Machine learning (ML)

Machine learning (ML) is the family of techniques in which algorithms learn patterns from historical data instead of following rules a programmer wrote by hand. Feed a model thousands of labeled purchase order lines and it learns to classify new ones; feed it years of price history and it learns to flag quotes that break the pattern. ML supplies most of the working intelligence behind procurement AI.

Examples

Spend classification: A manufacturer trains a classifier on 60,000 historically coded PO lines. New lines auto-code at 93% accuracy; the 7% below the confidence threshold queue for human review, which takes one analyst about two hours a week instead of a quarterly cleanup project.

Anomaly detection: A model learns that a stamped bracket's price tracks steel coil with a six-week lag. When a renewal quote lands at $3.85 against a predicted $3.38 to $3.52, the buyer digs in and finds a surcharge from a 2024 steel spike that was never removed.

Silent drift: A lead-time model trained on pre-disruption history keeps predicting 8-week deliveries while actuals stretch to 15. Monthly accuracy tracking catches the drift; retraining on recent receipts fixes it.

Definition

The contrast with rules-based software matters in practice. A rules engine for spend classification needs someone to maintain keyword lists (if the description contains BRG, code to bearings), and it breaks on every new phrasing. An ML classifier trained on historically coded lines generalizes to descriptions it has never seen, and its accuracy can be measured on held-out data rather than asserted.

Within AI broadly, ML is the learning machinery. Procurement applications cluster in three areas: classification (coding spend, matching duplicate supplier records), prediction (price and lead-time forecasting and supplier delivery risk, the engine room of predictive analytics), and anomaly detection (a quote or invoice that deviates from learned patterns).

The limits are unglamorous. Models need enough clean, labeled history; a category with 40 historical orders will not train anything useful. They learn the past, so structural breaks (a new tariff, a supplier change, a redesign) degrade them silently until retrained. And confidence varies by prediction, so good implementations route low-confidence cases to a human queue instead of forcing an answer.

*GARTNER is a registered trademark and service mark of Gartner, Inc. and/or its affiliates in the U.S. and internationally, and COOL VENDORS is a registered trademark of Gartner, Inc. and/or its affiliates and are used herein with permission. All rights reserved. Gartner does not endorse any vendor, product or service depicted in its research publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner research publications consist of the opinions of Gartner’s research organization and should not be construed as statements of fact. Gartner disclaims all warranties, expressed or implied, with respect to this research, including any warranties of merchantability or fitness for a particular purpose.