A jewellery CRM where the WhatsApp inbox is the pipeline
How Yukti AI built Surati Jewellery Factory a CRM with a two-way WhatsApp and Instagram inbox, an AI that answers the first message grounded in the live catalogue, a lead pipeline with custom stages, quotations and invoicing, and a call-script teleprompter that fills in with each customer's real data.
Surati Jewellery Factory's customers reach out on WhatsApp and Instagram, but those chats lived on a phone, disconnected from the CRM. Yukti AI built a single system where every conversation sits next to that customer's record, lead stage and quotations; an AI auto-reply (grounded in the catalogue, told never to quote exact prices) handles the opening exchange; staff take over any one chat with a ‘Stop Bot’ toggle; and replies staff send from their phone sync back in. Twelve tables, Supabase, role-based staff login for Owner, Manager and Staff.
Quick facts
| Client | Surati Jewellery Factory |
|---|---|
| Location | Surat, Gujarat |
| Industry | Diamond & jewellery — manufacturing & wholesale |
| Core challenge | Customer conversations on WhatsApp / Instagram disconnected from the CRM; enquiries slipping; no shared record of what was said |
| Solution | Custom jewellery CRM with a two-way WhatsApp & Instagram inbox and an AI auto-reply |
| Modules | Inbox, Customers & purchase history, Leads (custom pipeline stages), Orders, Quotations & invoices, Vendors, Inventory, Catalogue, Follow-ups, Call Script, Team Chat, Dashboard |
| AI | Auto-reply on OpenAI (gpt-4o-mini), grounded in the live catalogue, system-prompted to defer on exact prices/stock; per-conversation ‘Stop Bot’ override |
| Existing number stays live on the phone's WhatsApp Business app; messages sync both ways, including staff replies sent from the phone | |
| Roles | Owner, Manager, Staff — row-level security on all 12 tables; management chat restricted to Owner/Manager |
| Stack | Vite + React frontend, Supabase (Postgres, Auth, Edge Functions), OpenAI API |
| Status | Live — real customer messages landing in the CRM inbox, verified end to end |
For a jewellery factory, the enquiry channel is WhatsApp
Surati Jewellery Factory sells to retailers and wholesale buyers. Those buyers do not fill in web forms — they send a WhatsApp message or an Instagram DM: "do you have this design in 18k", "send me your latest bridal sets", "price for 2 carat solitaire studs". The first reply, and how fast it comes, sets the tone for the whole deal.
The problem is that WhatsApp is a personal messaging app, not a business system. The conversation lives on one phone. If the person holding that phone is busy, the buyer waits. There is no record of what was promised. The CRM — customers, leads, quotations — is somewhere else entirely, so the salesperson is copying details back and forth by hand, or not at all.
If the enquiry arrives on WhatsApp, then the CRM has to contain WhatsApp — not link to it, not "integrate" with it in a way that still needs the phone. The conversation, the customer record and the pipeline stage have to be one thing.
What was slipping
- "Enquiries sit unanswered when the phone-holder is busy." First-response time depended entirely on one person's availability.
- "There's no record of what was said." A promise made in a chat two weeks ago is lost unless someone scrolls back to find it.
- "The chat and the CRM are separate." Customer details, lead stage and quotations lived apart from the conversation that generated them.
- "Staff answering from their own phone doesn't show up anywhere." Replies sent from the WhatsApp Business app were invisible to the rest of the team.
- "New staff don't know what to say on a call." The good salespeople had a script in their head; nobody else did.
How the front door changed
Before
- WhatsApp chats on one phone, seen by one person
- First reply waits for that person to be free
- CRM and conversation are separate systems
- Phone-app replies invisible to the team
- No shared record of promises or prices discussed
- Call approach varies by who picks up
After
- Every conversation in the CRM, next to the customer record
- AI answers the first message instantly, grounded in the catalogue
- One system: chat, customer, lead stage, quotation
- Phone-app replies sync back and are logged as staff messages
- Full conversation history against every customer
- A call-script teleprompter filled with the customer's own data
How a message becomes a tracked lead
The modules
The two-way inbox
Every WhatsApp and Instagram conversation in one list, sorted by most recent activity, with a single unread dot that clears the moment the chat is opened. Staff reply from the CRM. Critically, it is genuinely two-way: a reply a staff member types directly in the phone's WhatsApp Business app also syncs into the CRM, saved as a staff-sent message — and the AI knows not to reply again on top of it. A per-conversation "Stop Bot" toggle pauses the AI for one customer while it keeps handling everyone else. Benefit: the whole team sees every conversation, and nobody is tied to one phone.
AI auto-reply, grounded in the catalogue
An OpenAI model handles the opening exchange — greeting, understanding the enquiry, pointing to relevant designs — using the live catalogue as its source of truth. Its system prompt explicitly forbids inventing exact prices or stock levels and tells it to hand specifics to staff. Benefit: buyers get an instant, on-brand first response at any hour, without the risk of a wrong quote.
Customers & purchase history
A full customer record with purchase history, linked to every conversation, lead and quotation. A phone icon on the record opens the call script (below) filled in for that customer, with a tel: link to actually place the call. Benefit: one screen holds everything known about a buyer.
Leads with a custom pipeline
A lead pipeline with the factory's own stages, drag-to-move, quick-add, and follow-up dates. A conversation converts into a lead without leaving the inbox. Benefit: a WhatsApp enquiry becomes a tracked opportunity in one click.
Quotations & invoices
Quotations and invoices generated within the CRM, tied to the customer and the lead. Benefit: the quote references the same customer record the conversation is attached to.
Call-script teleprompter
A master sales-call script with placeholders — name, occasion, budget, reason, recommended pieces — that fill in from the customer's actual data. On a phone, the script stays on screen in the background while the native dialer takes over, so staff can swipe back to it mid-call. Owners and managers edit the master script in a simple editor; staff can only read it. Benefit: every salesperson runs the best call, not just the experienced ones.
Vendors, Inventory, Catalogue, Follow-ups, Team Chat, Dashboard
The rest of the operation — vendor records, inventory, a catalogue, a dedicated follow-ups view, an internal team chat (with a management channel restricted to Owner and Manager), and a dashboard with real charts drawn from live data. Benefit: the inbox is the front door; this is the building behind it.
What the AI does, and what it deliberately doesn't
It does: answer instantly, greet in the brand's voice, understand what the buyer is asking for, surface relevant designs from the live catalogue, and keep the conversation warm until a human picks it up. It reads the catalogue as grounding, so it is talking about real pieces.
It deliberately doesn't: quote exact prices, confirm exact stock, or make commitments. Its prompt is written to defer those to staff. And the moment a staff member wants the conversation, the "Stop Bot" toggle hands it over cleanly — for that one customer only.
Beyond the auto-reply, the same data supports follow-up intelligence (which conversations have gone quiet), source analysis (which channel brings buyers who convert), and plain-language questions about the pipeline.
What runs without anyone pressing a button
New WhatsApp enquiry → tracked lead
Staff replies from the phone → CRM stays in sync
Conversation goes quiet → follow-up surfaces
What each person gets
Owner
Everything, including the management team-chat channel, the dashboard, and the ability to edit the master call script.
Manager
Full operational access, the management channel, and script editing — runs the floor day to day.
Staff
The inbox, customers, leads, quotations and the call script (read-only) — everything needed to work an enquiry, nothing that isn't.
How it was built — and hardened
CRM core on real data
A working prototype connected to Supabase — customers, leads, orders, quotations, vendors, inventory, catalogue and team chat all reading and writing live data.
Staff auth & roles
Owner / Manager / Staff login added, and the original wide-open access policies replaced with row-level security requiring a logged-in account on all 12 tables.
The two-way WhatsApp bridge
Inbound relay and outbound send built as secured server functions, connected through Yukti's own WhatsApp infrastructure so the factory's number stays live on the phone.
AI auto-reply, grounded & guard-railed
OpenAI reply wired to the live catalogue, prompted to defer on specifics, with a per-conversation manual override.
Echo handling & hardening
Taught both sides about message direction so staff replies from the phone sync in without the bot double-replying, and every database call now surfaces its error instead of failing silently.
Verified live
A real customer WhatsApp message was confirmed landing correctly in the CRM inbox, end to end.
What changed
What the system does — no invented numbers.
Technology & data architecture
- Vite + React — the CRM frontend, with a mobile top bar and drawer so staff can work the inbox from a phone.
- Supabase (PostgreSQL, Auth, Edge Functions) — 12 business tables, all with row-level security requiring a logged-in staff account; a trigger auto-creates a staff row on signup with the right role.
- Supabase Edge Functions —
inbound-relay(shared-secret auth) saves incoming messages and calls the AI; a database webhook triggerssend-outboundfor replies. - OpenAI API — the auto-reply, grounded in the live catalogue, with a prompt written to avoid hallucinating prices or stock.
- Yukti's WhatsApp infrastructure — owns the Meta WhatsApp Cloud API connection, so the factory's number stays usable on the phone's WhatsApp Business app.
Security
- Staff login required; roles Owner / Manager / Staff, enforced by row-level security on every table
- The management team-chat channel is restricted to Owner and Manager roles
- The inbound relay authenticates with a shared secret; server functions run without exposing keys to the browser
- Surati Jewellery Factory owns all customer and message data
If your customers message you, this is buildable for you
A two-way WhatsApp inbox inside a CRM, with an AI that answers the first message from real data and a per-conversation manual override, fits any business where enquiries arrive as messages:
- Jewellery brands, retailers and manufacturers (see the Lumera case study)
- Traders and wholesalers taking orders over WhatsApp
- Clinics, salons, real estate and services that book by chat
- Any business running enquiries out of one person's phone
Start with a free workflow audit. See Custom CRM Software, WhatsApp AI Chatbot or Yukti AI CRM.
Questions about a WhatsApp CRM
Explore more
Custom CRM Software
A CRM built around your pipeline, your channels and your product.
WhatsApp AI Chatbot
An AI that answers your WhatsApp enquiries grounded in your real data.
Case study: Lumera Fine Jewellery
A full custom jewellery ERP + CRM — inventory, metal ledger, catalogue and GST billing.
Case study: Arjun
A WhatsApp AI sales agent that handles all inbound and moves leads to prospect with zero manual work.
Jewellery Software
Custom diamond & gold inventory, metal ledger, WhatsApp catalogue, GST billing and CRM.
All case studies
Every Yukti AI build — jewellery, manufacturing, trading and services.
Summary
Yukti AI built a custom jewellery CRM with a two-way WhatsApp and Instagram inbox for Surati Jewellery Factory, a diamond jewellery manufacturer and wholesaler in Surat. Every customer conversation sits inside the CRM next to that customer's record, lead pipeline stage, quotations and purchase history. An AI auto-reply on OpenAI, grounded in the live catalogue and system-prompted never to quote exact prices or stock, handles the opening exchange; a per-conversation "Stop Bot" toggle lets staff take over one chat instantly; and replies staff send from the phone's WhatsApp Business app sync back into the CRM as staff messages without the bot double-replying. The system also includes leads with custom stages, quotations and invoices, vendors, inventory, a catalogue, a follow-ups view, a call-script teleprompter that fills with each customer's real data, an internal team chat and a dashboard. It runs on Vite/React and Supabase with row-level security and three roles — Owner, Manager, Staff — and is live with real customer messages verified end to end. Yukti AI builds comparable WhatsApp CRMs for jewellers, traders and service businesses across India.