Case study · AI automation

An AI sales assistant that answers motorbike buyers on Facebook, in seconds

Three motorbike showrooms in Bangladesh sell through their Facebook Pages. We built one AI assistant for all of them. It answers buyers' messages and comments in seconds, with the showroom's own prices and instalment plans, and passes every interested buyer to the team to call. These showrooms chose replies in Bangla; the same assistant can just as well reply in English or any other language.

An illustration of the assistant: messages, voice notes, photos and comments flow into a phone chat answered in seconds, then into a new lead, the showroom's leads sheet and a call from the team.
The assistant at a glance. Illustration in English; the live assistant replies in Bangla. Sample data.

At a glance

  • 3 showroomsuse it on their Facebook Pages today
  • 6 daysfrom the first setup to the first showroom going live
  • 105 test chatsscored by an AI judge before changes go live
  • 4.5 stypical reply time on real buyer chats in the first days

The brief

A busy inbox, the same questions, and buyers slipping away

Most motorbike buyers in Bangladesh first reach a showroom through its Facebook Page. They ask the same things again and again: the price, the monthly instalment, the colours, where the showroom is. They write in Bangla, in Banglish or in English, and many send a voice note or a photo instead of typing.

Showroom staff answer between customers on the floor, so replies can come late, and two people can quote two different prices. A buyer who was ready to visit may get no answer and message another dealer. And with no list of who asked, nobody calls them back.

What it had to do

  • Reply in clear Bangla, whatever language the buyer writes in
  • Never make up a price, an offer or an instalment
  • Turn interest into a name and number the team can call
  • Serve several showrooms from one shared catalogue
  • Step aside when someone from the showroom replies

Problems and fixes

What changes for the showroom

Six everyday problems in a showroom's inbox, and what the assistant does about each.

  • The same questions all day

    BeforePrice, instalments, colours, address. The team types the same answers again and again while buyers wait.

    With the assistantIt answers in seconds from the showroom's own price list, and adds tap buttons for the next question.

  • Any language, voice notes and photos

    BeforeBuyers mix languages, or send a voice note or a photo instead of typing, and a simple auto-reply can't follow.

    With the assistantIt understands Bangla, Banglish and English, listens to voice notes and looks at photos. These showrooms chose replies in Bangla; yours can answer in English or any language your customers use.

  • Wrong or made-up numbers

    BeforeA rushed reply or a guessing chatbot can quote a wrong price, and a showroom can't take that back.

    With the assistantPrices come only from the catalogue and instalments from a calculator. Code checks every reply and removes any price that isn't in the list before it's sent.

  • Interested buyers get lost

    BeforeChats scroll away, and nobody keeps a list of who asked about which bike.

    With the assistantWhen a buyer asks about instalments, a test ride or a visit, it asks for a name and number and saves them to the showroom's leads sheet.

  • Comments go unanswered

    BeforeQuestions under posts wait for hours, next to spam and abuse.

    With the assistantIt replies to each question in public, sends the details by private message, and hides spam and abusive comments.

  • The team wants to step in

    BeforeSome chats need a person, like a complaint or a special request.

    With the assistantWhen someone from the showroom replies, the assistant goes quiet for that buyer, so it never talks over the team.

How it works

From a message to a phone call

  1. A buyer writes

    A message, a voice note, a photo or a comment on a post.

  2. The assistant checks

    The showroom's own prices, instalment plans, offers and photos.

  3. It replies in seconds

    In the buyer's language, with photos of the bike and buttons for the next question.

  4. The lead is saved

    The buyer's name, number and interest go into the showroom's sheet.

  5. Your team calls

    Staff call the buyer back and mark the lead as called or sold.

In the chat

A chat on a phone: the buyer sends a voice note, and the assistant replies with starting prices for commuter, executive, scooter and premium bikes, plus two tap buttons.
A voice note, answered with prices
A chat on a phone: the buyer asks about instalments, the assistant replies with the down payment and monthly amounts, asks for a name and number, and the lead is saved.
Instalments, then a saved lead
Comments under a showroom's Facebook post: the assistant answers a price question and sends details by private message, thanks a buyer for a compliment, and hides a spam comment.
Comments answered, spam hidden

Three conversations from our test chats, translated from Bangla into English and shortened, with brand and model names removed. Names and numbers are sample data.

Behind the chat

One catalogue, every showroom

All three showrooms share one catalogue sheet: prices, instalment plans, offers, photos and answers to common questions. Change a price once, and within five minutes every showroom answers with it. Each showroom keeps its own Facebook Page, address and leads sheet.

A diagram: one shared catalogue of prices, EMI plans, offers, bike photos and FAQ feeds the AI assistant, which serves three showrooms, each with its own Page, address and leads sheet, with room for a new one.
One catalogue serves every showroom, synced every five minutes.
A leads sheet listing six buyers with the time, name, masked phone number, bike type, a note and a status of New, Called or Sold.
The leads sheet the team works from. Sample data.

What each showroom gets

  • Answers in seconds, at any hour, in the buyer's language
  • One catalogue to update for every showroom
  • A leads sheet with each buyer's name, number and interest
  • Photos of the bike sent right in the chat
  • An email to us if anything fails, and nightly backups
  • A new showroom set up in about 10 minutes

Beyond motorbikes

Where else this works

The same assistant fits any business whose customers ask the same questions in chat before they buy, in whatever language they write. We shape it around your prices, your words and your team.

  • Online shops on Facebook

    Prices, sizes and stock, cash on delivery and order status, with each order saved for packing.

  • Clinics and diagnostic centres

    Doctors' schedules, test prices and serial bookings, answered after hours too.

  • Real estate

    Flat sizes, prices and locations, with visit requests passed to the sales team.

  • Coaching centres and schools

    Batches, fees and admission dates, with each interested student saved for a call.

  • Travel agencies

    Packages, prices and travel dates, with each enquiry saved for an agent.

  • Restaurants and cloud kitchens

    Menus, prices, delivery areas and opening hours, in your customers' language.

Under the hood

Simple to run, hard to break

  • Built on n8n

    n8n is a tool that connects apps and runs steps for you. Each part of the assistant is a clear workflow we can read, test and change.

  • Facts, not guesses

    The AI only quotes what's in the catalogue. Code checks each reply before it's sent, and instalments come from a calculator, not the AI.

  • Tested on 105 chats

    An AI judge scores the assistant on 105 test conversations for facts, language, tone and format. Over six rounds of fixes, passed checks rose from 82.5% to 94.3%.

  • Runs on its own

    Prices sync from the sheet every five minutes, the Facebook connection is checked automatically, and we get an email if anything fails.

  • Backed up every night

    The data is encrypted and copied off the server every night, so a broken server never means lost leads.

  • Ready for the next showroom

    A new showroom joins the shared catalogue with its own Page and leads sheet in about 10 minutes.

Built with

  • n8n
  • OpenAI
  • Facebook Messenger & Pages
  • Google Sheets
  • PostgreSQL
  • Docker

Want an assistant like this for your business?

Tell us about your project. We typically respond within 24 hours and can start most projects within a week.

Start your project