Predicting customer conversion in multi-channel digital marketing: an interpretable modelling approach

Abstract

Digital channels now account for roughly three-quarters of global advertising expenditure, yet marketers still lack clear evidence on which campaign attributes actually drive customer conversion across channels. This study examines the determinants of conversion using a customer-level dataset of 8000 records containing demographic, campaign, and engagement attributes, supported by a campaign-level dataset of 300,000 records spanning four social platforms. A binary logistic regression model was fitted with class weighting to address an 87.6%/12.4% outcome imbalance, and benchmarked against random forest and gradient boosting classifiers on an identical train–test split. Three complementary analyses of the same dataset point in a consistent direction. Chi-square testing found campaign type strongly associated with conversion (χ2 = 84.454, p < 0.001), while campaign channel was not (χ2 = 2.785, p = 0.594). Logistic regression identified a conversion-oriented campaign objective as the strongest driver (odds ratio 2.078), followed by engagement signals click-through rate, time on site, and email interaction (odds ratios 1.47–1.57), while demographic attributes were near-neutral (age 0.982, income 1.048). Quartile analysis further revealed diminishing returns to advertising spend, with conversion plateauing above the third spend quartile. Because 87.6% of customers converted, the no-information accuracy of always predicting conversion is 0.876; once all classifiers are equally class-weighted, none exceeds this baseline on accuracy (logistic regression 0.754, gradient boosting 0.829), so accuracy is uninformative here. The models are therefore compared on discrimination of the minority non-converting class, where gradient boosting is strongest (average precision 0.63 versus 0.47 for logistic regression, against a no-skill baseline of 0.12). The value of the logistic model lies in interpretability rather than accuracy. The findings indicate that campaign intent and customer engagement, rather than channel selection, are the higher-value levers in multi-channel marketing.