Heterogeneous customer requirements and competition push enterprises to produce individualized products, increasing variant production and quality assurance demands. 100% inspection of all product characteristics is standard in variant production but expensive. Sampling inspection and key characteristics are a solution, but statistical knowledge and implementation effort hinder its application in variant production. This thesis develops an algorithmic procedure for applying sampling inspection in variant production that reduces effort and requires little statistical knowledge. It considers these requirements of enterprises producing variants by automatically defining key characteristics and reducing the need for 100% inspection. The procedure includes: (1) a model describing elements and sampling plans of the procedure, (2) algorithms grouping redundant characteristics and selecting key characteristics, (3) algorithms defining similar characteristics across variants and enabling sampling inspection, and (4) risk-based performance measures. The adaptive inspection planning procedure for variants automatically evaluates historical data. The procedure is applied during the product production and not its development phase to ensure the availability of historical data. Machine Learning and production engineering concepts are used to automate the procedure. The procedure includes easily interpreted visualizations of the results and performance measures to aid the transition from 100% inspection to sampling inspection. Simulations with artificial data validate the adaptive sampling inspection procedure for variant production, and a benchmark identifies the best algorithms for key characteristics, mixed lots, and performance measures. The top algorithms for identifying key characteristics are based on pairwise clustering, SPFA, risk analysis, and process capability analysis. For identifying mixed lots, pairwise clustering with a self-updating distance matrix and hierarchical clustering are best. Accuracy is the most suitable performance measure. The effects of various factors on the algorithms were evaluated, and suitable algorithm specifications were determined. The procedure and algorithms were tested on data from four enterprises and successfully identified key characteristics and grouped characteristics for sampling plans, reducing inspection costs with manageable risk. The procedure is recommended for variant production.