Artificial intelligence (AI), machine learning (ML), and the Industrial Internet of Things (IIoT) are reshaping maintenance from failure reaction to decision-centric value creation. This chapter synthesizes the state of the art into a practitioner-oriented framework that connects sensing, analytics, and execution across strategic, tactical, and operational levels. We propose a task-driven taxonomy spanning diagnostics, prognostics (RUL with quantified uncertainty), prescriptive optimization, and multi-site orchestration, mapped to data modalities (tabular/events, time-series, images, text, and physics-informed digital twins) and governance constraints (explainability, latency, privacy, maturity). We consolidate enabling architectures—edge/cloud pipelines, data fabric and semantics, MLOps, drift management, XAI, and federated learning—and outline when to prefer classical models, deep learning, probabilistic/survival methods, hybrid physics-informed models, or reinforcement learning. Building on reported industrial impacts, we present a staged roadmap for first deployments with indicative effort, skills, and KPIs, emphasizing how to translate predictions into scheduling, inventory, and risk-based decisions. Finally, we discuss emerging agentic/LLM copilots and Industry 5.0 servitization, highlighting their implications for safety, resilience, and sustainability. The result is a concise guide to select models, design trustworthy pipelines, and scale maintenance optimization with measurable business outcomes.