Running restaurant customer support on WhatsApp properly
Restaurant complaints arrive exactly when nobody can answer them. A ticket queue with priorities means a missing-item complaint at 9 pm is visible at 11 pm rather than lost, and the routine questions about timings and menus never reach a person at all.
- Peak complaint time is peak service time, which is the whole problem.
- Separate the urgent (order wrong now) from the routine (timings, menu).
- A complaint answered the same evening rarely becomes a public review.
- Tag complaints by cause; the pattern usually points at one dish or one rider.
Why restaurant support fails at exactly the wrong moment
The shape of the problem is unusual. For most businesses, enquiries arrive during working hours and can be answered during working hours. For a restaurant, the busiest hour for messages is the busiest hour for cooking.
At 9 pm on a Saturday, somebody messages that their order is missing a dish. The phone is in the kitchen, the manager is expediting, and nobody looks at it until eleven. By then the customer has eaten a partial meal, waited two hours and written a review.
The same window also carries the easy questions: are you open, do you deliver to this area, is the paneer dish vegetarian. Those do not need a person at all, yet they sit in the same queue as the complaint that does.
Separating the two is most of the solution. The routine questions should be answered instantly by a bot; the complaint should be visible, prioritised and impossible to lose.
There is also a second peak nobody plans for: the hour after closing. Customers who ate at nine write at eleven, when the kitchen is shutting and the manager is cashing up. Those messages set the tone of the review written the next morning.
What the bot should handle alone
Roughly two-thirds of restaurant messages need no judgement whatsoever.
- Opening hours, including whether you are open right now.
- Delivery areas and rough timings.
- Menu questions: what is in a dish, whether it is vegetarian or contains nuts.
- Table availability and how to book.
- Directions and parking.
All of these come from your own material. A knowledge base with the menu, the allergen notes, the delivery pin codes and the hours lets an AI agent answer them accurately at any hour, in Hindi or English as the customer writes.
Allergen questions deserve care. Put the exact wording you would want a member of staff to use into the knowledge base, and instruct the agent to hand over to a person for anything it cannot answer from that text precisely.
Keep the menu in the knowledge base rather than as a PDF link. A customer asking whether a dish is spicy wants a sentence, not a download, and a bot that can only offer a file is barely better than a menu board.
What must reach a person, and fast
Three categories, and they should be high priority the moment they arrive.
Order problems in progress. Missing items, wrong order, food that arrived cold. These have a short window in which they can be fixed well.
Any mention of illness. A message suggesting somebody felt unwell after eating needs a manager immediately, both for the customer and for you.
Refund and compensation requests. A bot should never decide these.
With the Support Desk switched on in Convo360, each of these becomes a ticket with a priority and a reply deadline counted in your service hours, escalating to the manager's phone if nobody has answered. The complaint at 9 pm is still visible at 11 pm with a red marker, which is the difference between an apology and a review.
Setting deadlines that fit a restaurant
Standard office SLA thinking does not fit here. Your working hours are not ten to seven, and the urgency of a complaint decays much faster than in most businesses.
| Type | First response | Why |
|---|---|---|
| Order problem in progress | 15 minutes | Still fixable |
| Illness report | Immediate, escalate | Manager decision |
| Refund request | 2 hours | Needs a decision, not speed |
| Feedback or praise | Next day | Worth answering, not urgent |
Set your working hours to your actual service hours, including late evenings, so a 10 pm complaint is measured against a clock that is running rather than one that starts at ten the next morning.
Be realistic about who is on duty. If nobody can respond between seven and ten on a Saturday, set the target accordingly and make sure the escalation reaches a phone that somebody is actually holding. A deadline measured against nobody is decoration.
The recovery message that prevents a review
Most bad restaurant reviews are not about the food. They are about nobody responding to a problem with the food.
A short, specific, same-evening reply changes the outcome more reliably than any compensation. Acknowledge the exact item, say what you are doing, and make it right without arguing about it.
Save it as a canned reply so whoever picks it up at 11 pm does not have to compose it while tired. Keep it short, keep the customer's name in it, and always edit one line so it does not read as pasted. Never use a saved reply unedited for a complaint, because a prewritten apology is obvious.
Tag complaints and read the pattern monthly
The support queue is the cheapest quality-control system a restaurant has, and almost nobody uses it that way.
Tag every complaint with its cause: missing item, cold food, late delivery, wrong order, quality. Five tags is enough. At the end of the month, count them.
The pattern is usually specific and actionable. One dish generating most of the quality complaints. Late deliveries concentrated on one aggregator or one rider. Missing items clustered on large orders, which points at packing rather than cooking.
None of that is visible from reading complaints one at a time during service. It becomes obvious in five minutes once a month when they are grouped, and fixing the underlying cause removes the complaints permanently rather than handling them faster.
Share the monthly tag counts with the kitchen, not just with whoever runs the phone. Complaint data that stays in support changes nothing; the same five numbers read out at a Monday briefing change what gets plated.
Example conversation
Industry guide: AI agents for Restaurants & QSR — Reservations, orders & feedback on autopilot.
Stop losing complaints in the rush
Start free in minutes — no code, no credit card. Bring your own AI keys and pay no per-message markup.
Frequently asked questions
Should the bot handle refunds automatically?
No. Refunds and compensation need a person, both because the decision varies and because a customer who is already unhappy reacts badly to an automated ruling on their complaint.
What about orders placed through Swiggy or Zomato?
Those complaints usually arrive through the aggregator, but many customers message the restaurant directly as well. Treat them as tickets the same way, and tag them by platform so you can see which channel generates more problems.
How do we handle messages after closing?
Let the bot answer routine questions at any hour, and set service hours so complaint deadlines start when you next open. A holding message that says when you will reply is better than silence.
Is this worth it for a single outlet?
If you take more than about twenty messages a day, yes, mainly for the complaint visibility. A single outlet feels the cost of a lost complaint more than a chain does, because the review carries more weight.
Can the bot take orders as well?
It can, through a flow ending in a payment link, though most restaurants do better fixing support first. An order bot on top of an unanswered complaint queue makes the underlying problem worse.
Related reading
Available across India — see WhatsApp & AI calling automation by state and city.