Every week a construction site generates paper: delivery notes for materials, supplier quality certificates, foreman's daily reports, subcontractor invoices. Someone has to read it, classify it and enter it into the system before it becomes useful, and that job usually falls on the technical office or admin staff, running behind what is actually happening on site. Summum AI applies automatic document classification and computer vision so that information enters the system on its own and real site progress is visible without waiting for the weekly report.
The problem is not volume, it is format. Every supplier sends its delivery note however it wants — a scanned PDF, a phone photo, a paper slip hand-signed on site — and every material certificate arrives on its own template. The technical office spends hours manually extracting what is already written on the document: material reference, quantity, batch, date, destination site. That repetitive work is exactly what an AI document-classification system does in seconds, with the added benefit that it never gets tired and never drops a field at month-end.
In parallel, physical site progress — foundations poured, slab cast, walls up — is today reported almost entirely by eye: someone visits the site, estimates a percentage and writes it down. A computer-vision system trained on periodic site photos (drone, fixed camera, or simply the foreman's phone) can estimate progress per work item — structure, envelope, installations, finishes — and compare it against the schedule, raising an alert as soon as an item drifts from its planned deadline. It does not replace the official certification of works or the judgement of the site management team: it is a supporting data point, more frequent and more objective than a fortnightly visit, connected to the construction ERP so the deviation shows up where the project is already managed.
This has a regulatory reading worth stating clearly from the design stage. Regulation (EU) 2024/1689 (the AI Act) classifies as high-risk, in Annex III, point 4(b), AI systems intended to be used to monitor and evaluate the performance and behaviour of persons in the context of a work-related relationship. A vision system that measures the physical progress of a structure — concrete poured, brickwork laid, pipework installed — does not evaluate any individual's performance and therefore does not fall into that category; we design it explicitly around the work item, not the worker. That said, if the photographs capture workers in frame, Article 89 of Spain's Organic Law 3/2018 (LOPDGDD) comes into play, requiring employers to inform workers in advance, expressly, clearly and concisely about the use of cameras on site. We build this into the rollout: an information notice and a camera framing that excludes rest areas or changing rooms, exactly as the LOPDGDD requires.
We review which documents arrive each week (delivery notes, certificates, foreman's reports, subcontractor invoices), in what format they arrive, and which fields need to be extracted from each. In parallel, we identify the work items worth tracking by vision — structure, envelope, installations, finishes — and the photo source available (foreman's phone, fixed camera, drone).
We train the recognition on your actual suppliers' and subcontractors' real templates, not generic ones. The system classifies each document by type, extracts the relevant fields and leaves them ready to push into the ERP with no manual intervention except in the cases it flags as uncertain.
We set up periodic photo capture per work item and train the vision model on historical data from similar projects when available, or on the first weeks of the current site. The output is a progress percentage per work item, updated with every batch of photos.
We connect both flows — documents and progress — to your construction ERP (certifications, scheduling, cost control) so the extracted data and the progress percentage appear where your team already works. We configure deadline-deviation alerts per work item at whatever threshold you decide.
The operational detail: what we deliver as part of the work and what we keep alive afterwards.
Automatic classification of delivery notes and certificates
OCR and classification by document type (delivery note, material certificate, foreman's report, subcontractor invoice) with extraction of each one's key fields.
Site-progress dashboard by computer vision
Estimated progress percentage per work item (structure, envelope, installations, finishes) from periodic photos, compared against the schedule.
Deadline-deviation alerts
Automatic notice when a work item falls behind schedule, with the photo history that backs up the estimate.
Integration with the construction ERP
Classified documents and progress percentages pushed into the construction ERP, with no double data entry between systems.
Regulatory framing of the rollout
Assessment of the system's position against Annex III of the AI Act and an information notice to staff under Article 89 of the LOPDGDD before any camera is installed.
Monthly review and retraining
Adjustment of both the classification model and the vision model using the real documents and photos accumulated each month on site.
Document classification and progress vision supply the data; Summum Sistemas' construction ERP is where that data becomes certification, scheduling and cost control. They are deployed as separate, complementary layers, each run by the group brand that owns it.
No. Summum Sistemas' ERP (erp-construccion) manages site control, certifications, Verifactu and subcontractors. This Summum AI page covers the layer before that: automatically classifying incoming documents and estimating physical progress by computer vision. Both layers integrate, but each is deployed and maintained by the group brand that specialises in it.
No. It is a supporting data point, more frequent and more objective than a periodic site visit, but it does not replace the judgement of the site management team or the official certification of works, which follows its usual procedure.
Regulation (EU) 2024/1689 (the AI Act) classifies as high-risk, in Annex III, point 4(b), AI systems intended to monitor and evaluate the performance and behaviour of persons in a work-related relationship. The system we deploy measures the physical progress of work items — structure, envelope, installations — not any individual worker's performance, and it is deliberately designed that way to stay outside that category. The obligations under Article 6 and Annex III of the Regulation, in any case, apply from 2 August 2027.
Article 89 of Spain's Organic Law 3/2018 (LOPDGDD) requires employers to inform workers in advance, expressly, clearly and concisely about the use of cameras at the workplace, and bans installing them in rest areas or changing rooms. We build this into the rollout: an information notice to staff and a defined camera framing before the system goes live.
The ones that generate the most volume on a site: material delivery notes, supplier quality certificates, foreman's daily reports and subcontractor invoices, in whatever format they arrive (scanned PDF, phone photo, paper signed on site).