- ai
AI Copilot for Business Data: A Real Case Study
How we connected a restaurant business's POS, inventory and loyalty systems into its own data infrastructure, then made it queryable by AI through MCP.
Last updated on
Almost every company running more than one system has the same problem: the data exists, but it’s scattered. Sales live in one system, inventory in another, customers in a third, and nobody can ask that information a simple question without cross-checking three screens by hand. This AI copilot case study is the story of a real client, a restaurant business in Spain, for whom we built the opposite: a copilot that knows the business, because it has direct access to the real data, all in one place.
The problem: three systems that don’t talk to each other
The project didn’t start as an AI project. It started as a scattered-data problem. The business used Ágora as its point of sale, GStock to manage inventory, production and purchasing, and Cheerfy as its loyalty system. Each did its job well, but none of them talked to the others.
Answering something as simple as “which products are about to run out, compared with what sold this week, and which loyal customers buy them most?” meant logging into three tools and piecing the answer together by hand.
What we built
We connected to all three systems through their APIs:
- Ágora for sales data.
- GStock for recipes, production runs, waste, inventory, purchases, incoming delivery notes and invoices.
- Cheerfy for CRM data, email campaigns and the purchases of customers enrolled in the loyalty program.
With that information we cleaned the data (the same products under different names in each system, dates in different formats, duplicates) and built the pipelines that keep it up to date. The result is a data infrastructure that belongs to the business, not to us, connected to the same three systems it uses every day. If you’re interested in the engineering side, we explain how we connected POS, inventory and loyalty data in a separate article.
From data infrastructure to AI copilot
That infrastructure is exposed through an MCP server, so an AI model like Claude can connect to it directly. That’s what turns it into a copilot: someone on the team asks in plain language how sales are going this week, or which supplies need restocking before the weekend, and gets an answer based on real, current data from all three sources at once, not an estimate. If MCP is new to you, start with what an MCP server is.
Having clean, connected data in one place also made it much simpler to build new internal tools: a dashboard with the business’s information in one view, and a purchasing and production tool that previously would have meant months of development, because the hardest part (reliable data in one place) was already solved.
What began as a systems integration ended up being, in practice, an AI project. The data infrastructure was the step that made everything else possible. You can see the full project, La Diez, in our work.
FAQ
Does this replace my POS, inventory system or loyalty program?
No. It connects to the ones you already use; it doesn’t replace them.
Do I need to switch systems to have something like this?
No. What matters isn’t which tools you use, it’s that there’s a way to get data out of them.
How long does a project like this take?
It depends on how many systems need to be integrated, but on average the full project takes 8 to 12 weeks.
Where to start
If your business also has its data spread across systems that don’t talk to each other, this is exactly the kind of project we build as part of our custom AI services. For the bigger picture, read our guide to AI agents for business. Tell us which systems you use and we’ll see how quickly something like this could come together for you.
