All Case StudiesHospitality

The Pho Shop That Stopped Feeding Uber

How a family-run Vietnamese restaurant built its own AI ordering channel and kept the 30% margin Uber Eats was taking.

4 min readUpdated 2026-04-03Based on Claude Sonnet 4 / GPT-4o

The Real Problem

Linh and Thanh run a Vietnamese pho restaurant in Mt Albert. They've been there six years. The broth takes 12 hours. The regulars know their order by heart. It's the kind of place people drive across town for.

More than half their revenue comes from delivery. That sounds healthy -- until you look at where the money actually goes.

Last month, they did $5,000 through Uber Eats. Uber took roughly $1,500 in commissions. That's 30% -- gone before Linh counts a single ingredient cost. And it gets worse: Uber doesn't reduce its commission when it runs promotions. When Uber offers "20% off your next order," the restaurant absorbs the discount on top of the commission. One West Auckland Vietnamese restaurant, Lixiri's Kitchen, told NZ Herald they received just $85.36 from $300 in Uber Eats orders. That's 28 cents on the dollar.

To compensate, most restaurants inflate their Uber Eats prices by $5-10 per dish. Customers notice. They complain in reviews. And the restaurant's reputation takes the hit.

Linh and Thanh know they need their own ordering channel. They tried building a basic website with online ordering. It technically works -- but it's a static menu with no personality, no customisation options, and no reason for customers to choose it over the convenience of Uber Eats. Customers still default to the app they already have on their phone.

Why Existing Tools Don't Solve This

Plenty of NZ Thai and Vietnamese restaurants already run their own ordering -- Narm Thong Thai, Thai Niyom, Thai E-Sarn, Le Xom. This proves the model works. But these are basic web ordering forms: pick a dish, add to cart, checkout. No personalisation, no conversation, no memory of what you ordered last time.

The problem isn't technology -- it's friction. Uber Eats is one tap. A basic website is five minutes of scrolling, selecting add-ons, entering your address, fumbling with payment. For a tired customer at 7pm, Uber wins every time.

Your POS doesn't help here. Lightspeed and Square handle in-store transactions. They don't give you a customer-facing ordering channel that competes with the convenience of a delivery app.

How AI Solves This

Linh and Thanh set up an OpenClaw AI assistant on WhatsApp and their website. It handles the full ordering conversation -- from menu browsing to payment to pickup time.

A regular customer messages:

Customer: "Hey, it's Sarah. The usual please."

AI: "Hi Sarah! Your usual -- large pho tai with extra bean sprouts and no coriander, plus two summer rolls. Pickup from Mt Albert?"

Customer: "Yep! And add a Vietnamese iced coffee this time."

AI: "Done! Your order:

  • Large pho tai (extra bean sprouts, no coriander) -- $16.90
  • Summer rolls x2 -- $12.00
  • Vietnamese iced coffee -- $6.50

Total: $35.40

Ready for pickup in 20 minutes (6:45pm). Pay now via the link below or pay when you collect. See you soon!"

A new customer messages:

Customer: "What's good here? I like spicy noodle soups but I can't eat shellfish."

AI: "Great taste! Here are our spicy noodle options that are shellfish-free:

  • Bun bo Hue (spicy beef noodle soup) -- our most popular spicy option, $17.90
  • Pho with chilli oil -- classic pho with our house-made chilli, $15.90

Note: our tom yum noodle soup does contain shrimp paste, so I'd avoid that one. Want me to add anything to your order?"

The AI knows the full menu with photos, handles customisation (protein choice, spice level, add-ons, dietary restrictions), processes payment via Stripe, and estimates pickup time based on current kitchen load. It remembers returning customers and their preferences.

Why Customers Switch

The AI gives Linh and Thanh something Uber can't: a direct relationship with their customers. The ordering experience is personal -- the AI remembers names, past orders, and preferences. And customers see the real prices, not the inflated Uber prices.

Linh sends a WhatsApp broadcast to her regulars once a week: "Friday special: bun bo Hue with free spring rolls, order through us!" No commission. No middleman.

How We Set This Up

None of this works if the AI is just a standalone chatbot with no connection to your actual business. That's why BestAI builds a custom integration program -- a piece of software that bridges your AI assistant with the systems you already use.

For this kind of setup, that means:

  • Connecting the AI to WhatsApp and your website so customers can order through channels they already use
  • Integrating with Stripe for secure payment processing -- customers pay via a link in the chat
  • Building your full menu into the AI's knowledge base, including photos, customisation options, allergen information, and pricing
  • Setting up a customer memory system so the AI recognises returning customers and recalls their preferences
  • Connecting to your kitchen workflow so pickup time estimates are accurate

Here's our process:

  1. We map your current workflow -- We sit down with you and figure out how orders currently flow, what tools you use, and where the bottlenecks are.
  2. We build the connections -- Our developers write a custom program (an API connector) that lets the AI talk to your payment system, your menu database, and your messaging channels. No manual data entry, no copy-pasting between apps.
  3. We test end-to-end -- Every ordering scenario gets tested with real orders before going live. Nothing launches until it works reliably.
  4. We maintain it -- When your menu changes or you add a seasonal special, we update the system to match.

You don't need to be technical. We handle all the development -- you just tell us how your kitchen runs, and we make the AI fit into that.

The Result

  • 30% margin recovered -- orders through your own channel pay zero commission
  • Regular customers converted -- the AI remembers preferences, making reordering faster than Uber
  • Customisation handled naturally -- spice level, protein choice, add-ons, and dietary needs through conversation
  • 24/7 ordering -- customers can place orders for tomorrow's lunch at midnight
  • Direct marketing channel -- weekly specials go straight to your customer base, no algorithm in between

What AI Can't Do Here

  • AI won't deliver the food -- you still need your own delivery driver or a pickup-only model
  • AI won't replace Uber Eats entirely -- some customers will always prefer the app, and that's fine
  • AI won't manage your kitchen capacity -- you need to set limits on how many orders you accept per time slot
  • AI can't force customers to switch -- you need to actively promote the channel (table cards, social media, WhatsApp broadcasts)

Who This Is For

  • Vietnamese, Thai, and Asian restaurants where delivery is a major revenue channel
  • Restaurants losing 25-30% of delivery revenue to platform commissions
  • Family-run operations that can't afford to give away $1,000+ per month in fees
  • Any restaurant with a loyal regular base who would happily order direct if it were easy enough

Want This for Your Business?

Book a 45-minute workflow review and we'll show you exactly how this applies to your specific situation, no obligation, no fluff.

More Hospitality Case Studies

Replying to Google Reviews After a 14-Hour Shift

How AI drafts personalised review responses that reference specific dishes and feedback — so your Google presence stays active even when you're exhausted.

Read case study

Every Missed Call Is a Lost Booking

How AI answers your restaurant's phone during service — handling reservations, menu questions, and hours enquiries without pulling staff off the floor.

Read case study

Your Instagram Hasn't Been Updated Since Last Month's Special

How AI creates a weekly social media calendar from your daily specials, food photos, and events — keeping your feed alive without hiring a marketing team.

Read case study

One Video, Five Languages, Zero Translators

An AI tool that takes a single promo video and produces localized versions in English, Mandarin, Japanese, Korean, and Hindi with native voiceover and lip-sync.

Read case study

You Don't Know What Each Dish Actually Costs You

How AI turns supplier invoices into real-time food cost tracking — so you know your margins before they disappear.

Read case study

The Same 10 Questions, 50 Times a Week

How AI handles repetitive customer enquiries across WhatsApp, Facebook, and Google — so you stop typing 'Yes, we have gluten-free options' every day.

Read case study

A Professional Menu with AI Food Photography in Under an Hour

An AI tool that takes your menu items and generates a beautifully designed, multilingual menu with photorealistic food images for every dish.

Read case study

The Saturday Morning Catering Call You Never Answered

How AI captures corporate catering enquiries during your busiest rush -- so you stop losing $800 orders to the cafe down the road.

Read case study

The Menu Nobody Could Read

How AI bridges the language gap between Chinese restaurant kitchens and Kiwi customers -- turning 'What's Mapo Tofu?' into a $22 order instead of 'Just sweet and sour pork, thanks.'

Read case study

Thirty Cents of Every Dollar Goes to Uber Eats

How one Auckland Chinese restaurant built its own ordering channel and kept the margin -- without learning to code.

Read case study

Raj and Priya Can't Answer the Phone During Dinner Rush

How AI takes every takeaway and catering call during your busiest hours -- so you stop losing orders while you're cooking.

Read case study

Fifteen Catering Enquiries Before Diwali. Three Replies Sent.

How AI qualifies catering leads, collects requirements, and prepares quotes -- so you never lose a wedding booking because you replied too late.

Read case study

Twenty Seats. Forty People Waiting. No System.

How AI turns a chaotic ramen shop queue into a smooth virtual waitlist -- so customers wait from the cafe next door instead of leaving.

Read case study

Forty Pieces of Salmon Nigiri in the Bin Every Night

How AI predicts exactly how much sushi to prepare each hour -- so you stop throwing profit in the bin at closing time.

Read case study

The Korean BBQ That Nearly Closed Its Dining Room

How AI handles bilingual bookings and guides first-time diners through Korean BBQ -- so you don't need to choose between answering phones and running service.

Read case study

Chimaek Without the Commission

How a Korean fried chicken shop built AI-powered direct ordering and stopped giving 30% of every delivery to Uber Eats.

Read case study

No Peanuts, No Problem

How AI gives every customer instant, accurate allergen information for every dish -- so your staff never have to guess about fish sauce or shrimp paste again.

Read case study

The Friday Night Phone Problem

How AI answers every call at a fish and chips shop during the Friday rush -- so two people can keep frying instead of one running to the phone.

Read case study

How an Independent Pizza Shop Fights Back Against Domino's

How AI phone ordering and smart upselling help a single-location pizza shop compete with chain-level technology -- without chain-level budgets.

Read case study

The Weekend That Books Itself

How AI handles bilingual hotpot reservations, manages the waitlist, and guides first-timers -- so your staff stay on the floor instead of the phone.

Read case study

The AYCE Profit Protector

How AI tracks consumption, predicts prep quantities, and protects margins at an all-you-can-eat BBQ restaurant -- without awkward waste penalties.

Read case study

The Monday Prawn Problem

How AI predicts exactly how much food to prepare for each buffet session -- so you stop throwing away $200 of prawns on quiet nights.

Read case study

The Tapioca Timing Problem

How AI tells your bubble tea shop exactly when to cook the next batch of pearls -- so you stop throwing away rubbery tapioca at closing time.

Read case study

One Photo, Four Platforms, Two Languages

How AI turns a single product photo into bilingual social media content for Instagram, TikTok, WeChat, and Xiaohongshu -- in minutes, not hours.

Read case study