DareData

The NOS work is the anchor: a GenOS-orchestrated invoice pipeline that now handles around 65% of roughly 20,000 supplier invoices a month end to end, matching them to SAP purchase orders and routing the ambiguous rest to a 20-person review team, plus UVA, a retrieval assistant over 15,000-plus documentation pages that operators rate at 87% answer accuracy on the 20% of calls where it is used.

Outside the telco, the Heineken routing engine (an OSRM duration matrix plus constraint optimization) reports €100k in fuel savings and 10% more locations visited per day, and the COFICAB cable-design platform runs surrogate models and an optimizer over 50,000-plus geometry and material combinations, with the engineers on each study named.

Two things to weigh: NOS acquired 20% of DareData, so the four NOS studies are partly work for a shareholder, and the COFICAB and UVA pages both label their outcomes as still under measurement, so the hard numbers sit mostly in the invoice and routing work.

The weaker fit is buyers who want a model-level technical narrative—the studies name integrations, volumes and team members, not the language models or evaluation setup behind GenOS.