Home › Success stories › Club House Arenas del Sur
🧾 AI Engineering · Case #02

Carla: the AI management agent of Club House Arenas del Sur

In a restaurant, money leaks through small details nobody notices until the end of the month: a drink charged at the wrong price, an invoice that went up without notice, or an advance that was never deducted. By the time the data shows up, it's too late to fix it.

That's why we built our own agent that detects these inconsistencies the same day they happen, so they can be fixed before they turn into losses.

ClientClub House Arenas del Sur
SectorRestaurant
DisciplineAI Engineering
RuntimeOpenClaw
Same day
Deviations caught before they hit the month-end close
Less administrative load
Invoices, emails and operational checks processed automatically
Continuous financial control
Costs, cash, payments and margins monitored with up-to-date data
Orderly staff management
Clock-ins, hours, advances and payroll validated systematically
Who Carla is

It's not a dashboard. It's a manager with judgment that works the same day.

Carla is the management agent of Club House Arenas del Sur, a restaurant with its own team in kitchen, floor and administration: around 16 people on time tracking, more than 400 costed recipes and dozens of supplier invoices per week. The operation is that of any well-run restaurant — with all the complexity that implies: moving costs, supplies arriving at a different price each month, payroll to settle and cash to reconcile every day.

The problem wasn't a lack of data. It was that it arrived late. The books closed at month's end, and only then could you see whether the result had been within expectations. The small errors — a bad entry in the POS, an invoice that went up without warning, an advance left dangling — piled up invisibly for 30 days.

What used to be done by hand

Daily operations concentrated critical, manual and error-prone tasks.

Costing dishes with a spreadsheet and old prices, or by eye (price ÷ 1.4).

Entering invoices one by one, with no check on whether the cost went up.

Settling payroll with a calculator and waiting for the accountant's payslip to know the number.

Reviewing time records by hand, with no hours bank.

Looking at the cash without cross-checking advances against wages.

Finding out about money problems at month's end, when they can no longer be fixed.

Which areas she helps manage

One agent integrated across different areas of operations.

Administration

Processes supplier invoices, validates prices and costs, reconciles cash movements, and cross-checks bank and payment-method data to spot discrepancies.

Human Resources

Checks clock-ins against the plan, manages the hours bank, and prepares the information needed for payroll and advances.

Kitchen & costs

Costs recipes ingredient by ingredient, updates prices, calculates COGS, and flags cost or margin deviations out of range.

Management & leadership

Generates daily sales reports, compares results with prior periods, and consolidates information to ease economic and operational tracking.

Communication

Reviews incoming emails, replies when appropriate, and routes each query to the right person, following up on pending administrative matters.

How it interacts with the team

Part of the internal team. Every question goes to the right person.

Nacho / Luchi
Owner · Administration

Main points of contact. They review money decisions, enter payments, reconcile cash and close the month with her.

Lucas
Head chef

He passes her weights, brands and yields of supplies. Carla returns costs, weekly suggestions and per-dish stats. She doesn't show him revenue or amounts: they're not his to see.

José
Floor manager

He brings her invoice questions and operational topics, in plain language, no jargon.

A rule it respects: supplies → kitchen. Invoices → floor manager. Structural decisions → owner. And per-person permissions aren't a minor detail: it adapts the level of detail to whoever it's talking to. To suppliers it speaks plainly; to the owner it gives the full detail.

Concrete examples of impact

What changed, with real cases.

01

The revenue leak no one could see

On the floor, drinks were being rung up with the kiosk product (cheaper) instead of the table product. On drinks alone, the error was on the order of tens of thousands of pesos per quarter, invisible until Carla cross-checked sale by sale and caught it. Fixed in time, it stops piling up.

02

A measurement error that moved millions

It found that revenue was being measured without subtracting discounts at the ticket level. The correction moved the month's number on the order of millions — a figure that changed how the business read. It wasn't a money leak: it was a wrong picture.

03

Settlements that match the accountant to the peso

It rebuilt a payroll settlement from the time records and it matched the accountant's payslip to the peso. That builds confidence: the restaurant can close payroll without depending on waiting for the external payslip.

04

Cash actually reconciled

It detected salary advances that inflated the cash because the inflow was recorded but not the outflow when the employee got paid. Once fixed, the cash started reflecting real money.

05

Real autonomy, with review where it matters

Circuitos que corren 100% solos (reconocimiento y aplicación de facturas, reporte diario de ventas, revisión de mails, chequeos de control). Circuitos donde propone y el dueño confirma (cambios estructurales, cambios de precio de venta, cierre de liquidación de sueldos). Nada crítico se aplica sin revisión; todo lo repetitivo corre solo.

In a restaurant, money leaks through small details no one sees until month's end. We went from finding out at month's end to detecting it the same day.

The technical side

How it's built

It's not plug-and-play. It was built piece by piece, around how the restaurant works — with a calibration and learning phase per supplier, invoice by invoice. That's the difference between using a generic assistant and having your own agent.

Integrated with the venue's systems

It reads and writes directly at the heart of the operation: sales, costs, payroll and cash.

POS FudoOwn ERPGmailDrive / SheetsBankMercado PagoWhatsAppTelegram

Invoice recognition

They arrive as a photo or PDF. It reads them, extracts line by line and matches against the catalog with per-supplier learning profiles — tuned invoice by invoice until they're sharp.

What runs alone vs. review

Automatic: invoices, daily sales report, email checks and control checks. With human review before applying: structural decisions, sale-price changes and payroll close.

Work frequencies

Invoices several times a day. Sales report every morning. Emails every 15 minutes. Time records on Mondays. Payroll settlement at month-end close. Control checks, daily.

Per-person permissions

It adapts the level of detail to whoever it's talking to: it doesn't show the head chef amounts or revenue; to suppliers and the floor manager it speaks plainly; to the owner it gives the full detail.

Rollout

It's been running since early 2026. There was a calibration phase: the invoice-reading profiles and product mappings were tuned invoice by invoice. It's an agent that develops in the real operation.

The key isn't the number. It's when it's detected: the same day, not at month's end.

A mischarged drink, an invoice that came in higher, an advance that wasn't deducted. The sum of those small leaks, caught in time, is what ends up moving the month's result.

Same day
The correction lands while it can still be fixed, not once the month has already closed.

How would an agent like this fit your operation?

Club House is a restaurant case, but the logic carries over to any business with daily operations of costs, invoices, payroll and cash. Your own agent develops in the real operation, learns how each supplier works and executes only what can be done on its own — leaving human judgment for where it matters.

Series Success stories · AI agents in real operation Case #01 — Catena, Seotronix's AI SEO Specialist.