment (I learned this the hard way)

Here's where I messed up.

I thought we could just throw all the data into a model and magic would happen. We had a dish on the menu called "Grandma's Special." No ingredients. No tags. No context. The AI looked at it and basically shrugged.

It recommended nothing useful.

I learned the hard way that metadata isn't an IT chore — it's the whole ballgame.

So we started small. We mapped transaction logs. What did people order together? Did weather or time of day change their choices? We reshaped the menu layout to make the top slots smarter. The result? Faster decisions and higher tickets.

The Cold Start Chaos

Then came the classic startup headache: what about new users? New dishes? New outlets?

We layered in logic like a safety net:

• For new users: we leaned on outlet-level trends and time-of-day signals.
• For new dishes: we used NLP to tag ingredients and map them to similar known items.
• For new outlets: we leaned on broader cuisine patterns and device behavior.

And when there was literally zero data? We defaulted to good old category-level common sense.

The Invisible Superpower

We knew not everyone would log in. So we treated anonymous clickstreams like gold.

What did they tap on? How long did they hover? What did they add and remove from their cart?

No names. No emails. Just enough behavioral breadcrumbs to shape the next screen. Personalization without the creepiness. Smart, respectful, and fast.

For our logged-in loyalty members, we took it further. We showed match scores next to every dish — "87% match based on your past orders."

That little nudge gave people the confidence to try something new. And their average spend went up.

Trust Wasn't a Checkbox — It Was the Core

We worked across Singapore, Australia, and the EU. So privacy wasn't just a compliance headache — it was foundational.

We had one clear rule: earn trust before you ask for data.

Guest users stayed fully anonymous. Loyalty members stayed in control — they could see what we knew, opt out, or delete everything with a single click. All our training data was aggregated and encrypted. No creepy profiles. No shady tracking.

Because in hospitality, trust is everything. Once you lose it, no amount of AI wins it back.

The Quiet Impact

What started as a side experiment slowly became the engine of the platform.

Restaurants noticed it first. Bills crept up. Lines moved faster. Diners stopped staring blankly at the screen.

We weren't shouting about AI or flashing tech for the sake of it. We were just quietly nudging people in the right direction. Making discovery feel easy. Even fun.

We started in Singapore. Scaled to Malaysia, Australia, Indonesia, and the Philippines. Each country brought new flavors and quirks. But the core behavior was universal: people don't want more options. They want better ones.

Looking Back

This project changed how I think about building products.

Personalization isn't about showing off your algorithms. It's about empathy. It's about spotting the small frictions. It's about being helpful without being loud.

We didn't build this because AI was trendy. We built it because people needed help deciding what to eat.

Once we saw that clearly? Everything else just followed.