Case study

90s Coffee Roasters

Two WhatsApp assistants connected to the POS of a specialty coffee shop in Buenos Aires

The owners ask on WhatsApp and get the real number in seconds. No logging into the system, no spreadsheets, no waiting until the end of the month to find out what happened.

IndustryCoffee shop (Buenos Aires)
Duration3 months (live in production)
PeriodMay 2026 — present

The challenge

90s Coffee Roasters is a specialty coffee shop in Buenos Aires, run by two partners who work behind the counter themselves. The POS system recorded everything, but the numbers only showed up for whoever stopped to look for them — and when you're making coffee, you don't stop.

What we did

01
WhatsApp query assistant
The owners ask in plain language — "how much did we sell yesterday", "what sold the most this week" — and get an answer with real POS data in seconds.
02
Expense logging by message
An expense is logged by sending a message to the group chat. No app, no form, no putting it off.
03
Anti-hallucination architecture
The assistant only answers with numbers that came from the system. When it doesn't know, it says so — it never makes up a plausible value.
04
Executive report with combination analysis
An association rules study of the sales history, revealing which products sell together and how strongly.
05
Automatic sales sync
A routine that runs every night consolidating the day's sales, building a searchable history the POS didn't offer on its own.
06
Low stock alert
A query that shows what is running out before it runs out, flagged by level of urgency.

Results

Numbers from the client's real operation, live since July 2026.

2WhatsApp assistants in production
<1minFrom question to the real number from the system
4.11×Strength (lift) of the most relevant product association found
5Question types understood in plain language
0Spreadsheets opened to look up a number
24/7Available, even outside store hours

Try it yourself

This is the same technology running at the coffee shop, with made-up data from a sample store. Ask your question however you like — the assistant understands plain language, not commands.

☕
Assistant · Café Venezia
online
Hi! I'm Café Venezia's assistant. Ask me about revenue, best-selling products or stock ☕
How much did I sell yesterday? What sold the most this week? How many cappuccinos did I sell this month? Is anything running low? Tell me a joke

The numbers above are made up for demonstration. The real 90s Coffee assistant replies in Spanish, using the store's POS data — the same system works in any language.

How it worked

Assessment of existing data
A survey of everything the POS was already recording and what could be extracted. First executive report with the full history consolidated.
Automating the biggest pain first
Expense logging by message came first, because it was the daily headache: writing things on paper and entering them later.
Query assistant
Built with a clear rule: it only answers with numbers that came from the system. Each question type was tested individually before going live.
Infrastructure built to grow
A central router sends each WhatsApp message to the right assistant. Adding a new assistant is configuration, not a rebuild.
Live and used every day
Both partners use it from behind the counter, between one coffee and the next. Asking takes five seconds and the answer arrives before the next customer.

Tools used

n8n WhatsApp POS API (Fudo) LLM (Groq) Python Google Sheets

Want something like this for your business?

A quick 20-minute call to understand your operation. No commitment.