Burgos packs the largest industrial estate in Castilla y León, and a good share of the region's export industry, into very little ground. Villalonquéjar holds close to 500 companies and more than 13,000 jobs across a little over eight million square metres in four phases, with automotive, food and chemicals as the heaviest blocks (source: Asociación de Empresarios del Polígono de Villalonquéjar, consulted July 2026). Orbiting the large plants is a layer of suppliers, engineering firms and service companies working to tight deadlines, with highly specific terminology and an admin load that keeps growing: requests for quotation, part approvals, technical datasheets, commercial follow-up in several languages. That is where applied artificial intelligence pays off first, rather than in a laboratory project. Summum IA deploys sales and marketing process automation with n8n and language models across Burgos, along with RAG systems over internal documentation, back-office and voice agents, and the document processing that feeds the ERP. The group has a presence in Burgos as a service-area office, with no published street address and its head office in Valladolid: the work is done remotely, with occasional visits when the project warrants them.
The urban area of Burgos lives off its factories. According to El Correo de Burgos (3 February 2025), reporting on Gonzalo Andrés's study for Fundación Caja de Burgos, 42 plants account for 68% of industrial employment in the urban area, which totals 20,533 people, with Campofrío, Bridgestone, Antolín and Benteler among the largest. That concentration explains the real commercial problem facing a Burgos SME: it sells to a handful of demanding customers, through supplier portals, with requests for quotation arriving by email, specifications attached as PDFs and short response windows. Automating that circuit with n8n and a language model as the decision node (read the request, pull out part numbers and quantities, find the precedent in past quotes, draft the offer and flag it to whoever signs it) is cheap to start and easy to measure. The same applies to marketing: segmenting the contact database, producing the product sheet in the customer's language and keeping the CRM current without manual work.
The second trait of Burgos is that it sells abroad. The province exported 4,344.4 million euros in 2025, up 9.26% year on year and the highest figure in its historical series, with Germany as the leading destination (885.2 million), chemical products in front and machinery growing 19% (source: BURGOSconecta, 2 April 2026). At regional level, the Council of Chambers of Commerce analysis records a 7.6% fall in the number of exporting companies in Castilla y León between 2023 and 2025, from 5,706 to 5,271 (source: Cámara de Comercio de Burgos, 26 May 2026). Selling more with a thinner commercial structure means leaning on tooling. A RAG system over manuals, certificates and correspondence lets you answer a German customer citing the exact document instead of reconstructing it from memory; a voice or chat agent handles the first enquiry in several languages and hands the case over ready for the person who closes it.
Burgos is not starting from scratch on technology. ITCL, a technology centre based in the Villalonquéjar estate itself, closed 2025 with 108 staff, more than 207 clients and 24% revenue growth, and works on FPGA architectures and neuromorphic computing (source: El Correo de Burgos, 8 June 2026); the Higher Polytechnic School of the University of Burgos feeds technical profiles into that same environment. What is still missing from many deployments is the regulatory side. Any agent that talks to a person, by voice or by chat, must comply with Article 50 of Regulation (EU) 2024/1689 (AI Act) from 2 August 2026: stating unambiguously that it is an AI system and technically marking the content it generates. Summum IA delivers that technical layer already in place and leaves the system's regulatory risk assessment to Summum Consultoría, which signs off on that part across the group.
We walk through with sales, quality and admin the actual path a request for quotation or an order takes from arrival to invoicing, noting where a value gets copied by hand, where everything waits on someone to reply and where a document goes missing. That produces a short list of candidate processes, ranked by hours recovered and by tolerance for error, which is what decides what gets automated first and what is better left alone.
We build the first flow in an isolated environment: the triggers, the connectors to the tools you already use, the exact point where the language model classifies or drafts, the output validation that runs before anything executes, and the escalation path to a person. What the flow is allowed to decide on its own, and what always waits for a signature, is agreed in writing before anything gets wired. The prototype is tested against your own real emails and documents, not demo examples, before it goes anywhere near production.
We connect the flow to the ERP, the CRM or the document manager through APIs or the MCP protocol, without asking for system migrations. When the process needs to consult in-house knowledge (manuals, part approvals, terms agreed with a specific customer), we layer the RAG system on top and tune the sector terminology, which in automotive, food and chemicals looks nothing alike from one case to the next. None of this forces you to change ERP or rebuild the document manager already in place.
Before the flow faces the outside world we put the Article 50 mechanisms of Regulation (EU) 2024/1689 in place: notice that the person is talking to an AI system, marking of generated content and an auditable record of conversations. Then we measure which cases resolve on their own, which escalate to a person and for what reason. Those figures drive the adjustments to the existing flow and the choice of the next one worth building.
The operational detail of a Burgos project: what we hand over and what we keep alive afterwards.
Commercial and marketing circuit diagnosis
A map of how an opportunity arrives and gets handled today: forms, email, supplier portals, requests for quotation and follow-up. We finish with a backlog ranked by hours saved and by risk of error, not by whatever looks most impressive in a meeting.
Automation with n8n and language models
Flows installed on your own infrastructure that read the incoming request, extract the data, check past cases, draft the reply or the product sheet in the customer's language, and escalate to a person the moment the case falls outside what was foreseen.
RAG over technical, quality and product documentation
A conversational search layer over the company's manuals, certificates, part approvals and commercial correspondence, with every answer citing the source document and page. It earns its keep above all when whoever wrote the original has left the company.
Back-office and voice agents
Assistants that take the first enquiry by phone or chat, check the status of an order or a shipment in the ERP, and either resolve the case or hand it over ready for the person who closes it, in the language the customer used to ask.
Document processing and classification
Automatic reading of delivery notes, invoices, material certificates and datasheets arriving by email or shared folder, extracting the key fields and pushing them into the ERP or the document manager, with no manual capture in between.
AI Act transparency and team training
AI-interaction notice, technical marking of generated content and conversation logs for any conversational component of the project, plus training for the team that will work alongside the tool, eligible for FUNDAE funding in Spain, with the material handed over so it can be reused internally.
An AI project at a Burgos company touches two distinct layers. Summum IA builds the technical part: the n8n flows, the RAG system, the agents, the ERP integration and the Article 50 transparency mechanisms. Summum Consultoría handles the regulatory layer: classifying the system under the AI Act, determining which obligations apply and documenting the data processing. Each layer is signed by whoever answers for it, and the split is written down from the start.
Classic email marketing sends a fixed sequence to a list. Automating with artificial intelligence is a different thing: an n8n flow watches what actually happens (a form, a request for quotation arriving by email, a catalogue download), a language model reads that material and decides what to do with it, and the system acts accordingly, classifying the contact, drafting the reply in the customer's language, updating the CRM and alerting whoever needs to call. The machine prepares the text, the commercial decision stays with a person, and every step is logged so it can be reviewed later.
Yes, though the work moves elsewhere. When the customer base is small and demanding, the cost is not in winning business but in responding: requests for quotation with specifications attached, supplier portal forms, part approval questionnaires, quality claims, periodic reports. All of that can be read, extracted and prepared automatically, leaving the person the part that needs judgement, which is the price, the delivery commitment and the exception. With a short customer list the return comes from response time and from documentation errors avoided, not from new lead volume.
No. The group has a presence in Burgos as a service-area office, with no published street address, and its head office is in Valladolid. Diagnosis, building the flows, testing and follow-up are all done remotely, by video call and with whatever system access is agreed. We keep travel to Burgos for the milestones that justify it: the initial discovery session with the team, an in-person training session or the go-live. To discuss a project, the group's phone number is +34 983 515 290.
Exactly as it affects one serving a customer in Burgos. Article 50 of Regulation (EU) 2024/1689 requires, from 2 August 2026, clear disclosure that the party on the other side is an AI system and technical marking of any synthetic content it generates. It applies across the whole European Union regardless of where the company is established or where the customer sits, so a Burgos manufacturer exporting to Germany carries exactly the same obligations. Summum IA delivers that technical layer ready; the system's regulatory risk assessment is signed by Summum Consultoría.
It depends on the design, and that is decided before anything is built. n8n is installed on your own infrastructure, so the flows and the credentials stay inside. What does leave, if configured that way, are the calls to the language model, which is why we define from the start which information travels and which does not. For sensitive documentation (drawings, cost breakdowns, agreed commercial terms) we evaluate locally deployed models or dedicated infrastructure. The access policy and the legal basis for processing are coordinated with Summum Consultoría before go-live.