Digitalising Is Not Digitalising for Sustainability: Decomposing the Digital Transformation–ESG Relationship in African Listed Firms

Purpose and problem. Digital transformation is widely promoted as a lever for corporate environmental, social and governance (ESG) performance, yet the empirical record is contradictory. An influential strand of that literature attributes the inconsistency to nonlinearity, proposing that digitalisation depresses ESG performance before improving it. This study tests that proposition directly and, finding no support for it, offers an alternative explanation grounded in what firms actually do with digital capability. Design and methods. We analyse a balanced panel of 80 firms listed on the Johannesburg, Nigerian, Nairobi and Ghana exchanges over 2015–2024 (800 firm-year observations). We estimate two-way fixed-effects models and apply the Lind–Mehlum test for U-shaped relationships and a bootstrapped Hansen threshold search, the two procedures the nonlinearity claim requires but which prior work in this area has largely omitted. We then decompose digital transformation into digital intensity and the orthogonal component of digital capability that is purposively directed at sustainability processes, and validate the decomposition using difference GMM, quantile regression, instrumental variables, subsample exclusions and lagged specifications. Findings. The nonlinear hypothesis is rejected on both tests (U-test p = 0.320; threshold bootstrap p = 0.707). Digital intensity on its own bears no relationship to ESG performance (β = 0.029, p = 0.952). Sustainability-purposed digital integration, by contrast, is robustly and positively associated with ESG performance (β = 4.494, p < 0.001), and holding it constant reveals a positive intensity effect (β = 2.556, p < 0.001) that the aggregate measure conceals. Climate risk attenuates rather than amplifies the intensity pathway (β = −0.360, p = 0.025); institutional quality and state ownership do not moderate either pathway. Effects hold across all three ESG pillars and the whole ESG distribution, but do not survive a one-year lag. Value. The contradictory findings in this literature are better explained by measurement aggregation than by nonlinearity: composite digital transformation indices mix a component that moves ESG performance with one that does not, so the estimated effect depends on the mix. We report a null on the headline construct, three refuted moderation hypotheses and one robust positive result, and we disclose openly that the analysis rests on a partly synthetic dataset, which bounds the inferential claims accordingly.