How 150 stores in 10 countries run their whole back of house on AI agents

A franchise playbook is only worth something if an operator can use it today.

A franchisee does not start from zero. They buy into products that are already developed, suppliers that are already tested and lessons the brand has collected across years and markets. That knowledge is part of what makes the franchise valuable.

Churromania’s version of it lived across POS systems, delivery platforms, supplier catalogs, reports and the people who knew where to look. An ordinary week could produce very different questions: did new advertising in a food court change sales, how do we split 1,000 promotional prizes without leaving a store behind, how does one inventory request get split across three distributors.

Clave built an ontology of the brand before it automated anything.

The menu and its ingredients, products and pack sizes, stores and franchise groups, suppliers and regional availability, local area marketing, market demographics, country rules, connected systems and the people allowed to make each decision.

That ontology is a living knowledge base, not a diagram. A product can have one supplier in Florida and another in Chile. A promotion can matter in one market and not another. A correction from an operator becomes reusable knowledge in the places Churromania authorizes it.

Ordering, forecasting and reporting stopped being separate jobs.

An operator says what the location needs. Clave identifies the items, splits the request by distributor, builds the correct order for each one and tracks it through delivery. Items Sysco carries go to Sysco, the rest is routed to the right regional distributor. Clave checks the supplier connection and the user’s authority to buy, and stops before submitting when either is missing.

Clave also watches sales and inventory, flags where a store is about to run short and shows the products, quantities and expected demand behind the suggestion. Its forecast reads historical sales with current trends, weather and local events, because a hot weekend, a nearby event or the Fourth of July change what a location sells.

Reports are workflows now. One workflow sends a Spanish month to date comparison at 9:00 every morning. Chile gets its own performance report at 8:00 AM Santiago time, with in store and delivery sales separated correctly, to seven recipients. Puerto Rico has its own daily view. Twenty two reports are currently active.

Operators text Clave from the floor, so nobody had to learn a new habit.

Restaurant operators work between locations, on email and in WhatsApp. An operator can text Clave or send a voice note, ask what happened yesterday, request a report, start a workflow or approve an action in the same conversation. Across four months that is 1,144 email sessions, 200 WhatsApp sessions, 432 web sessions and 40 mobile sessions.

General AI can answer a question. Purpose built agents finish the restaurant workflow, in the channel the operator already had open.

Corporate can see the network while it is operating, not a month later.

In Chile, the team noticed PedidosYa Cash Collection was missing from delivery reporting. Once corrected, Clave kept the definition for every later question and report. It also surfaced a store where the Uber Eats connection to Toteat was not active.

Local rules stay local. Corporate sees the questions operators are asking, the reports being generated, the orders being placed and the definitions being corrected, and the next franchisee inherits all of it instead of rediscovering it.