Hospitality · D2C retail

Bombay Sweet Shop's Claude agent captured 1,200+ sales leads in its first two months

Boondi is a WhatsApp concierge for Bombay Sweet Shop, built by KnackLabs on Claude. It answers 250 to 500+ customer questions a day in Hindi, English, and Hinglish, quotes live prices and stock, logs every sales lead, and hands off to a person with a written summary.

WhatsApp Voice Hindi · English · Hinglish Built on Claude
B
Boondi
online · Bombay Sweet Shop
Bhaiya, where's my Diwali order? 🪔
Aapka order #BSS-4821 dispatch ho chuka hai, out for delivery, ETA aaj shaam 6 baje. 📦
Can I send 200 boxes for corporate gifting?
Bilkul! I've noted your bulk gifting enquiry, our team will reach out today with assorted options. 🎁
→ LEAD CAPTURED TO CRM
Client
Bombay Sweet Shop
Hunger Inc Hospitality
The agent
Boondi
support & sales
Channels
WhatsApp + Voice
one agent
Built on
Claude API
Sonnet 5 + Haiku 4.5 on WhatsApp
Status
In production
WhatsApp and voice live
The challenge

Every WhatsApp message was answered by hand, and festival rushes broke that.

Bombay Sweet Shop, a Hunger Inc Hospitality brand, sells mithai across India: everyday D2C orders, retail enquiries, and corporate and wedding gifting. Most customers reach it on WhatsApp, and demand spikes hard around Diwali, Raksha Bandhan, and wedding season.

Customers want answers now: What's in stock? Is it eggless? Where's my order? Can I send 200 boxes for Diwali?, over WhatsApp and over the phone, in whatever mix of Hindi and English they speak.

Before Boondi, the support team answered every message by hand. During festival rushes they fell behind. Replies were slow, chats went unanswered, and bulk and corporate orders got lost, because leads were logged by hand and often missed.

By hand
every WhatsApp reply, before Boondi
Missed
bulk and corporate leads in the rush
How Boondi handles a conversation

One agent. Two channels. Three jobs.

A customer messages Boondi on WhatsApp in whatever language they speak. Claude reads the intent, and Boondi answers, sells, or hands off.

01 · Support

It looks up the real answer

"Where's my order? Is the Mithai Box of 9 in stock?" Boondi looks up the live order, price, and stock, and Claude writes the reply.

→ reads orders, prices, and stock
02 · Sales

It captures the lead while it's warm

"I want 200 boxes for Diwali gifting" doesn't get a generic reply, it becomes a captured lead the sales team can pick up immediately.

→ logs the lead in the CRM automatically
03 · Hand-off

It knows when to bring in a human

When Boondi can't resolve it, or the customer asks for a person, the thread moves to the team with a written summary, so no one starts from zero.

→ hand-off with a written summary
live WhatsApp conversation
Boondi answering a Bombay Sweet Shop customer on WhatsApp — nearest store lookup and pincode serviceability

FIG.01 · Boondi answering a customer on WhatsApp

agent inbox · human hand-off
Boondi read-only agent inbox — a conversation with per-turn latency, LLM cost, and the human hand-off tool call

FIG.02 · the agent console: per-turn latency, the Claude model, and the hand-off call (customer details blurred)

Why it works

It's wired into the real business, not a scripted FAQ bot.

Built for the spike.
Handles the everyday long tail year-round and absorbs the festival surge, without a proportional scramble to hire and train.
Speaks like the customer.
On WhatsApp, Claude replies in Hindi, English, and Hinglish, the way Indian customers actually message.
Sells, not just supports.
A gifting enquiry becomes a captured lead in the CRM, not a generic reply.
The hand-off is seamless.
When a person should take over, they do, in the same thread, with the whole conversation already there.
1,200+
sales leads captured automatically in the first two months
~3.5 s
reply time, even during festival rushes
250-500+
customer questions answered a day
$0.02
cost per conversation
The results

More leads, faster replies, and far less manual work for the team.

Measured on the live WhatsApp deployment in its first two months of production.

Before
With Boondi
Sales leads
Logged by hand, often missed
1,200+ captured automatically in two months
Reply time
Slow during festival rushes
About 3.5 seconds
Support team
Answered every WhatsApp message by hand
Much less dependent on manual replies. People step in when Boondi hands off
In their words

"KnackLabs understood our business, not just our technology. They built Boondi to speak like our customers, access real orders and capture gifting leads—even during the festival rush. They owned the outcome, not just the build. A rare team that feels like an extension of your business."

Hunger Inc Hospitality
Nisha Bhatia
Technical Project Manager, Hunger Inc Hospitality
Under the hood

How Boondi is wired

On WhatsApp, Claude does the reasoning, through the Claude API with Claude Sonnet 5 and Claude Haiku 4.5. Boondi runs on Gantry, our agent runtime, which handles intent routing, memory across channels, security, and audit. Its connections to Bombay Sweet Shop's stack are reusable modules. Voice runs through Ukti, our voice runtime, on a separate model stack.

Channels in
WhatsApp
via Interakt · Claude
Inbound calls
Ukti voice · telephony
→
Boondion Gantry
route
intent detection
memory
across channels
lang
Hi · En · Hinglish
audit
security + logs
→
Business systems
Shopify
live order look-ups
CRM
sales leads captured
Human inbox
clean hand-off
See it in action

Every conversation and lead, on one console

The read-only inbox: sessions, per-turn latency and cost, and every lead Boondi captures.

customer conversations · read-only console
Boondi operator console — customer conversations, catalogue and pincode settings, and auto-captured leads
This is for you if…

…you field support and sales over WhatsApp and phone, and your volume spikes seasonally.

You want an always-on agent that sounds like you, and knows when to bring in a human. We don't advise on AI, we build the agent and run it with you.