← All case studies

Software for food distributors: the WhatsApp order module

A Rosario frozen food distributor went from WhatsApp order to delivery note: 20 seconds per order instead of 5 minutes, and 93–94% accuracy on its own.

Year2026
IndustryFood wholesalers · WhatsApp order module
DurationSince March 2026 · panel in production since July 2026
What we did
  • WhatsApp orders turned into delivery notes
  • AI item and customer recognition
  • Panel to resolve doubts and confirm
  • Connection to the existing management system
20 s · per order; before, over 5 minutes
  1. Comes in The salesperson forwards the customer’s WhatsApp.
  2. Reads Customer, location, quantity and unit.
  3. Builds Real catalog, discounts and stock.
  4. A person checks Picks only where it’s unclear.
  5. Note goes out In the usual system, right away.

The case on video, explained.

Go to the questions ↓

One module, one job: the WhatsApp order goes out as a delivery note

Software for a food distributor has to understand how people order. This Rosario frozen food distributor also sells dry, fresh, dairy and disposables to restaurants, bakeries, delis, hotels and clubs, with salespeople on the road and a catalog of thousands of items. Orders came in over WhatsApp the way each customer wrote them: “mozz”, “cream”, “2 ham”. Someone in the office translated them from memory and retyped them into the management system. There we installed a single Mithandir Core module, WhatsApp orders, without touching its management system. Today it builds the order from the message, looks up each line in the real catalog, flags whatever it isn’t sure of so a person can pick, and issues the delivery note in the same system as always. Each order is entered in about 20 seconds, as soon as it arrives.

Food warehouse: shelves with boxes, oil jugs and sacks, next to the door of a cold room Generated image
Fig. 01 — Without the module
Traditional distributor The chat and someone typing
Cold room with a digital delivery note floating in the air and a worker with a pallet jack in the back Generated image
Fig. 02 — With the module
Frozen food distributor Rosario · Argentina
The customer orders
“mozz”, “cream”, “2 ham”
The same: in their own words
Who translates it
Someone in the office, from memory
The real catalog and the sales history
If stock is short
Found out on delivery
No half delivery notes
Doubts
By message, while the order waits
In the panel, with the options
When it’s entered
All together, at 2 p.m.
As it arrives, in 20 seconds
The management system
Its vendor’s
The same: the note comes out there
Generated image

Entering a food order was not copying it

The customer places the order with the salesperson however it comes out: a text, a pasted list or a photo. The salesperson forwards it and, in the office, someone enters it again line by line to issue the delivery note. And entering was not copying: “mozz” or “cream” can be several catalog items depending on who orders, one customer has several locations, oil is ordered in liters and sold in jugs, and ham is ordered by the piece and charged by the kilo. That person set the pace of the business: each order took over 5 minutes, and they were batched to be entered all at 2 p.m. Until then, stock knew nothing. On Mondays, when restaurants restock, the queue was longer.

Orders forwarded to the office
  1. deli at the park
    downtown store
    mayo x 3 kg
    50 lts oil
  2. PANADERÍA SUR
    4 flour 000
    1 case butter
    3 yeast
  3. Photo · Handwritten order
  4. hotel plaza: 2 cream
  5. bar 9 de julio: 2 cooked ham
  6. resto la costa: 6 mozz
To type in6
  1. 7:48 · Text Pick the location and look up each item
  2. 8:05 · Pasted list Convert the case of butter into units
  3. 8:20 · Photo Read the photo and type it in line by line
  4. 8:41 · Text Ask which: there’s 1 L and 5 L cream
  5. 9:02 · Text Decide whether it’s 2 pieces or 2 kilos
  6. 9:30 · Text Check stock before issuing the note

Goods went out at the pace of the typing. On Mondays, the queue was longer.

A Monday morning · example

The module, step by step: what comes in, what it reads, what it builds, what a person checks and what goes out

This is how it works today, with a real order. Since October 14, 2026 the WhatsApp doesn’t reply or ask: it receives, and everything else happens in the panel.

  • 1 · What comes inThe salesperson forwards to the distributor’s number the order the customer sent: a text, a list or a photo of a handwritten list. Only registered salespeople’s numbers count; an unknown number builds nothing. Voice notes aren’t read yet.
  • 2 · What it readsAI reads the message or the photo and pulls out the customer, the location, each item as the customer wrote it, the quantity and the unit: “50 lts oil”, “2 cooked ham”, “1 case butter”.
  • 3 · What it buildsIt finds the customer and location, and each line in the catalog and in what that customer bought before. It turns liters into jugs and cases into units, keeps ham by the piece, applies the customer’s discounts and checks stock. A draft lands in the inbox.
  • 4 · What a person checksThe salesperson or the office opens the draft in the panel. What the system resolved comes marked; what’s unclear comes with the closest options and the person picks, about 1 in 5 lines. That choice is learned for that customer.
  • 5 · What goes outOn confirming, the panel validates and issues the delivery note in the distributor’s management system, with its number, and stock is deducted right away. If a line is out of stock, it doesn’t go out half done: it stops and says which. The order stays as a row, with what was picked and the delivery note number.

AI item recognition, with rules that make it trustworthy

The salesperson still sends the order over WhatsApp, as always. The system reads it with AI and turns it into a built order: customer and location, items, quantities and units. Each line is looked up against the real catalog and against what that customer bought before: the system started out with the sales history. What it is sure of, it resolves. What it isn’t, it doesn’t guess: it flags it in the panel with the possible options and a person picks; that choice is learned for that customer. In this business, a mistake shows up when the truck reaches the restaurant. So, when nothing is left in doubt, someone confirms and the delivery note is issued in the same management system as always, with the customer’s discounts and checking stock in both the warehouse and the store; if something is short, it doesn’t go out half done.

What it’s sure of, it resolves
It’s right on 94% of what it resolves without asking.
What it isn’t, it doesn’t guess
About 1 in 5 lines goes through a person, who picks among the options.
Out of stock, no half orders
The order stops and says which line fails.

What changed in the office

The office stopped typing: it reviews what the system flagged, picks and confirms. There are two different times: how long an order takes to be entered, and how long it takes from the customer sending it to it being in the system. Both shrank.

≈ 250 hof order entry saved a month: more than one full-time store employee

And it’s a single module. At Vinesco, with the whole system, about 370 hours are saved a month.

20 sto enter an order

Before, over 5 minutes: finding each item, asking about what was unclear and typing it in. Now a person reviews and confirms.

Al llegarthe order enters the system, from when the customer sends it

Before, it waited until 2 p.m., when they were all entered together, and stock knew nothing until then. Now it goes in as soon as it arrives and stock mirrors it.

93–94%accuracy on what the system resolves without asking

In the late-September 2026 measurements. About 1 in 5 lines goes through a person, who picks among the options.

How the module is used, person by person

The module slotted into the workflow without asking anyone to switch channels. Customers still message their salesperson; the salesperson forwards or enters the order; the office settles whatever was left in doubt and issues it; management watches the dashboard. Against a traditional distributor, the difference is in the last column: what each of them stopped doing.

What they do What they see What they no longer do
Customer What they doOrders from their salesperson over WhatsApp, in their own words. What they seeTheir order, complete, on the truck for their area. What they no longer doAnswering “which cream?” and waiting for the reply.
Salesperson What they doForwards the message, list or photo, or enters it in the panel. What they seeTheir orders and the status of each one. What they no longer doAnswering menus over WhatsApp.
Office What they doPicks wherever there was doubt and issues the delivery note. What they seeThe day’s inbox, the stock and each customer’s discounts. What they no longer doTyping every line, finding it in the catalog and entering everything at 2 p.m.
Management What they doWatches the dashboard by delivery date. What they seeOrders, amounts and salespeople, up to date. What they no longer doCounting orders by hand at the end of the day.

Customers change nothing · salespeople learn no new screen · the delivery note comes out of the usual system

The order and the catalog, learning side by side

The distributor started in March 2026, after seeing the system working at Vinesco, with real sample orders and its item list. Before reading a single order, we loaded the sales history so the system would know how each customer orders. Meanwhile, its vendor was building the API to issue delivery notes: the first one went out on July 8. From then on, we measured every week what it got right and what it didn’t, and every mistake became a rule: ham by the piece, store stock added to warehouse stock, milk that comes in closed packs. At the end of August we trained the salespeople, and in October we removed the WhatsApp buttons: doubts are resolved in the panel.

The orderThe catalog
  1. 26 mar 2026 Starts, with real sample orders
  2. may 2026 Each customer’s sales history
  3. 5 jun 2026 The vendor finishes the issuing API
  4. 8 jul 2026 First real delivery note from the panel
  5. 13 aug 2026 Warehouse and store stock, added up
  6. 31 aug 2026 Salespeople trained; order photos over WhatsApp
  7. 7 sep 2026 “1” means a piece, not a kilo
  8. 30 sep 2026 94% accuracy on what it resolves alone
  9. 14 oct 2026 No buttons: doubts go to the panel
  10. 19 oct 2026 Milk in closed packs, by the pack
The delivery note · In the same management system
  1. oct 2026 Every order, entered as it arrives

What using it made us change

We designed for how the team works, not for how we imagined it would. Every change is dated, with what happened and what we did.

  1. Rev. 01 · 6 apr 2026 What to do if an item in the order is out of stock? The distributor’s call: no half delivery notes. The order stops and says which line fails.
  2. Rev. 02 · 15 may 2026 The management system’s vendor asked that nobody touch its database. The panel doesn’t write to its database: it validates and issues through the API the vendor built.
  3. Rev. 03 · 10 aug 2026 “1 ham” was invoiced as one kilo, and the piece weighs seven. Since September 7, “1” means one piece. The rules for each unit are the house’s.
  4. Rev. 04 · 7 oct 2026 Almost nobody answered the button menus we sent over WhatsApp. On October 14 we removed the buttons: doubts are resolved in the panel, and messaging costs went down.
  5. Rev. 05 · 15 oct 2026 “12 milk” went out as 12 packs, 144 liters, for the brands sold in closed packs. Since October 19, for those brands twelve units are one pack; the one sold loose stays by the unit.

Under the hood: the real catalog, the history and the usual delivery note

The panel is built with Next.js and TypeScript on Supabase, and connects to WhatsApp Business and to the distributor’s management system, which wasn’t replaced: that’s where its customers, stock, price lists and delivery notes live. AI reads the message; the math (units, kilos, jugs, stock) runs on fixed rules. What used to be a message, someone searching and someone typing is now a row: every order with its status, what was picked and the delivery note number.

What it was
  1. A message 50 lts oil, 2 cream
  2. The catalog
  3. Someone typing Oil 5 L x10
Becomes a row · example
Order · Rotisería del Parque
Status
Note issued
Picked
“cream” → 5 L
Note
0001-00004521
History
  1. 7:48 message arrived
  2. 7:49 built, 1 flag
  3. 7:55 note issued

Answer and we’ll tell you if this works for your company.

We review it quickly. If it’s a fit, we set up a meeting.

Assessment

Five questions to find out if what worked for this distributor fits your company. Two minutes.

What does your company do?
How do orders reach you today?

You can pick more than one.

Pick up to 4.

Roughly how many orders come in per week?
Where do you issue delivery notes or invoices?
When do you want to solve this?
Last step: who should we write to?

We only use it to reply about this assessment.

Other systems in production