Deep Reinforcement Learning-Based Self-Learning Model for Adaptive Marketing Strategy Formulation

Marketing strategy formulation is a sequential decision problem: each campaign action changes the audience state, platform competition landscape, and brand state in ways that affect future reward-a dynamic that supervised learning's static policy and A/B testing's slow convergence structurally cannot exploit. This paper applies a Proximal Policy Optimization (PPO) Deep Reinforcement Learning agent to adaptive marketing strategy formulation across 36 companies, 4,800,000 customer interactions, and 240,000 campaign decisions (180,000 training/60,000 test, 75/25 chronological split, 24-month horizon). The DRL agent achieves composite marketing performance r=0.864 vs. expert benchmarks, outperforming supervised ML (0.764; +13.1% [(0.864-0.764)/0.764]), A/B testing (0.684; +26.3%), and rule-based marketing (0.524; +64.9% [(0.864-0.524)/0.524]). Deployed outcomes include +34.2% conversion rate, −28.4% customer acquisition cost, and +48.6% campaign ROI vs. supervised ML baseline. Sub-channel performance: Digital Display$(\mathrm{r}=0.904)$, Email Personalisation (0.884), Social Media (0.844), Search/SEM (0.784). Cold-start degrades performance to$\mathrm{r}=0.724(-16.2 \%)$and high market volatility to 0.784 (−9.3%), each with a deployed mitigation.

Hali tarjima qilinmagan