The AI landscape doesn't move in one direction — it lurches. Some techniques leap from experiment to table stakes in a single quarter; others stall against regulatory walls, technical ceilings, or organisational inertia that no amount of hype can dislodge. Knowing which is which is the hard part. The State of Play cuts through the noise with a rigorously maintained index of AI techniques across every major business domain — classified by maturity, evidenced by real-world adoption, and updated daily so you always know where you stand relative to the field. Stop guessing. Start knowing.
A daily newsletter distilling the past two weeks of movement in a domain or two — delivered to your inbox while the index updates in the background.
Each dot marks the weighted maturity of practices within a domain — hover for a brief summary, click for more detail
AI that monitors equipment health through vibration, acoustic, thermal, and visual sensors and alerts on anomalies. Includes multi-sensor fusion and threshold-based alerting; distinct from remaining useful life estimation which predicts future failure rather than detecting current conditions. Scope covers ML-based anomaly detection and sensor fusion; traditional threshold alarms and statistical process control without ML are out of scope.
ML-based condition monitoring is a solved technology problem with an unsolved adoption problem. Multiple GA platforms from major vendors deliver proven ROI -- 30-50% downtime reduction, 40% maintenance cost savings, payback within 18 months -- for organisations with the digital maturity to absorb them. The practice uses vibration, acoustic, thermal, and visual sensors to detect equipment degradation in real time, distinct from remaining-useful-life estimation, which predicts future failure rather than flagging current conditions.
The vendor ecosystem is consolidated, the analyst coverage is deep, and independent case studies span steel, energy, automotive, and manufacturing with consistent positive outcomes. That places this firmly in good-practice territory: the question is how to roll it out, not whether it works. Yet only about 22.5% of organisations report their programmes as "effective" -- a figure unchanged since 2023 -- and 60-80% of implementations underperform or stall within two years. The bottleneck is organisational, not technical: data quality, workflow integration, technician adoption, and change management determine success far more than algorithm sophistication. A sharp bifurcation persists between digitally mature sectors achieving production-scale returns and commodity industries still blocked by infrastructure and readiness gaps.
Three platform-scale vendors dominate: Siemens Senseye (new wins including Octapharma; 80% failure forecast accuracy in automotive welding), GE Vernova (SOCAR Turkiye achieving 20% reactive maintenance reduction; deployments at Xcel Energy and Sasol; May 2026 reports covering 350+ equipment types with $1.6B cumulative customer losses avoided and 3.41-month average ROI), and AWS IoT SiteWise with native multivariate anomaly detection. AWS discontinued standalone Lookout for Equipment (October 2026 end-of-life), signaling that even GA production services lack sufficient adoption momentum for standalone viability. Market projections remain aggressive -- USD 70.7B by 2032 at 26.5% CAGR -- and named deployments at BlueScope Steel (1,950 avoided downtime hours), Cepsa refineries, Nissan (10,000+ assets), and a North American refinery (USD 1.89M losses prevented) show 40-50% cost reductions at digitally mature sites. Real-world case studies validate deployment value: coal plant avoided $1.84M emergency repair (19-day early detection); 310 MW hydroelectric station avoided $2.2-3.1M generator replacement via partial discharge detection (94% accuracy, 67% false-positive reduction).
Adoption is accelerating among digitally mature organizations but remains blocked at mainstream entry. May 2026 independent survey (Fluke, 600+ manufacturers) shows predictive maintenance adoption doubled from 9% to 18% year-over-year, confirming market growth momentum but revealing critical barriers: 78% of respondents cite skills gaps as primary implementation obstacle, not technology maturity. A German practitioner assessment (TeepTrak, April 2026) identifies three reliably working use cases with documented ROI: vibration-based bearing prediction (80-90% accuracy, 10-30 day lead time, 8-18 month payback), motor current signature analysis (70-85% accuracy), and process drift detection (4-10 day lead time). However, critical limitations persist: 15-25% of failures remain unpredictable, training data requirements are substantial (6-18 months of historical operation), and multi-cause failure classification remains difficult. For SMEs achieving adoption, 3-year ROI ranges from 3-6x with 20-35% downtime reduction and 150-400k€ annual benefit.
The broader sector reveals bifurcation: industry benchmarking with major operators (Saudi Aramco, Equinor, Rio Tinto) documents 30-40% cost reductions where governance maturity is high, yet 60-80% of implementations underperform or discontinue within two years. Real-world failures are data engineering, operational integration, and model-drift management problems—not algorithm sophistication. Implementation timelines of two to six months and integration complexity continue to gate mainstream entry. Peer-reviewed research in aviation confirms deep learning dominance in prognostic algorithms but highlights deployment constraints: data heterogeneity, explainability requirements, and regulatory certification prevent transition from experimental models to operational use. Edge-based acoustic anomaly detection has achieved 91.80% accuracy on industrial motor signals, while multi-modal sensor fusion architectures are producing 94%+ fault detection with 30-90 day advance warning in production settings.
— Consulting ROI analysis: automotive production avoids $2.3M per hour of downtime; multi-sensor detection (vibration, thermal, acoustic) identifies degradation 30-90 days before failure; typical 3-year ROI of 3-6x with 8-18 month payback.
— Named deployments: coal plant avoided $1.84M emergency repair (19-day advance detection); 310 MW hydroelectric station detected partial discharge degradation, avoided $2.2-3.1M replacement cost; 94% detection accuracy with 67% false-positive reduction vs threshold-based systems.
— GA platform covering 350+ equipment types with AI/ML-driven anomaly detection; vendor reports $1.6B cumulative customer losses avoided and 3.41-month average ROI from production deployments.
— Independent survey of 600+ manufacturers showing predictive maintenance adoption doubled from 9% to 18% YoY; signals growth but highlights skills gaps (78% cite lack of expertise) as primary implementation barrier.
— Peer-reviewed synthesis of 20 aviation studies: deep learning dominates prognostic methodologies, but deployment remains constrained by data heterogeneity, explainability, and regulatory certification requirements.
— Industry benchmarking with named major operators (Saudi Aramco, Equinor, Rio Tinto) showing predictive maintenance delivers 30-40% cost reductions and 30-50% downtime reduction; highlights governance maturity as critical success factor beyond technology.
— Peer-reviewed edge-based CNN for acoustic anomaly detection on industrial motors achieves 91.80% accuracy; demonstrates practical condition monitoring at equipment level without manual threshold configuration.
— Practitioner assessment of German SME context: identifies three mature use cases (vibration bearing 80-90%, motor current analysis 70-85%, process drift detection) with realistic ROI (20-35% downtime, 8-18 month payback) and hard limits (15-25% unpredictable failures).