← All case studies

AI agents for distributors: four agents working around the clock at Vinesco

AI agents for distributors: at Vinesco, purchasing, finance, sales and supplier deals each have an agent that advises with data refreshed every 5 minutes.

Year2026
IndustryWholesale beverage distributor · AI agents per role
DurationOn top of Vinesco's system · role-based assistant in production since October 6, 2026 · in testing with administration
What we did
  • AI agents for distributors
  • Purchasing and restocking agent
  • Checks and finance agent
  • Sales and collections agent
  • Supplier deals and credit notes agent
Tuesday · 3:12 a.m.
Purchasing 3:12

Buy what's needed and never run out of stock.

For the buyer Malbec Reserva: 9 days left. Suggested: 18 cases.
Finance 3:12

Never let the company go a day without cash.

For administration Friday the 7th goes negative. There's a week.
Sales 3:12

Every customer looked after, every account collected.

For each salesperson 3 customers in your area haven't bought in a month.
Deals 3:12

Make every supplier pay back what it owes.

For management October's credit note came in short. 15% is missing.
24 h · four agents awake, one person who decides

The case on video, explained.

Go to the questions ↓

This distributor is profitable and this one isn't. Why?

Both sell the same things, in the same city, with the same number of people. One runs out of its best-selling wine, finds out about the bounced check on the day it bounces, has customers who stopped buying without anyone noticing, and accepts whatever credit note the supplier sends. The other doesn't. The difference is that the second one has four agents working around the clock. It's Vinesco, a wholesale beverage distributor in Rosario, Argentina. On top of the system that already enters whatever comes in over WhatsApp, we added four AI agents, one per role: purchasing, finance, sales and supplier deals. They don't replace anyone: they advise the person in charge of each area, with the metrics they need to decide.

4 agents
one per role: purchasing, finance, sales and deals
Every 5 minutes
what they watch is refreshed, at any hour
0 changes
the agents look things up; the person decides
The aisle of a beverage warehouse with wrapped pallets and a pallet jack in the middle, by day Generated image
Fig. 01 — Without agents
A distributor as usual Any warehouse
The same warehouse at night, empty, with four blue panels glowing between the shelves Generated image
Fig. 02 — With four agents
Vinesco Rosario, Santa Fe · Argentina
Stock
You notice it's missing once it's gone
An alert with days of cover left
Cash
Cash flow, once a week
The negative day, flagged ahead
Customers
Whoever stops buying is lost
The salesperson knows whom to call
Suppliers
The credit note is taken at face value
It's checked against what was owed
The people
The same team
The same team

The purchasing agent: never run out of what sells

Its job is one: buy what's needed, when it's needed. It watches each product's actual sales, the stock and each supplier's lead time, and works out how many days of sales are left. Below 21 days it says it's time to restock; below 10, that it's critical. When the buyer asks what to order, it answers with the suggested order per supplier, priced at the last purchase and with how much it went up. It also looks the other way: anything with more than 45 days of stock, or that hasn't sold in months, is money sitting still, and the agent shows it so it doesn't get bought again. The order is built with one button and goes out by WhatsApp, email or Excel. The buyer is the one who sends it.

  • Under 10 dayscritical: order now
  • Under 21 daysrestock
  • Over 45 daysoverstock: money sitting still
Suggested order · Bodega del Sur · example
  1. 01Malbec Reserva 750 · 9 days left · 18 cases
  2. 02Malbec Roble 750 · 19 days left · 6 cases
  3. 03Torrontés 750 · 33 days left · not needed
  4. 04Last purchase price · up 10%
  5. 05Goes out by WhatsApp, email or Excel, when the buyer decides
Monthly sales × months of cover − stock. The agent works it out; the buyer sends it.

The finance agent: never a day without cash

At a distributor, money comes in and goes out in checks, e-checks and transfers on different dates, and the cash flow used to be built in Excel once a week. The finance agent builds it day by day and keeps it current on its own: incoming customer checks, the company's own outgoing checks, supplier invoices coming due, and the taxes and payroll administration enters. With that, it flags the first negative day, 15 or 30 days out, 3 months or a year. If administration asks which payment it can push back, it simulates invoice by invoice how each day ends up, without changing anything. And it keeps the month's VAT position, which in August and September matched the management system's to the peso, plus the withholding files for the tax agency.

  1. 01What comes inCustomer checks and e-checks, with their dates
  2. 02What goes outOwn checks, suppliers, taxes and payroll
  3. 03Day by day15 or 30 days, 3 months or a year out
  4. 04The first negative dayFlagged ahead, with time to act
  5. 05Administration decidesWhat to collect sooner or which payment to push

The sales agent: every customer covered, and properly collected

A salesperson with a large portfolio can't reach everyone, and the customer who stops buying is lost quietly. The sales agent watches each salesperson's portfolio and tells them whom to call: who hasn't ordered in a month, who's coming back from their season and which product someone stopped taking after always taking it. The other side is collections. Before every due date, the salesperson gets on their phone, over Telegram, what their customers owe, overdue first. They see their account, how each customer pays and how far they are from the 80% floor of their target. And the partners get the week's summary every Sunday. We estimate that collecting current accounts speeds up by 30% and that losses from selling to overdue customers drop by 10%.

  • Portfolio coverageWho stopped buying, who's coming back and what they stopped taking.
  • 30 %
    Collections before due dateOver Telegram, overdue first.
  • 10 %
    Selling to who owesThe salesperson sees what's overdue before taking the order.
  • 80 %
    The monthly targetWho's below the floor, before the month closes.

The deals agent: every supplier pays back what it owes

At a beverage distributor, a good part of the margin isn't in the price: it's in supplier deals. Bonuses, discounts and monthly promotions the supplier pays back later, as credit notes, with limits per customer and per month. If nobody keeps count, the credit note comes in short and gets accepted, and nobody knows how much credit they have with each supplier. The deals agent keeps that count. Each deal has its limit and its usage for the month, and it alerts if the brand changes a limit. It works out what's owed before the note arrives; when it does, it matches it to its deal and shows the gap. Each deal's balance stays current, and the claim is ready for management to send.

  • What's owedPer deal and per supplier, before the note arrives.
  • What arrivedEach credit note, matched to its deal.
  • − 15 %
    The gapReady to claim, with its backing.
  • Credit with each supplierEach deal's balance, up to date.

They advise, they don't decide

No agent buys, pays or files a claim on its own. They look up the data without being able to change anything and give the person in charge what they need to decide. Each one speaks to its role, and each role only asks about its own area: a salesperson sees their own portfolio, not someone else's, and the most sensitive data stays out of reach for anyone who shouldn't see it. You ask them in plain words, from the platform, or they come to you: the salesperson's alert, the partners' Sunday summary. What used to be a two-day report is now instant.

You ask It answers with What it won't do
Purchasing You ask“What do I order from each supplier?” It answers withThe suggestion, days of stock and how much each went up. What it won't doSend the order on its own.
Finance You ask“Which day do I go negative?” It answers withDay-by-day cash flow, checks coming due and VAT. What it won't doMove a payment.
Sales You ask“Whom do I call and who owes me?” It answers withTheir portfolio, overdue accounts and the monthly target. What it won't doShow them someone else's portfolio.
Deals You ask“What does each supplier owe me?” It answers withWhat's expected, what arrived and the gap. What it won't doSend the claim.

Each agent reads without changing anything · each role asks only about its own area

What changed for each manager

The buyer stopped building orders by eye, finance stopped finding out about the negative day on the day itself, every salesperson knows whom to call and who owes them without asking administration, and credit notes are checked against what was owed. Figures marked “estimated” don't come from a measurement: they're Mithandir's and the project lead's own count.

4 agentesworking around the clock, one per role

Purchasing, finance, sales and deals, on data refreshed every 5 minutes.

Generated texture
Al instantewhat used to be a 2-day report

You ask the agent in plain words, from the platform.

30%faster collection of current accounts · estimated

Each salesperson gets what's coming due before the due date, overdue first.

10%fewer losses from selling to overdue customers · estimated

The salesperson sees the customer's overdue balance before taking the order.

≈ 150 ha month of spreadsheets no longer built · estimated

The order to each supplier, the cash flow, each salesperson's accounts and the credit note claims.

Under the hood: the agents sit on top of the data entry layer

An agent advises only as well as the data it watches. That's why, at Vinesco, the agents are the second layer. The first is data entry: whatever salespeople forward over WhatsApp (orders, payments, supplier invoices) ends up entered on its own, with a person confirming. On top, a copy of the management system refreshed every 5 minutes brings everything into one database. The four agents work on that database, with Claude's language model, read-only and with the role of whoever is asking. Since October 6, 2026, the role-based assistant has been in production, and the alerts to salespeople and partners go out over Telegram.

24 h · from the data to the decision
  1. It arrives over WhatsApp Orders, payments and invoices, as always.
  2. Data entry It's entered on its own, once.
  3. One database Current every 5 minutes.
  4. Four agents They watch, compute and alert.
  5. The person decides Buys, pays, calls or claims.

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 see whether these agents would help your distribution company. It takes under a minute.

What does your company do?
How many salespeople take orders?
What happened to you in recent months?
Where are your sales, stock and collections data today?
When would you like to solve it?
Last step: who should we write to?

We only use it to reply about this assessment.

Other systems in production