SAS Institute

Contribution to AI

Analysis stopped being a sequence of one-off runs and became a program. The system that grew out of agricultural field-trial work at North Carolina State — funded first by the National Institutes of Health, then by a consortium of southern experiment-station statisticians after that grant lapsed — divided the work into a DATA step that read observations one at a time and PROC steps that did the statistics. Because records passed through memory singly, files far larger than a machine's core could be cleaned, merged, reshaped, modelled and reported by one script that could be rerun, corrected and handed to somebody else. That script is the ancestor of the analysis pipeline everyone now writes in Python. The 1985 rewrite out of IBM's PL/I into C carried the arrangement off the mainframe and onto everything else, and the transport file format published alongside it was still, decades later, the layout in which the FDA required clinical trial datasets to arrive — a vendor's internal record structure hardened into regulation. Selling by annual renewal rather than one-time purchase established that analytical software was a service kept current, an arrangement the whole industry later adopted. By 1998 Enterprise Miner arranged predictive modelling as a sequence of connected nodes — sample, explore, modify, model, assess — which put decision trees, clustering and neural networks in front of analysts who would never write the algorithms themselves.

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