Why AI matters for a jewellery business, and where it doesn't

I come from a family that has run a wholesale jewellery business for decades. I grew up around it and saw the problems firsthand. So when people ask me about AI for jewellery business owners, I don't start with the technology. I start with how the work actually moves through a shop, because that is where AI either earns its place or it doesn't.

This is me thinking out loud about both sides: where I think AI really helps, and where I think it should stay out.

How does a jewellery business actually run day to day?

Most of the jewellery businesses I know don't run on software. They run on conversations.

An order is confirmed on WhatsApp. A delivery date is promised on a call. A design arrives as a photo. A salesperson sends a voice note to say the piece is with the karigar. The owner keeps the important things in a khata or an order book, and a lot of the rest in their head.

Everyone keeps registers: orders, repairs, karigar work, samples. In the shops I have sat in, nobody keeps them fully up to date. Not because people are careless. The work is happening in the chats, and the register is a second job that comes after.

Then there is the shared phone. Many shops run one WhatsApp number that several people answer. In one demo, an owner's first question to me was simply, "What happens if three people use this WhatsApp?" That one question says a lot about how the trade works.

And a customer's question is rarely one message. They ask for today's rate. Then whether the same design comes in 22K and 18K. Then whether their old chain can go towards it. Each of those is a quote, a follow-up and sometimes a visit.

This is getting more common, not less. The head of the jewellery business at one of India's largest jewellery chains told The Economic Times this month that more than half of the transactions in its stores now include some form of gold exchange. In Kerala, jewellers agreed to standardise 18-carat pricing from October 15, as demand for lighter, lower-karat pieces grows. More exchange and more variants mean more conversations per sale, and more things someone has to remember.

What goes wrong when the work lives in chats?

What I noticed is that the problems are rarely about skill. Jewellers know their trade better than any software ever will. The problems are about memory and follow-through.

A promise made on a call lives in one person's head. If that person is busy, on leave or leaves the job, the promise goes with them. Some businesses I know employ people whose whole job is collecting updates from others and passing them on. I have seen owners forward messages by hand so the right person sees them.

The question I kept hearing, in different words, was "What did we tell him?" Someone has to scroll back through weeks of chat, or call a staff member, to find out what was promised and when.

Repairs and karigar work have the same problem. A piece goes out, a date is given, and the only record is a chat and a line in a book.

Billing and stock systems handle stock, bags and invoices well. They were never built to hold the conversation with the customer. So the most important part of the relationship sits outside every system the shop owns.

What should AI for jewellery business owners actually do?

This is the part I find most interesting, and also the part I want to be careful about.

For the first time, software can read the way the trade already works. It can read a WhatsApp chat, listen to a voice note in Hindi, Gujarati or Telugu, and read a photo of a handwritten order pad. That matters because staff don't have to change how they work for the business to get organised.

Where I think it really helps:

• Follow-through on promises. When a customer is told "Thursday", AI can notice that and make sure someone is reminded before Thursday, not after the customer calls to ask.
• Turning conversations into tracked work. People often say one thing across four messages. AI can wait, read all of it together, and turn it into one clear piece of work for the right person.
• Finding what was said. "What did we tell him about the bridal set?" should take seconds, with the actual message shown, not someone's memory of it.
• Reminders that don't depend on one person. Repairs, karigar dates, pieces sent on approval. These need nudges, and nudges are exactly the kind of work people forget.
• Voice notes in any language. A lot of the trade talks rather than types. AI can turn a voice note into text the owner can read and search later.

None of this replaces anyone. It takes the memory work off people so they can do the work they are actually good at.

At first I thought owners would be happy using this on a computer at the counter. I was wrong. One of the first things owners asked me was, "Can I use it on my phone?" Salespeople walk the floor and take calls. They are not sitting at a laptop. If AI is going to help a jewellery business, it has to meet people where they already are.

Where should AI never be used in a jewellery shop?

This section matters more to me than the one above.

Prices and gold rates. AI should never calculate a price or a gold rate. Those numbers come from the shop's own sheet or, where a shop connects its ERP, from that system, and AI should only read them from there. A language model can produce a number that looks right and is wrong. In this trade, a wrong number is real money and a customer's trust.

Weights, purity and hallmarking. Fine weight, gross weight, karat, HUID. These are measured and certified, not guessed. AI can help find a record. It should not make one up.

Valuation and old-gold exchange. Testing and valuing old gold is a judgement made with instruments and experience, in front of the customer. AI can keep track of an exchange that is in progress. It should not value the gold.

Decisions and customer messages. AI can draft a reply, but a person should approve anything that goes to a customer. AI can point out that something needs attention, but the owner decides what happens.

Trust and relationships. Customers buy from people they trust, often families they have known for years. No software replaces that, and nothing should try.

I learned part of this the hard way while building. I told the software not to store things like order dates or phone numbers as its own "knowledge", and no amount of instruction wording made it fully obey. So the rule now lives in code. My takeaway is that the guardrails that matter can't depend on the AI behaving well. They have to be built in.

Another honest one: early on, a Hindi voice note came back labelled as Urdu. It was small, and it was fixed quickly. But it reminded me that AI makes mistakes in exactly the places where the trade is most specific, which is why a person has to stay in the loop.

Will owners and staff actually trust AI?

Honestly, this is the hardest part, and it isn't technical.

A business WhatsApp number is often also someone's personal phone, with family and the accountant sitting right next to customers. When software asks to connect to that, people want to know exactly what it keeps, who can see what, and what goes out in their name. Those are fair questions, and they deserve plain answers before anything is connected.

And in family businesses, it is rarely one person's decision. In demos I have heard "I'll have to show my brother" and "let me show my sister" more than once. I think that is healthy. AI that touches the customer relationship should be decided carefully.

What I'm still unsure about

I don't think anyone has fully worked out how much an AI should do on its own in a business like this. Too little, and it becomes another list the owner has to check. Too much, and it starts acting without the person who is responsible. I lean towards AI doing the follow-through and people making the calls. I'm not certain where that line sits for every shop, and I expect the businesses I work with to show me.

What I am sure of is this: AI should help a jewellery business remember and follow through. It should not decide prices, value gold or speak for the owner.

This is why I'm building Reva, an AI workspace for jewellery businesses, around exactly these limits.

Pratik Jain, founder of Reva

Drafted with help from my AI team.