Custom software
with AI

It's a program that uses a language model to understand a request and then, with the permissions you give it, does something in your systems: creates the customer record in the CRM, checks stock in the ERP, prepares a quote or sends an email. Where a traditional chatbot would give you a link or a phone number, an agent finishes the task. That's why it's built with clear limits: • It only touches what you authorise; for example, it can read orders without being able to delete them • Every action is logged with its date and result • Anything it can't solve goes to someone on your team A real example: in AgentPrice, the system we built for Electrotodo, the AI compares prices with competitors and proposes a new price, but nothing changes in Shopify until a person approves it.

It depends on what you have. We mostly come across three situations: 1. The software has an API, like Shopify, PrestaShop, Microsoft 365 or most modern CRMs. We connect through it. 2. There's no API, but there is a database. We read from it in read-only mode and never write to its tables. That's how we work with Navarro Hermanos' ERP. 3. Neither: legacy software, spreadsheets or documents that arrive by email. Then we use scheduled exports or read the documents with OCR, as we do for Gesttia. Your team keeps using the software it knows. What goes away is copying data from one to the other.

Here's what we actually do, so you can judge for yourself: • As a rule, each client has its own server and its own database. We don't mix two companies' data on the same machine. • Backups run daily, encrypted and stored off the server in a European Union data centre, with a 30-day deletion lock. • For client projects we normally use OpenAI models through Azure. Under Microsoft's terms, the data sent is not used to train models. • If your data can't leave your network, the system is installed on your own servers. Navarro Hermanos, for example, runs it inside its own network. We sign the data processing agreement required by the GDPR and, if you need one, a non-disclosure agreement.

We don't give a general figure, because it depends on how many hours that task takes today. What we do is work it out with you before starting, using your numbers: how many people do it, how long it takes them and what a mistake costs. A sample calculation: if three people spend two hours a day copying orders from the store into the ERP, that's around 130 hours a month. With that figure and the quote side by side, you can quickly see how many months the project takes to pay for itself. If the numbers don't work, we tell you and we don't do it.

We build and maintain what we propose. At the end you have a working system, and we're still around when it needs changing. A few things you can check: • The case studies on this page name the company, and almost all of them are still in production. • We have three products of our own (AgentBot, AgentDesk and AgentSearch) running in real companies. • Juan Luque, who runs Nexttia, was in charge of e-commerce, logistics and customer service at Perfumerías Primor from 2011 to 2023. • We're based at the Andalusia Technology Park and we're a Shopify Partner.

Yes, and it's usually what we recommend. 1. Free assessment: we go through your processes with you and tell you where we'd start. 2. A first, tightly scoped project with a result you can measure. For example, getting quote requests that arrive on WhatsApp into the ERP automatically. 3. If it works, we extend it to other processes. If it doesn't, we stop there. You don't need to commit to 12 months or approve a huge budget to get started.