Artificial Intelligence-Driven Personalization in Digital Marketing: Enhancing Consumer Engagement and Purchase Intent
Year: 2026
Type: conference-paper
Abstract: Digital marketing personalisation has transitioned from competitive advantage to consumer expectation, yet most deployed systems remain anchored to rule-based segmentation heuristics that lack the granularity and adaptivity required to capitalise on individual behavioural trajectories across multiple channels. This paper proposes an Artificial Intelligence-Driven Personalisation Framework (APF) integrating a sequential deep-learning behavioural model, a real-time contextual bandit content-selection layer, and a cross-channel attribution engine, evaluated across 42 digital marketing campaigns from 14 enterprises (e-commerce n = 6, retail$\mathbf{n}=\mathbf{5}$, fintech$\mathbf{n}=\mathbf{3}$) encompassing 284,000 consumer interaction sessions (198,800 training / 85,200 test, 70/30 chronological split) and 18,600,000 raw behavioural data points over a 24-month evaluation horizon. The APF achieves a composite personalisation-effectiveness score of$\mathbf{r}=\mathbf{0. 8 7 2}$against expert-curated consumer-outcome benchmarks, outperforming an ML-only recommendation engine ($\mathbf{r}=\mathbf{0. 7 4 8}$; +16.6% [(0.872-0.748)/0.748]), rule-based segmentation$(\mathbf{r}=\mathbf{0. 6 3 4}; +\mathbf{3 7. 5 \%}[(\mathbf{0. 8 7 2 - 0. 6 3 4}) / \mathbf{0. 6 3 4}])$, and generic mass-marketing$\quad(\mathbf{r}=\mathbf{0. 5 1 2}; \quad+\mathbf{7 0. 3} \% \quad[(\mathbf{0. 8 7 2 - 0. 5 1 2}) / \mathbf{0. 5 1 2}])$. Deployment outcomes across the enterprise cohort include +42.8% consumer engagement rate, +57.4% click-through rate (CTR), +38.6% purchase-intent score, and -34.8% cartabandonment rate relative to pre-deployment baselines. Subdimension performance: Behavioural Targeting$(\mathbf{r}=\mathbf{0. 9 1 8})$, Real-Time Content Adaptation (0.879), Purchase-Intent Prediction (0.847), and Cross-Channel Personalisation (0.808). Sparse-interaction consumers (< 20 logged behavioural signals) degrade personalisation accuracy to r = 0.791 (-9.3% [(0.872-0.791)/0.872]), and cold-start new users to 0.726 (-16.7% [(0.872-0.726)/0.872]), with deployed mitigation strategies documented in Section VI.
Language: English
Subfield: Marketing
Domain: Social Sciences
Open Access status: gold
OpenAlex ID: W7225834783