NVIDIA's New Model Sharpens Predictions on Tabular Data

NVIDIA introduced Kumo Tabular, a model built to make predictions from tabular data — the kind stored in databases, CRMs and corporate spreadsheets. According to its developers, the model sets a new balance between prediction accuracy and computing cost, outperforming earlier approaches.

Tabular data is the backbone of everyday business: stock levels, deal stages, transaction histories. When a model predicts such data more accurately while using fewer resources, tasks like demand forecasting, credit-risk scoring or spotting suspicious transactions get faster and cheaper — without needing expensive infrastructure.

Check which of your recurring forecasts and reviews across 1C, CRM or spreadsheets still run on manual guesswork. A similar principle already runs inside 1Cifer: agents read data from every connected system at once and flag deviations a person would likely miss.