Deals don't die in WhatsApp because nobody cares. They die because a chat is not a pipeline. Here's how to make it one.
October 2026
6 min read
You can turn WhatsApp into a sales pipeline by moving your number to the official WhatsApp Cloud API, working from one shared team inbox, and letting AI read every message and photo to extract structured rows: what's offered, what's wanted, at what price. Those rows get matched against demand, stock and history, and people still close the deal.
If your customers buy and sell through WhatsApp, you can turn those chats into a real WhatsApp sales pipeline. Move the number to the official WhatsApp Business API, give the team one shared inbox, and let AI read every message (and every photo) and extract a structured row: what is offered or wanted, condition, price, next step. Then match those rows against open requests, stock and past sales. People still do the selling. The AI just makes sure nothing scrolls away.
I built exactly this for a jewellery and pre-owned watch retailer. Customers sent photos of watches they wanted to sell, or asked for a specific reference. All of it landed on one phone. And most of it scrolled away.
That's not a people problem. A chat is just not a pipeline.
WhatsApp is where your customers already are. That's the good part. The bad part is everything else:
This applies to far more than jewellers. Car dealers getting trade-in photos. Real estate agents getting "is this one still available?". Second-hand and B2B traders swapping stock lists in chat. Wholesalers taking orders by voice note. Same pattern every time: the deal flow is real, it just lives somewhere you can't query.
You can't build anything on top of the WhatsApp app on a phone. You need the official WhatsApp Cloud API from Meta. Once the number moves there, messages arrive through a webhook in your own system, and a WhatsApp Business API shared inbox becomes possible: the whole team works from one place, chats get assigned, history stays.
I've built this twice now. For a trading-education company it was mainly about the team: several people answering from one number instead of passing a phone around. For the jewellery retailer it became the base for everything else. Their inbox has templates, welcome and away messages (also in English), order lookup by phone number in the webshop, product cards with shop links, and an importer for old chat exports so history isn't lost.
Two rules you need to know before you start:
Also: moving the real number off the phone app is a migration, not a toggle. Test on Meta's test number first, write a runbook, then move.
This is the part I like most. Inside the retailer's inbox, Claude does a few small jobs: translate a message, suggest a reply, polish a text before it goes out. Useful, but not the point.
The point is AI WhatsApp lead extraction. Claude reads the whole conversation, including the photos, and turns every watch or piece of jewellery that's offered or wanted into one structured row:
A photo of a watch plus "what would you give me for this?" becomes a row someone can sort, filter and act on. That's the moment WhatsApp stops being a chat app and starts being a WhatsApp CRM.
A row on its own is just tidier chaos. The value comes from matching. Every extracted row is checked against three things:
Stock and sales history live in the owner's own Google Sheet. I didn't replace it. It syncs read-only every hour, changes only, and the sync aborts if the result looks suspicious. When there's a match, an internal note appears in the chat, visible to the team, never to the customer. There's also a market search link per reference, so pricing doesn't start from a guess.
A conversation that would have scrolled away is now a deal with context. You can see the whole build in the jewellery retailer case.
The building blocks are the same for almost any business that trades in chat:
What I keep human, on purpose:
And privacy, because these are your customers' messages. The AI runs through the commercial Claude API, where customer data is not used to train models. No consumer chat apps, no copy-pasting chats into ChatGPT. Keep a retention policy (this client keeps chats for 24 months), give each role only the chats it needs, and never use real chats or photos in demos.
Honestly, not everyone needs this.
If you're somewhere in between, start with the shared inbox. Add extraction when you notice you're losing deals, not before.
More builds like this are on my work page. If your deals live in WhatsApp and you want to know what this would look like for you, book 30 minutes with me. No pitch, just your chats and your process.
Not really. The app is built for one phone and has no proper way to read messages into your own system. For a sales pipeline you need the official WhatsApp Cloud API, which sends every message to your software through a webhook.
It's one inbox where your whole team answers from the same WhatsApp number, built on Meta's official API. Chats can be assigned, templates and away messages are built in, and the history stays with the business instead of on someone's phone.
Yes. A model like Claude can read the conversation and the images together and return structured fields, such as brand, model, condition and asking price. Those rows can go straight into a table and be matched against stock or open requests.
Not in this setup. The AI runs through the commercial Claude API, where customer data is not used to train models, and chats are kept under a fixed retention period.
It doesn't have to, and I'd advise against it for valuable deals. In the setup I built, AI suggests replies and extracts data, but a person always decides the price and sends the message.
You can send free-form replies within 24 hours of the customer's last message. After that window closes, you can only start the conversation again with a template that Meta has approved.