The situation
If you run a distribution business in India, this probably sounds familiar. Retailers send orders on WhatsApp, in whatever words they like. "5 case of the 20 rupee pack, 2 of the new flavour, same as last time." Someone in the office reads every message, works out what it means, opens Tally, and types the whole thing in again. Then they make a proforma invoice and send it back.
That was the daily routine at a B2B FMCG distributor in Ahmedabad I work with. Nobody wanted to change how customers order. WhatsApp works. The problem was everything that happened after the message arrived.
Why retyping orders hurts more than it looks
On a quiet day it is just slow. On a busy day it is where the mistakes come from.
- The rules live in someone's head. Cases have to be converted to pieces. "Buy N, get one free" schemes scale with carton size. Some schemes are paid as free product worth a percentage of the order. Prices differ by customer. GST is split differently for in-state and out-of-state buyers, then rounded off.
- Every order is typed twice. Once by the customer on WhatsApp, once by your team into Tally. Twice the work, and a second chance to get a number wrong.
- One wrong number is a real cost. A missed scheme or a wrong rate means someone gets billed incorrectly, and fixing it later costs more time and trust than entering it right.
Why "just use AI" is not the answer
The obvious idea is to paste the WhatsApp message into an AI tool and let it produce the invoice. It will look right most of the time. That is the problem. An AI that is right most of the time is fine for reading a message and not fine for deciding what a customer owes you.
So the system was built around one rule: the AI reads, it never calculates.
What we built
The order still starts on WhatsApp. Behind it, four things happen:
- The AI reads the message. It turns loose wording into a clean list of products and quantities. That is all it does.
- Fixed rules work out the bill. Case conversions, schemes, customer pricing, GST and round-off are calculated by plain business rules, the same way every time. Those rules were checked against real past invoices before anything went live.
- A person approves it. Nobody gets an invoice and nothing goes into Tally until someone in the office has looked at the generated proforma invoice and clicked approve. The retyping is gone; the decision still belongs to a person.
- It lands in Tally on its own. The approved order is written into Tally Prime as a sale order, and the customer gets their proforma invoice.
One more safeguard matters more than it sounds. Every customer group on WhatsApp is mapped on purpose. If a message comes from a group the system does not recognise, it simply does not reply. Sending one customer another customer's negotiated price would be far worse than sending nothing.
The scare before launch
A few days before go-live, Tally crashed on the office computer during routine data work. No business data was lost, but it made the stakes very concrete. So instead of patching around it, we set new working rules:
- The backup refuses to run while Tally is open.
- The first write always goes to a test company, never the live one.
- The office computer is never reachable from the internet. It reaches out to collect approved orders; nothing reaches in.
The lesson was simple. With a system that writes into your accounts, deciding what it is never allowed to do matters more than adding features.
What changed
Orders now become reviewed sale orders and customer-ready proforma invoices without anyone retyping the business in between. Customers order exactly as before. The team checks instead of types.
If you are thinking about automating your own orders
A few questions worth asking before you start, whoever builds it:
- Where do your orders actually arrive? Automate that channel. Do not ask customers to move to a new app.
- Which rules are written down, and which live in one person's head? The second list is the real project.
- Who approves before money or stock moves? Keep a person there. It is cheap, and it is what makes the rest safe.
- What should the system refuse to do? Decide that first.
Questions people ask
Can WhatsApp orders be entered into Tally automatically?
Yes. An AI model can read a free-text WhatsApp order and turn it into line items, a rules engine can work out the bill, and an approved order can be written into Tally Prime as a sale order without anyone retyping it.
How do I automate data entry in Tally?
For orders that arrive as messages, an AI model reads each one into line items, fixed rules work out the bill, and approved orders are written into Tally Prime automatically, so nobody retypes them.
Do my customers have to change how they order?
No. They keep sending orders on WhatsApp the way they already do. The automation works in the background after the message arrives.
What happens if the AI misreads an order?
Nothing reaches Tally or the customer until a person has reviewed and approved the generated invoice. If the AI misreads an order, that check is where it gets caught, before the customer ever sees it.
Does the AI calculate prices, schemes and GST?
No. The AI only reads the message. Every number, including case conversions, schemes, GST and round-off, comes from fixed rules that are checked against real past invoices.