The first analysis of a latent fingermark, carried out when a trace is recovered at a crime scene, decides whether the mark is searched against the national database, kept only for exclusion, or set aside. The marks that reach this stage are often partial, faint, or deposited on difficult substrates, so the analysis is slow and varies between operators. This paper reports the operational validation and national deployment of MinutIA, an AI-based decision-support tool that detects minutiae and helps assess fingermark usability. MinutIA pairs a YOLOv8 detector for minutiae localisation with a Detectron2 segmentation module for anatomical zone assessment, and it was built to assist criminal identification technicians and fingerprint examiners, not to replace them. Because a degraded mark has no single correct number of minutiae, the tool was judged against qualified human reviewers rather than against an idealised answer. Validation used a sequential crossed review defined by a 2023 national instruction, in which field technicians (CIC), a quality-control reviewer, (SPCQ), National Automated Fingerprint Database (FAED) specialists (DFAED), and an expert unit (DED) annotated the same marks. On 689 reviews of 390 traces by 36 examiners, the detector reached a pooled precision of 0.77, recall of 0.88, and F1 of 0.82 against reviewer-corrected references. A formal equivalence test shows that MinutIA’s disagreement with the quality-control reference is no larger than the disagreement the CIC and DFAED tiers show against that same reference, at a margin of four minutiae. A separate analysis of the national usage log, covering 29,206 valid calls from 177 units over eleven months, shows a single drop in active units after the first month followed by stable use, with volume concentrated in a core of heavy users. The results support MinutIA as a credible decision-support tool and locate the main barrier to wider use in onboarding rather than sustained dissatisfaction.