How to build a WhatsApp chatbot for your cloud kitchen

Start from what customers actually send. Customers ask where their order is, whether an item is available today, whether you deliver to their area, and why the order they got differs from the one they placed.

Then look at what goes wrong today:

• Zomato/Swiggy commissions eat your margin.
• Customers order once, platforms own them. Zomato knows your customers. You don't. They can recommend competitors on your next order.
• Repeat orders leak to the aggregator. A customer who ordered direct last week orders through Zomato this week, because nothing reminded them you exist.

Each of these is a conversation with a predictable shape — a question, a lookup, a confirmation. That shape is what a chatbot handles well.

1. Menu — categories as a list message: Biryanis, Starters, Breads, Desserts, Combos.
2. Items and customisation — spice level, portion, add-ons as buttons.
3. Address — saved for regulars, checked against the delivery area.
4. Total and time — with delivery charge.
5. Pay — link or cash on delivery.

Notice what the business side does: it answers with the specific detail (time, reference, next step), it asks for one thing at a time, and it gives the customer a one-word way to reply. Those three habits matter more than how clever the bot is.

Most direct customers order the same few things. "Usual" or a tap on [Order again] on their last delivery message shows the last order to confirm, with the address and a time. Two taps from hungry to ordered is what makes direct ordering easier than the aggregator app.

Reputoo supports both, and most cloud kitchens end up using both:

• Keyword flows — a fixed menu or trigger word ("MENU", "BOOK", "STATUS") leads down a set path. Predictable, easy to audit, right for anything with a fixed answer.
• AI chatbot — reads the customer's own words and answers from your knowledge base (website, price list, policies). Right for the long tail of questions nobody wrote a menu item for.

A sensible start for a cloud kitchen: keyword flows for the top two or three tasks, AI for everything else, and a person behind both.

The bot reads the order status and answers directly — preparing, picked up, rider's name, estimated time. If the order is late, it says so and alerts staff. It never guesses a time it does not have.

When a dish runs out, the bot offers a replacement or removal with buttons before the kitchen starts, and adjusts the total. Marking items unavailable on the kitchen screen stops the bot from selling them.

• Complaints, refunds and anything where the customer is upset.
• Anything involving money the bot cannot see — a disputed bill, a custom quote.
• Health, safety or legal questions.
• Any customer who asks for a human. Do not make them ask twice.

For cloud kitchens specifically: wrong or missing items, refunds and food-safety complaints need a person quickly — delivery complaints escalate fast.

Handover on Reputoo moves the chat into the shared inbox with the conversation so far, so staff never ask the customer to repeat themselves.

1. Connect your WhatsApp number to Reputoo (official API; takes minutes).
2. Add your website and documents to the knowledge base so the AI answers in your cloud kitchen's own terms.
3. Build keyword flows for the top tasks above. What to connect: aggregator dashboards plus a kitchen display or POS; direct WhatsApp orders should land in the same queue as everything else.
4. Choose who gets handed-over chats, and during which hours.
5. Test from a personal phone, then share the number or a QR code with customers.

Wrong, missing or cold food is reported with the order and a photo; the bot apologises and hands to staff immediately. It does not decide refunds.

The bot knows opening hours and last orders, and during peak rush can give longer, honest delivery times — or pause orders if the kitchen is overwhelmed.

Orders arrive in many phrasings and languages: "2 chicken biryani ek raita". Keyword flows cover dish names; the AI chatbot handles the rest in the customer's language.

• Order with customisation; check the kitchen ticket.
• Reorder the usual.
• Order from outside the delivery area.
• Mark an item unavailable mid-order.
• Report a missing item with a photo.

Watch direct orders per day, reorders, and "where is my food?" questions answered without staff. Read handovers weekly for dish names the bot misses.

Subscribers can check this week's menu, skip a day, pause for travel or change the delivery address — each in a tap — and renew by link before the plan ends. The kitchen sees the day's count updated automatically.

Offices order lunch for teams and parties order platters. The bot asks the date, time, headcount and dishes, checks lead time, and hands to the manager to confirm pricing. Regular office orders can be reordered in one tap.

"Leave at the gate", "call when downstairs", "ring bell 3" — the bot saves delivery instructions with the address and prints them on the rider's ticket. Fewer calls, fewer cold meals.

The bot sends a payment link at order or offers cash on delivery, confirms payment, and sends a receipt. Payment problems go to staff.

Complaints, severe allergy questions, bulk orders and payment problems go to staff with the order and the customer's words. The bot tells the customer when to expect a reply — quickly, because food complaints cannot wait.

A short message after delivery — "Enjoy your meal! How was it?" with [Great] [Something was wrong] — catches problems while they can be fixed. "Something was wrong" goes straight to staff with the order. Keep it to one question and never tie it to an offer.

Some customers live close enough to collect. The bot offers pickup with a ready time, skips the delivery charge, and sends "ready for pickup" with directions to the pickup window — useful for kitchens in buildings with no storefront.

The bot suggests one add-on that fits — a drink with a biryani, a dessert with a family meal — once per order. Customers who decline are not asked again.

If a customer messages one brand's number asking for another brand's dish, the bot can tell them it is from the same kitchen and share the other brand's link or number, rather than losing the order.

The bot answers "is your kitchen FSSAI licensed?" and "do you cook veg and non-veg separately?" from what you have written, factually. Anything it cannot answer goes to the manager.

The bot follows the kitchen's schedule: it opens ordering at opening time, stops at last orders, and pauses if the kitchen is overwhelmed, telling customers when ordering will resume. A bot that keeps taking orders the kitchen cannot cook creates the worst customer experience of all.

At 8 pm on a Saturday, delivery takes longer. The bot quotes times from the kitchen's current queue, not a fixed number, so the promise matches reality.

Direct customers collect a stamp with each order and can type "stamps" to see their balance. Rewards apply automatically at checkout per your rules — a reason to keep ordering direct rather than through an aggregator.

When a dish is added, removed or repriced, update it once and the bot's menu lists change with it. Stale menus lead to orders the kitchen cannot fulfil.

List only the changes the kitchen can make reliably — spice level, no onion, extra gravy — as buttons, with a short free-text note for anything else. Notes the kitchen cannot honour should be declined politely before the order is placed, not discovered by the customer when the food arrives.

New customers often give incomplete addresses. The bot asks for a landmark or a location pin when an address looks incomplete, and saves it for next time. Fewer lost riders means hotter food and fewer calls.

When one person orders for a group — a team lunch, a family dinner — the bot can send tracking to that person and, if they ask, share a status link others can check. Nobody has to keep calling the kitchen to ask when lunch arrives.

Office customers can pre-order lunch in the morning for a set delivery time. The bot takes pre-orders until a cut-off, and the kitchen plans quantities from the list before the rush begins.

Once a week, read the handovers and the orders the bot could not place. Most fixes are small — a dish name customers spell differently, a customisation the kitchen can actually do, a delivery area that needs expanding — and each one makes the next week's orders smoother.