SAS Institute (Expansion)

Contribution to AI

Preparing data and analysing it were separate jobs done in separate tools until the DATA step put both in one language. Reading records, merging files, reshaping and subsetting all sat alongside the procedures that fitted the models, so the path from raw transaction logs to a fitted regression became a single script that could be rerun, audited and handed to a colleague. By the mid-1980s that script was portable too: the system was rewritten in C, and the Multivendor Architecture that followed meant code written for an IBM mainframe ran on a VAX, a Unix machine or DOS with the same syntax and the same answers. Version 5's macro language, its full-screen Display Manager and SAS/AF for building interfaces over procedures let statisticians hand working applications to people who were not statisticians. The consequences were institutional rather than intellectual. Banks, insurers and drug companies built their reporting and their risk models on one platform, and clinical trial data reached American regulators as SAS transport files, which quietly made one vendor's file format a public standard. Enterprise Miner in 1998 placed decision trees, clustering and neural networks behind a common interface under a named workflow — sample, explore, modify, model, assess — so that predictive modelling became a documented business procedure with a defined sequence of steps, roughly a decade before anyone was calling the same activity data science.

Drafted with AI and edited by hand (claude-opus-5, reviewed 2026-08).