Case study · Fabrics & furnishings · India
How F&F scaled order processing with an AI agent
F&F supplies furnishing fabrics to dealers across India, and almost every order starts on WhatsApp. Each dealer had their own group with F&F's team, so growth depended on how many groups a few people could keep up with. We built an agent that answers every dealer and places the order itself. Every new dealer is now one more conversation for the agent, not more work for the team.
- The problem
- Sales stuck at what a few people could answer across ~1,000 WhatsApp groups.
- What we built
- An agent on WhatsApp and email that checks live stock and places orders in Business Central.
- The result
- Every dealer answered at once, at any hour. Growth no longer waits on hiring.
- Status
- Live in production. A person still handles every exception.
The same problem shows up in
- Textile and fabric distributors
- Tiles, laminates and building materials
- Wallpaper and flooring
- Auto and industrial spare parts
- Any B2B business whose dealers order on WhatsApp
If every new customer means another WhatsApp group for your team, your sales can only grow as fast as you can hire.
Before and after.
Before
Growth = headcount
One group per dealer, and a small team reading all of them.
- 1A dealer asks for stock in their group
- 2The message waits until someone sees it
- 3That person checks stock and batches in the ERP
- 4They reply, the dealer confirms, the order is typed in
- 5The SO number goes back to the dealer
Each order took a person's time. Twice the dealers meant twice the team.
After
Growth ≠ headcount
Ten more dealers or a thousand, the team's work stays the same.
- 1The dealer messages F&F
- 2The agent answers and places the order
No longer limited by people
- Reading every group
- Checking stock and lots
- Replying to the dealer
- Typing the order in
~1,000
Dealer groups before
Watched by hand by a limited team.
500–1,000
Orders a day
Since go-live on 21 Sep 2026.
0
Added headcount
More dealers, same team.
20
Items in one message
A whole order list answered in one reply.
The problem
Why growth got stuck.
Every dealer was a group to watch
New dealers added work faster than the team could add people.
Busy hours lost orders
When every group needed an answer at once, some dealers waited. A dealer with a customer in front of them asks another supplier.
Answers stopped at closing time
Orders sent in the evening or on Sunday sat until the next working day.
Knowledge sat in a few heads
Which lot matches, which warehouse, which date. Only experienced people could answer, and they couldn't be copied.
Why not just an app?
F&F already had an app. Dealers still ordered on WhatsApp.
So we didn't ask dealers to change how they order. We put the agent where they already were.
- 01
Not every dealer is comfortable with apps
Many dealers run their shop on WhatsApp and a phone call. Logins, menus and forms put them off, and WhatsApp doesn't.
- 02
WhatsApp is already open all day
An app is one more thing to open and remember. WhatsApp is where the dealer already talks to their customers and suppliers.
- 03
Dealers don't order the way forms ask
They type a name the way they say it, a code without the prefix, or send a photo of the label. A form needs the exact code.
- 04
Fabric needs a conversation, not a number
“Is it one piece? Which lot? Can half come from Mumbai?” An app shows a quantity, but it can't talk the dealer through the choices.
- 05
Long lists are painful in an app
A dealer pastes 20 items in one message. In an app, that's 20 items added to a cart one by one.
What they asked for
Grow the number of dealers without growing the team. And never promise stock or a date the ERP doesn't show.
One more: don't just answer, close the order. Give the SO number and the PDF, and don't ask the dealer to confirm the same thing twice.
So the goal was a salesperson who never runs out of time, not a chatbot.
In production
What it does now.
Every dealer, at once
A hundred dealers messaging at the same time all get an answer.
Open after hours
Evening and Sunday orders are placed, not left for Monday.
Live stock, both warehouses
Answers from Business Central at that moment, from Dharuhera or Mumbai by the dealer's zone.
Sold the way fabric is sold
Shows which lots fill the order, so the warehouse ships what was promised.
An order, even with no stock
Gives a date from open purchase orders and lead times, and takes the order instead of losing it.
The team on exceptions only
Credit holds, customised items and changes reach the right person with the full chat.
It never overrides a credit hold, never gives a discount, and never changes an order once it's placed. Those stay with people.
The conversation
A 9:40 pm order, placed by 9:42.
- Fabric A Ecru — 40 MTR available, in one piece.
- Fabric B Blush — 25 MTR available, not in one piece:Lot 1 — 31 MTR1. 18 MTR — Batch 10421 (DH)2. 13 MTR — Batch 10424 (DH)
- Fabric C Ivory — 2 ROLL available.
Please confirm:
- For Fabric B, these batches, or shall we choose?
- Delivery to your usual address?
The dealer sent three items in one message. The agent checked all three against live stock in both warehouses, showed lots only for the item that isn't in one piece, and asked everything it needed in one reply.
One answer from the dealer, and the order is in Business Central, with the GST PDF back in the chat. Nobody at F&F had to be awake.
Trust and data
Safe to scale.
An agent that takes orders at scale makes mistakes at scale too, if it's careless. So it had to be careful before it was fast.
Only what the ERP says
Stock, dates and prices are read live from Business Central. It never guesses.
Only what the dealer saw
An order holds only the lots the dealer was shown.
Each dealer, their own data
A dealer sees only their own orders and invoices.
People own the money decisions
Credit, discounts and changes stay with F&F's team.
The process
How we worked.
- 01First
Their way of selling became the rules
F&F's sales playbook became the agent's rules.
- 02Sandbox
Evaluated on a copy of their ERP
Tested on real dealer scenarios in a copy of Business Central. No real order was ever at risk.
- 03Pilot
A few trusted dealers first
Released to a few trusted dealers first. F&F's team reviewed every chat, and we fixed anything they'd have answered differently.
- 04Ongoing
Watched, and tested before every change
Live chats are monitored. Every change must pass evaluation on 147 real dealer scenarios first.
The hard part
Not worse than their team, ever.
The obvious build: a chatbot that answers questions and hands the dealer a form to fill in.
Why that fails
Dealers already have a team they trust
For years, F&F's team has known their names, their lots and their usual address. A slower or more robotic experience sends them back to the phone, and the agent is never trusted.
What we did instead
F&F's own team became the bar
Every reply had to be one their team would have sent. Where it couldn't be as good, the agent hands over to a person.
| The team did | The agent does |
|---|---|
| Talked in F&F's words: lot, SO number, PI | Uses the same words |
| Placed the order and sent the PDF | Raises the order, with the SO number and GST PDF in the chat |
| Asked only what it needed | Asks every open question once, at the end of one reply |
| Kept messages short | Keeps every message readable on WhatsApp, with no lots the dealer didn't need |
| Passed tricky cases to the right colleague | Sends cases like a credit hold or a custom size to the right person at F&F, with everything the dealer already said. That person picks up from there, and the dealer doesn't have to explain it all again |
Where it is better than before
Speed
An answer in minutes, not when someone gets to the group.
Hours
Evenings and Sundays answered.
Consistency
Every dealer gets the same rules, read from live stock.
Long lists
A 20-item order answered in one reply.
How we held that bar
During the pilot, F&F's team read the conversations and flagged any reply they would not have given. Every evaluation scenario says what a good reply must include, judged by how their team would answer.
Faster is easy. Faster without the dealer noticing a downgrade is the hard part.
Six months from now
How it stays accurate.
Before any change goes live, it's evaluated on 147 real dealer scenarios, three times each.
- 1Does every scenario that passed still pass?
- 2Does every item the dealer asked for get an answer?
- 3Does every exception still reach a person?
- Goes live
An example
The multi-item change made one type of reply worse. The tests caught it before any dealer saw it, and it was fixed before release.
What's next
Order intake was agent one.
F&F's leaders don't want a few AI tools. They want a business that runs itself, with people deciding only what people should decide. Order intake was the first step. We are building the rest with them.
The goal
A truly autonomous business, where every routine step runs on its own and every decision that matters still has a person behind it.
- Live
Order intake
Answering dealers and placing orders
- Next
Supply intelligence
Chasing vendors for dates, and reordering before stock runs out
- Planned
Warehouse operations
Watching dispatch, goods coming in, and stock audits
- Planned
Purchase accounting
Matching vendor bills to goods received, and handling excess or short deliveries
- Planned
Payments
Scheduling vendor and salary payments for approval
- Planned
Inventory lifecycle
Tracking every design from launch to clearance
One agent took the limit off sales. The next ones take it off supply, the warehouse and finance.
Find out where your growth is capped by headcount.
A short discovery session, then a pilot on fifty of your real WhatsApp orders against a copy of your ERP. Compare its answers with your team's before you commit.
Book an AI KickoffOr email hello@avestalabs.ai