How Clave brings stone age QSR software into the AI age

Generic AI expects a clean API. Restaurant operations do not have one.

Generic AI assumes every company has a modern warehouse, an MCP server, an API key and an engineer ready to wire it together. It works beautifully after somebody else finished the hard connection work.

Restaurant software is frustrating even in the United States. In South America the gap is wider: country specific accounting platforms, local vendors, terminal era back offices. The data may sit in NCR Back Office, an old SAP transaction or a desktop application, and some systems are not publicly reachable at all and only open through a private VPN tunnel. The person who knows what a number means is often the operator, not the vendor.

Those operators deserve plug and play too. Clave fought through SAP R3, NCR NBO, Restaurant PE, Contífico, private tunnels and terminal applications with its own models and internal agents, so that from the operator’s side the work is one thing: open Clave and ask.

The connection is the start. The brand ontology is what makes it useful.

Clave builds an ontology for each group: stores, products, modifiers, channels, suppliers, financial definitions, roles, recurring reports and operating rules. That gives the agent a working model of the business instead of a pile of fields.

A sale is not just a sale. A modifier may be free by design, priced wrong or used as internal shorthand. An empty store may be closed, not failing. A change to average ticket can make a whole recurring report wrong. The agent has to know the difference before it can be trusted, and the operator still sees one place to ask.

SAP and NCR Back Office, in the language of restaurant operations.

Pardos Chicken runs 30 restaurants in Peru on SAP R3, Restaurant PE and NCR Back Office. Sales sit in ZPISD002 Reporte LOG. Inventory uses MB5B. Waste appears in MB51 movement types 971 and 972.

In one week Clave mapped the data behind Pardos’s sales, inventory and waste workflows. The goal was never to teach an operations manager transaction codes. Instead of navigating three systems to reconstruct an answer, the manager works through Clave. The systems stay where they are.

The agent learned how the brand actually operates.

Cultura Açaí runs on Contífico, an Ecuadorian platform that plays a QuickBooks like role locally. Its menu recorded toppings as $0 modifiers, and Clave found 12,461 topping uses in a single month, then turned that into an operating question: are these intentionally free, or is the business leaving money on the table.

The same catalog work caught duplicate naming between ACAI BOWL GRANDE and BOWL GRANDE. When the owner challenged an average ticket calculation, Clave corrected the logic, resent the report and carried the approved definition forward. A location with zero sales was recognized as permanently closed, a store status fact rather than a demand collapse.

From a sales question to the purchase the team has to make next.

Rogers also runs on Contífico, and its team uses Clave from WhatsApp. A manager asks about yesterday’s sales. Clave compares stores, separates the channels, identifies what changed in product demand and turns that into the raw material purchase that follows.

The agent does not stop at a chart, and the same conversation becomes the scheduled report. Nobody has to think about Contífico as a data source, the report as a separate job or WhatsApp as a disconnected channel.