A lead inbox for car dealerships

The context

Car buyers no longer start at the showroom door. They message it. The first contact is a WhatsApp on a Saturday night asking whether a used Civic is in stock, and the second is the same question on Instagram from someone who never gets a reply because nobody was watching that inbox.

Most of those messages are not buyers. They are people comparing, browsing, or asking about a car that sold last month. A salesperson who answers all of them personally spends the day typing, and the two people who were actually ready to visit wait until Monday.

What I was solving

1. 01Enquiries arrive on WhatsApp, Messenger and Instagram, and each app is a separate tab that somebody has to remember to open.
2. 02They arrive in Roman Urdu as often as English, frequently in the same sentence, and a reply in the wrong register reads like a machine.
3. 03A generic auto-reply is worse than none. Answering "we will get back to you" to someone asking a price is the point at which they message the next dealership.
4. 04Every conversation looks equally urgent in a list. Nothing tells a salesperson which two out of forty are ready to come in.
5. 05WhatsApp closes the free reply window twenty-four hours after the customer’s last message. Miss that and the follow-up simply will not send.
6. 06A dealership with more than one showroom needs to know which branch a lead belongs to, and who owns it.

How I built it

1. 01Put all three channels into one thread view, so a conversation is a conversation regardless of which app carried it, with the channel shown as a badge rather than as a separate inbox.
2. 02Gave the assistant the live stock list instead of a script, so it answers with the car that exists: the 2022 Civic RS Turbo, 38,000 km, Rallye Red, automatic, at its actual price, and says so when nothing matches.
3. 03Had it reply in whichever of English or Roman Urdu the buyer wrote in, keeping the words dealership staff actually use rather than translating them.
4. 04Scored every conversation from zero to one hundred on how close it is to a visit, and defined qualified explicitly: condition chosen, brand chosen, model chosen, variant chosen, visit requested.
5. 05Built the lead record around what a salesperson needs before they pick up the phone: brand, model, variant, matched stock, payment preference, status, stage, owner, branch and appointment, with anything unanswered left blank rather than guessed.
6. 06Tracked the WhatsApp twenty-four hour window per customer, and switched the composer to approved templates once it closed, so a follow-up that cannot send is never silently swallowed.
7. 07Left the humans in charge: assign, close, pause the assistant, ask it to draft a reply, or type one yourself, clearly marked as sent by you rather than by the assistant.

The system

01 / 03

02 / 03

03 / 03

DealerShipOS on demo data: the dashboard, the conversation inbox with a lead record open, and the analytics view. The names, numbers and stock are sample data, not a real dealership’s pipeline.

The parts that only show up in production

Two details in the screenshots are the ones that decide whether software like this survives contact with a real showroom. The first is the twenty-four hour window. WhatsApp lets a business reply freely for a day after the customer’s last message, then only through a template it has approved in advance. A system that ignores this looks fine in testing and quietly stops delivering follow-ups a week later. Here the composer says the window has closed and points at the follow-up steps that can still reach the customer.

The second is the failed message sitting in the middle of Fatima’s thread, with the access token error printed next to it in red. A send can fail for reasons that have nothing to do with the dealership, and hiding that behind a grey tick means a salesperson believes they answered someone they did not. It is shown where it happened, in the thread, at the message that did not arrive.

The analytics view carries the same idea. Every figure counts leads created inside the selected window, and the page says so, because measuring new leads in one period against sales closed in another produces a conversion rate that looks precise and means nothing.

What this kind of build is for

A chatbot answers questions. This answers them from stock, in the buyer’s language, then decides whether a human should take over and tells that human what has already been agreed. The difference is not the replies, it is everything attached to them.

The names, numbers and vehicles in the screenshots are demo data for a sample dealership. They show how the system works, not a result it produced.