Our pilot restaurant · the proof behind everything we sell

Step inside our restaurant.

This is Wok & Karahi — our family's halal Chinese-Indian-Pakistani kitchen in Spring, TX. It had the same problems yours does: calls dying in the rush, three delivery tablets beeping, invisible to ChatGPT, and a 30% commission bill every month. This page walks you through exactly what we found, exactly how we fixed it, and what each fix was made of — so you can see your own restaurant in every step.

Wok & Karahi · Spring, TX · 4.6★ across 872 Google reviews · ~8 min walk-through

Abbas and his father Zoeb at their restaurant, Wok & Karahi in Spring, TX
Abbas & his father, Zoeb — the family behind the restaurant, and behind AZ.
2× saleswe took over a declining restaurant and doubled its sales in 18 months
~400 → 872Google reviews in about a year — now 4.6★, new ones running five-star
Top 10 Halalnamed by You Had Me At Halal, 2024 · 100% Zabihah-certified
Live AI linean AI phone agent answers the real line and writes orders into the POS — call it yourself

Sound familiar?

The week this restaurant used to have.

None of this made us a bad restaurant — the food was winning awards. It's just what happens to every independent when nobody has time to step back. See how many you recognize:

7:40 PM, Friday. The phone rings. Nobody can answer.

The kitchen's slammed, the counter has a line, and the caller — who wanted a $60 family order — gets voicemail. Most callers never call back; they just order from whoever picks up.

Cost: orders lost in the busiest hours, invisibly

Three tablets, all beeping, all night.

DoorDash, Uber Eats, and Grubhub each on their own tablet. Staff accepted each order, then re-typed it into the POS by hand. In a rush, orders came out late, wrong — or vanished.

Cost: lost & wrong orders · staff hours every shift

Invisible to the newest front door.

Ask ChatGPT "best halal food near Spring, TX" and we weren't in the answer — our website literally blocked AI crawlers (an HTTP 403), so no assistant could read our menu, hours, or address. Worse: AI thought we were two different restaurants.

Cost: absent from where discovery is moving fastest

The month-end commission bill.

The delivery apps took ~30% of every order they touched — and kept the customer's name, number, and habits for themselves. Direct orders ran just ~8.8% of our mix while 20%+ sat on the 30%-fee apps.

Cost: thousands per year, plus the customer relationship

What we did — and what each fix was made of

Watch the shape of each fix. That's the real method.

For every problem we asked the same questions: does a good existing tool already solve this? Can we connect what we already have? Only where nothing off the shelf was good enough did we build it ourselves — and where the highest-ROI fix wasn't software at all, we designed the moment. Each fix below is tagged with what it was made of.

Connected · an existing tool, set up right

Three tablets became one POS.

This didn't need custom software — it needed someone who knew which of the 100+ integration tools actually works, and how to wire it into Clover properly. Now every delivery order flows straight into the POS and prints to the kitchen. No re-typing, no misses.

Live for over a year. The entire manual step — and the lost, late, wrong orders that came with it — is gone.

3 → 1

tablets on the counter → one POS, one kitchen printer

Loved it? Scan me ⭐

QR

Brought to the table only when a guest is clearly happy — at the exact moment they'd say yes.

Something wrong? Call me — Abbas

The owner's personal card, handed to an unhappy guest before they go home and post about it.

Behavioral · a $2 card and the right words

Reviews doubled — without touching a line of code.

The highest-ROI fix in this whole story cost about $2. Two cards, one rule: happy guests get a "scan me" review card at the peak of their happiness; unhappy guests get the owner's personal card — so the complaint comes to us, not to Google.

~400 → 872 Google reviews in about a year, now 4.6★ with new ones running five-star. For an independent, research puts each star at roughly 5–9% of revenue. (Verifiable live on Google right now.)

Behavioral · the apps deliver our message for us

The delivery apps' own customers, converted to direct.

Every third-party order leaves our kitchen with a thank-you note tuned to that order type: "Paying up to 50% extra?? Order on our website — save big." Because the apps mark our prices up and stack fees, while direct means cheaper prices and $7.99 delivery within 5 miles.

Think about the mechanics: DoorDash's own drivers hand-deliver our customer-acquisition channel, order after order, at zero marketing cost. The opportunity it works against: ~8.8% direct today versus 20%+ sitting on the 30% apps.

Thank you! 🙏

Paying up to 50% extra??
Order direct at wokandkarahitexas.com — cheaper prices, real deals, $7.99 delivery within 5 miles.

QR
Built · nothing off the shelf was good enough

A website AI assistants can actually read.

We rebuilt the site in code as an AEO-native site: full Restaurant/Menu/FAQ structured data, every one of the 78 dishes statically rendered with its real POS item ID, local landing pages, and an llms.txt. The 403-to-bots problem is gone — and we consolidated the "two restaurants" identity confusion into one clean entity that every AI describes correctly.

So when someone asks ChatGPT or Gemini where to eat halal near Spring, we're in the answer. (Check it: wokandkarahitexas.com.)

403 → readable

blocked to every AI crawler → fully machine-readable, one consistent identity

+63%

higher ticket when a guest orders from both cuisines — now actively steered

Built · found in our own sales data

Our data found money we didn't know we had.

We analyzed ~1,500 of our own POS orders and found that guests who order from both cuisines — Chinese and Indian/Pakistani — spend a 63% higher ticket. Nothing was steering anyone there. Now the menu, the website, and the phone agent all nudge toward the pairing.

That's what "analytics" means here: not a dashboard for its own sake, but your own numbers finding your next dollar.

Built · our deepest piece of engineering

An AI employee on the phone — live, right now.

Nothing we could buy would say our dish names right, speak our guests' languages, or write a clean order into Clover. So we built it: a locked, human-sounding voice (English + Urdu/Hindi) that answers every call, takes the whole order, sizes catering from a headcount, remembers regulars, and handles an upset caller with genuine care.

Under the hood — because this is where demos die and real systems live: 28 integrated tools, the full 78-dish menu grounded in real POS IDs so it can never invent a price or a dish, a 15-persona adversarial test gate and 206 automated checks, and gated order-writing so it can never place a bad order. And it never falsely confirms — if something fails, it says so.

Don't take our word for it. That's the live line — order something, ask what's in the Crispy Beef, ask about catering for 40. You're talking to the exact system we build for client restaurants.

Call it now — (903) 602-4012
Live call · 8:52 PM
Thanks for calling Wok & Karahi — good to hear from you again, Priya! The usual butter chicken, extra naan?
Yes please — and is the karahi spicy tonight?
I can have the kitchen make it mild for you — perfect chilly night for it. Want a mango lassi to cool it down?

Knows the regular · speaks their language · knows the menu · writes the order into the POS — the same, on every call.

Built · the site that improves itself

8. A website that runs its own growth loop.

Most restaurant sites are built once and rot. Ours watches its own data — what people search, what AI assistants rank us for, what the POS says actually sells — and proposes its own improvements: new local pages, article drafts, sharper calls-to-action aimed at the goals that matter (calls, catering forms, direct orders). We approve by replying to an email; it ships itself, revertible; live A/B tests decide what stays.

Running on a schedule today, for ~$10–35/month all-in.

Incoming call · (832) ···-··41
NamePriya S. — returning caller
The usualButter chicken · extra naan
PrefersMild · pickup · last visit 12 days ago
“Good to hear from you again, Priya! The usual?” ✓
Built · a guest memory we own

9. The restaurant remembers its regulars now.

The apps kept our customers' names and habits for themselves — only 32 identifiable repeat customers in 1,501 orders. Now the phone recognizes a returning caller and offers "the usual," and a receipt-scan spin-the-wheel turns an app customer into a known, direct one on the spot. Loyalty in the first ten seconds of the second call.

Built · found by the data, not a hunch

10. The data found a catering business hiding inside the restaurant.

43 orders over $150 in 90 days — feast demand nothing was tracking, at margins far above dine-in. So we built it a door: the phone agent sizes catering from a headcount, a portion calculator turns people into exact trays, occasion pages catch the searches, and real inquiries route to a human.

“We're feeding about 40 people — what do we need?”
Chicken Chow Mein2 full trays
Chicken Karahi1 full + 1 half tray
Naan50 pieces
Quote routed to a human ✓
the owner's morning — the real reports the system sends
Built · command & control from the inbox

11. The whole system reports to the owner — and asks permission.

A daily sales report before the kitchen opens. Website improvements proposed by email and shipped on an APPROVE reply. An alert the moment something fails — heard from the system, never from an angry customer. A weekly digest of everything the phone captured. Nothing that matters happens without a yes — which is why the restaurant runs mostly without us standing in it.

How deep it actually goes

This isn't a bag of tricks. It's one system that keeps improving.

Every piece above shares the same data and reinforces the others — the phone agent recommends what the sales data says actually sells; the website ranks for the dishes people actually order. And it doesn't go stale:

A website that improves itself — with your approvalEvery week, an engine watches our numbers and what people ask AI assistants, proposes specific site changes by email, and automatically ships only the ones the owner approves. Running on a schedule at our own restaurant today.
Nothing irreversible without a humanOrder-writing started in dry-run and went live only after real verified transactions. Every risky action is gated and double-checked. That discipline is the difference between a demo and something on your real phone line.
You own everythingYour customers, your data, your website, your phone number. No lock-in, no vendor commissions, month to month. If we ever part ways, everything stays yours.

The honest scoreboard

What's proven, and what's still being measured.

Proven, verifiable today:

  • 4.6★ across 872 Google reviews — roughly double a year ago, running five-star. (Check Google now.)
  • Sales doubled in 18 months after we took the restaurant over.
  • Tablet chaos eliminated — 1+ year live, zero manual re-entry.
  • 403-to-bots → fully AI-readable, one consolidated identity. (Check wokandkarahitexas.com.)
  • The AI phone agent is live — call (903) 602-4012.
  • +63% ticket on dual-cuisine orders, found in ~1,500 real POS orders — now actively steered.

Still maturing (we won't claim what we can't show yet): the rebuilt site and phone agent went live recently, so clean before/after revenue windows on those two are still accumulating. The commission-recovery shift (~8.8% direct → the 20%+ on apps) is the open opportunity being worked with the thank-you-note engine. When we publish those numbers, they'll be real ones — that's the same standard we'll hold for your restaurant.

The takeaway: if these leaks were real at our own award-winning, 4.6★ restaurant, they're almost certainly worse at yours. The difference now: you don't have to be the experiment. Problem → what exists → customize → connect → build → measure. That's what we bring to your restaurant — including the problems we've never seen before.

Good questions

What owners ask about the pilot.

Is this a real restaurant or a demo?

It's a real, operating restaurant: Wok & Karahi, a halal Chinese-Indian-Pakistani kitchen in Spring, TX, owned by AZ's founder Abbas Zoeb — an ex-Google software engineer. 4.6★ across 872 Google reviews; Top 10 Halal 2024 (You Had Me At Halal). Every method AZ sells was proven here first — and when your restaurant needs something we haven't built yet, we build it the same way we built these.

Can I actually talk to the AI phone agent?

Yes — it's live. Call (903) 602-4012, place an order, ask about a dish or catering. You're talking to the exact system we build and tune for client restaurants.

What does "the site was 403 to bots" mean, and why does it matter?

The old site returned an HTTP 403 error to crawlers, so ChatGPT, Gemini, Perplexity and Claude couldn't read the menu, hours or address — invisible in the fastest-growing discovery channel. The rebuilt, code-built site is now fully machine-readable, with full schema and a statically-rendered menu.

Why does ordering direct instead of through the apps matter so much?

The apps take ~30% per order plus fees, and they own the customer. Direct ordering costs a fraction. At our place, direct runs ~8.8% of the mix against 20%+ on the 30%-fee apps — move even part of that to direct and the difference drops straight to the bottom line.

Can you prove the same results at my restaurant?

We prove it in your real numbers — sales tracked month over month straight from your POS and the apps, not a spreadsheet. And with strategy work you only start paying once the savings show up. We built and validated every play on our own restaurant so you get an enterprise-grade setup without being the experiment.

Your restaurant is the next walkthrough. 2 of 5 founding spots are taken — founding restaurants lock the lowest rate for life, and only pay once they're saving.

Book your free call
Book your free call