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 models that estimate the remaining useful life of components and systems to optimise replacement timing. Includes degradation curve modelling and multi-sensor fusion; distinct from condition monitoring which detects current anomalies rather than predicting remaining life. Scope covers ML-based predictive models; traditional statistical reliability methods (e.g. Weibull analysis) without ML are out of scope.
ML-based remaining useful life estimation has solved the technology problem. It has not solved the adoption problem. Forward-leaning manufacturers -- BlueScope Steel, Nissan, Schaeffler -- report 40-50% reductions in unplanned downtime and payback periods under 18 months, proving the practice delivers where conditions align: high-volume production, standardised equipment, and adequate sensor data. Yet the gap between vendor claims and median deployment reality remains stark: 60% of manufacturers report ≥26% downtime reduction, but many lack sufficient failure data, sensor coverage of failure modes that matter, or flexible maintenance workflows to respond to model outputs. Data engineering consumes roughly 70% of project effort, work management processes lag behind model capabilities, and mid-market firms with heterogeneous equipment struggle to replicate Fortune 500 outcomes. Recent research advances (physics-informed hybrid models, graph attention networks, asymmetric loss functions for safety-critical domains, domain adaptation for equipment transfer) continue to address methodological gaps in data scarcity, early-life prediction, and model generalization, indicating the field is maturing toward production robustness. Siemens' 2022 acquisition of Senseye consolidated vendor leadership, while AWS's withdrawal of Lookout for Equipment in 2024 narrowed the platform landscape further. However, a persistent manufacturing-wide adoption barrier remains: 76.4% of industrial AI projects fail to deliver intended business value, with data quality and model reliability requirements exceeding consumer AI standards. Deployed systems face operational challenges including silent model degradation (91% of production models degrade over time), with critical failure not typically visible until the 90-day mark, requiring robust drift detection instrumentation. The result is a practice where the vanguard has proven ROI decisively, but most organisations have not yet started -- making RUL estimation a textbook leading-edge capability whose next constraint is execution and operational monitoring, not invention.
Siemens Senseye is the dominant platform, claiming 50% downtime reduction, 55% productivity gains, and 40% cost savings across enterprise clients. BlueScope Steel's multi-year rollout -- 2,000+ hours of prevented downtime and 53 avoided process interruptions -- remains the most thoroughly documented deployment. Specialist entrants like Novity offer hybrid physics-ML alternatives, but the vendor field has narrowed significantly: AWS formally discontinues Lookout for Equipment on October 7, 2026, marking the end of a major managed RUL platform despite active Toyota and Koch Industries deployments, narrowing off-the-shelf options to Siemens-dominated software platforms. Toyota's AWS-based IoT deployment across global automotive manufacturing and Siemens Energy's 18-factory rollout (30 custom RUL use cases) demonstrate enterprise-scale adoption. Recent audits document broad adoption across 217 Tier-1 auto suppliers and 89 pharma contract manufacturers deploying sensor-driven RUL systems, with bearing prediction achieving 94% accuracy 7-14 days before failure. Manufacturing case studies show concrete outcomes: a pulp mill achieved 27-month asset lifespan (vs. 18-month fixed schedule) with $94K capital cost deferral; power generation deployments show 73% average downtime reduction with $4.2M Year 1 savings and 5.1-month payback. Vendor evaluations (2026) rate top performers at 20-40% downtime reduction with 6-12 month implementation timelines, signaling realistic expectations vs. vendor hype.
Market sizing reflects confidence in trajectory. The global predictive maintenance market valued at $17.11 billion in 2025 is projected to reach $116.8 billion by 2034 (24.3% CAGR). The energy sector alone projects $2.81 billion in 2026, growing at 25% CAGR, with operators like NextEra Energy reporting 23% outage reductions. Across manufacturing, consulting benchmarks document 30-50% downtime reduction with 300-500% ROI and EUR 50-150K investments generating EUR 200-800K annual value. Named deployments deliver: ENGIE saved $870K annually across 10,000 connected assets; an automotive OEM achieved $3.2M annual savings and 47% downtime reduction across 200+ machines; a refinery reported $5M+ annual savings. Manufacturing-specific RUL case studies show 18-25% maintenance cost reduction and 12% energy savings from efficiency degradation detection. Methodological progress continues: recent research advances hybrid physics-informed neural networks with asymmetric loss functions for safety-critical applications (IJPHM, USAF), graph attention networks for multi-sensor equipment, domain adaptation for equipment transfer across heterogeneous fleets, LSTM-GAN approaches for RUL under partial sensor failure, physics-informed digital twins for battery aging, and data-efficient approaches requiring <40% of typical training cycles. Yet structural barriers remain evident: only 28% of heavy machinery Tier-1 OEMs report full-scale PdM adoption despite 45% projections, with 18.6-month median payback periods, indicating equipment heterogeneity and retrofitting complexity as significant constraints. A critical operational challenge emerging in production deployments: 91% of deployed ML models degrade silently over time, with degradation often invisible until the 90-day mark, requiring robust drift detection instrumentation to maintain prediction accuracy. The binding constraint remains operational and organisational: practitioners consistently rank work management design, planning ownership, response pathways, and model maintenance above algorithmic performance as barriers to value. Adoption remains sharply stratified: Fortune 500 plants with standardised lines extract consistent ROI, while organisations with diverse equipment fleets, legacy sensor infrastructure, limited failure histories, or inadequate drift monitoring face a steeper path to production ROI.
— Global automotive OEM (10M vehicles/year) deployed AWS IoT-based predictive maintenance with Lookout for Equipment across manufacturing equipment; demonstrates enterprise-scale RUL adoption in automotive sector.
— AWS discontinues Lookout for Equipment October 7, 2026 despite active enterprise deployments (Toyota, Koch); signals vendor exit from productized RUL market despite customer demand, limiting off-shelf options.
— PLoS One peer-reviewed study on LSTM-GAN RUL for turbofan engines with sensor failures; addresses critical deployment barrier that RUL systems can maintain accuracy despite real-world sensor degradation.
— Fortune 500 deployment across 18 global factories with hundreds of connected assets and 30 custom RUL use cases; 6-month PoC at 5 sites followed by 18-factory rollout showing scale and operational maturity.
— GammaTech Engineering case study on physics-informed RUL prediction for battery aging in IoT devices; production deployment demonstrates hybrid physics-ML approach for resource-constrained environments.
— USAF-authored peer-reviewed research in IJPHM addressing uncertainty quantification for safety-critical RUL systems, bridging deployment gap between academic methods and operational aerospace requirements.
— Named pulp mill customer achieved 27-month equipment run (vs. 18-month fixed schedule), $94K deferred capital cost, zero unplanned downtime; industry-wide metrics show 41% unplanned failure reduction.
— Systematic evaluation of 27 PdM vendors in manufacturing (2026); top performers achieve 20-40% downtime reduction with 6-12 month implementation timelines, signaling realistic ROI expectations vs. vendor claims.
2018: Early academic papers validate data-driven RUL methodologies. AWS and Siemens integrate ML-based RUL into cloud and IoT platforms. Specialized vendors like Senseye claim high-accuracy forecasting. Early manufacturing deployments reported in steel and pharmaceutical sectors. Industry analysis highlights adoption barriers (data quality, organizational readiness) as core challenges, not algorithmic limitations.
2019: Nissan North America and Maxion Wheels report measurable RUL deployment outcomes. Market shows strong growth (24.5% CAGR in North America). Research and vendor discourse emphasize challenges: system-level scaling (Carnegie Mellon review), failure data scarcity, model maintenance, and ROI quantification. Kalman filtering and hybrid domain-aware methods emerge as refinements. Industry (PHM Society) recognizes model retraining and data drift as operational burdens alongside algorithmic advances.
2020: AWS and Siemens launch managed RUL platforms (Lookout for Equipment, Predictive Service Assistance) reducing ML expertise barriers. Domain adaptation and temporal convolutional networks advance methodologies for cross-equipment transfer and safety-critical systems (nuclear, aerospace). Petrochemical, automotive, and specialty manufacturing report production deployments with quantified uptime gains. Yet aviation MRO shows <5% adoption despite benefits, exposing persistent organizational and data barriers that platforms alone cannot overcome.
2021: RUL platforms mature into production systems with scale deployments: Senseye reports 10,000+ assets across Fortune 500 clients with €100M+ verified savings and ROI insurance backing. Ecosystem integration accelerates—Senseye-PTC partnership, AWS regional expansion. USAF achieves 35-40% unscheduled maintenance reduction and 3-6% mission capability gains in high-stakes aviation. Research advances probabilistic forecasting (VisPro with uncertainty quantification, 3x accuracy improvement) and bearing RUL methods (>90% accuracy with LSTM-autoencoders). Novel domains emerge: drone delivery (Aerialoop) demonstrates RUL feasibility for autonomous vehicles processing 10,000+ data points per flight. Data scarcity remains unsolved—physics-informed ML (PIML) shows promise with only 17 training simulations achieving 97% R² on exhaust manifold fatigue, but broader adoption still constrained by organizational barriers and failure data availability.
2022-H1: Siemens acquires Senseye, consolidating RUL software leadership and signaling market maturation. AWS continues Lookout platform development with schema detection and improved sensor ingestion validation. Research advances methodologies across bearing, battery, and aeroengine RUL domains with improved uncertainty quantification and domain adaptation. Independent simulation studies validate economic ROI of predictive maintenance despite inherent RUL prediction uncertainty. Critical assessments emerge highlighting persistent adoption barriers: high hardware/software costs, limited suitability for unique machinery, elevated cybersecurity risks, and accuracy-explainability tradeoffs in complex domains like aviation.
2022-H2: Siemens-Senseye consolidation completes, unifying Fortune 500 customer base (Nissan, TATA, Alcoa) under single vendor ecosystem. AWS and Siemens continue platform investment with managed RUL capabilities. Industry surveys show adoption maturation: 91% of practitioners report effective PdM programs, though satisfaction remains mixed. Research advances battery and electrical equipment RUL methods with deep learning (CNN-LSTM, CEEMDAN). Implementation barriers persist: practitioners identify cultural, operational, and organizational factors (inadequate ROI documentation, inconsistent application) as primary failure sources, not technical limitations.
2023-H1: Academic research expands methodological focus: physics-informed neural networks for power electronics (51.3% MSE improvement), random-coefficient methods for aeroengines, and LSTM approaches for manufacturing robotics (Volvo partnership). Applied research extends into new domains—electric utility asset renewal (Brazil)—validating cross-sector applicability. AWS Lookout and Siemens platforms continue post-GA refinement. However, practitioner surveys reveal persistent credibility gap: only 22.5% report PdM/RUL programs as effective, with 51.3% needing improvement due to data quality failures. Adoption remains concentrated in high-volume standardized manufacturing (automotive, petrochemical); construction, utilities, and discrete manufacturing remain on frontier.
2023-H2: Research methodologies advance on complex systems and batteries: TU Delft develops probabilistic RUL for turbofan engines, University of Huddersfield validates multi-indicator RUL with industrial furnace case study. AWS Lookout adds model retraining feature to address model drift burden. Market projections signal sustained growth (23.1% CAGR toward $8.7B by 2030). However, BCG industry report documents persistent implementation failures: data infrastructure struggles, organizational resistance, and difficulty demonstrating ROI—highlighting that vendor platform maturity outpaces organizational adoption capability.
2024-Q1: AWS Lookout and Siemens Senseye continue customer expansion with documented deployments at Siemens Energy, Koch, Cepsa, GS EPS, and Doosan Infracore. Senseye enhances platform with generative AI, with BlueScope steel adopting the feature for knowledge capture. Market data shows acceleration: global predictive maintenance market reaches $4B (2024), projected at 12% CAGR toward $11B by 2034. Consultancy-led deployments report concrete ROI: 40-70% unplanned downtime reduction and 73% achieving positive ROI within 12-18 months. Research advances continue on time-varying bearing RUL conditions and battery prediction methods (XGBoost hyperparameter tuning achieving 6.5% MAPE).
2024-Q2: Research methodologies advance with survival analysis approaches for RUL prediction on censored bearing data; comprehensive deep learning review published in Sensors highlights ongoing academic progress. Vendor platform evolution accelerates: AWS integrates Bedrock for generative AI-enhanced maintenance planning alongside Lookout for Equipment. UK government procurement confirms RUL product GA for public sector. Market growth sustained: global predictive maintenance market expands to $5.41B (2024) with 11% CAGR. Practitioner case studies show continued ROI validation: oil & gas deployments report 36% downtime reduction (10:1 ROI), manufacturing 45% reduction (7:1 ROI).
2024-Q3: Ecosystem signal reversal emerges: AWS announces discontinuation of Amazon Lookout for Equipment for new customers (effective October 2024), marking vendor retreat despite continued security/performance support for existing users—a negative indicator of platform viability. Academic research expands methodologically: Graph Neural Networks for RUL prediction emerge as cutting-edge direction for modeling system interdependencies; battery RUL multimodel integration achieves 0.14% RMSE (46.2% improvement over single models); mining equipment RUL expands domain applicability beyond traditional automotive/petrochemical. Siemens Senseye platform enhancement integrates vision sensor analytics, demonstrating continued ecosystem investment. Academic maturation signals strengthening: tribology journal synthesizes bearing RUL state-of-art, indicating research consolidation and sustained disciplinary focus.
2024-Q4: AWS end-of-support confirmed with December 2024 product discontinuation announcement. Siemens Senseye consolidates leadership with quantified customer outcomes (40% cost reduction, 55% productivity increase, 50% downtime reduction) at BlueScope and Schaeffler. Emerging methodologies advance: LLM-based approaches for turbofan RUL achieve SOTA on NASA C-MAPSS; cloud-native architectures with edge preprocessing and uncertainty quantification demonstrated on ball screw and complex equipment. However, frontline O&M professional skepticism emerged: Emory/Presenso study reveals skill shortages and hype distrust outweigh vendor marketing, indicating persistent adoption barriers despite technical maturity. Practitioner credibility gap remains acute (only 22.5% report effective programs), maintaining stratified adoption between standardized high-volume manufacturing and data-scarce/unique equipment sectors.
2025-Q1: Research methodologies advance with novel architectures: Cambridge synthesizes deep learning methods for sustainable manufacturing; hybrid ensemble models (CNN-Transformer-LSTM-stochastic) demonstrate superior NASA C-MAPSS accuracy; interpretability research at TU Delft achieves 5% accuracy gains via Counterfactual Explanations in Bayesian framework. Market adoption signals strengthen: AI-based predictive maintenance market expands to $922.65M (15.59% CAGR), with RUL estimation as key application segment. Production deployments continue: Senseye RUL system deployed at SCCC (Thailand) integrated with SAP PM; unnamed European automotive OEM reports 92% failure prediction accuracy 24-48 hours ahead with 18% throughput gain and 9-month ROI from welding robot RUL implementation. Ecosystem remains Siemens-dominant with AWS transition to end-of-life confirmed (October 2024 discontinuation for new customers). Organizational adoption barriers persist despite methodological maturity.
2025-Q2: Production deployment expansion continues across sectors: BlueScope Steel's global RUL rollout prevents 1,950 hours of downtime, unnamed automotive OEM connects 10,000+ assets with 12% downtime reduction, Sachsenmilch dairy deployment saves low six figures through early pump failure detection. Academic methodologies advance with Bidirectional TCN attention mechanisms, quantile regression state-space models, and ensemble deep learning approaches improving C-MAPSS accuracy. Domain expansion reaches consumer electronics (Fairphone battery RUL via DaCapo EU project). AI-based predictive maintenance market grows to $922.65M (15.59% CAGR), with RUL as key segment. Implementation barriers persist: consulting reports 40-70% downtime reduction and 73% positive ROI within 12-18 months, but practitioners cite legacy system integration, skills gaps, and difficulty proving ROI. Siemens maintains ecosystem leadership while AWS withdraws from new customer acquisition, maintaining support only for existing deployments.
2025-Q3: Specialist vendor Novity enters RUL market with hybrid physics-ML platform covering industrial equipment classes. Siemens Senseye confirms multi-year BlueScope deployment outcomes: 2,000+ hours downtime prevention across three years and 53 process interruptions avoided. Academic research advances with Attention-LSTM models for aircraft engines achieving SOTA on C-MAPSS and survival analysis methods (Cox PH) improving performance on industrial datasets. Market scales to $9.73 billion in predictive maintenance (23.03% CAGR), with RUL as key application segment. Methodological maturity increases while implementation stratification persists: adoption remains concentrated in standardized Fortune 500 manufacturing while heterogeneous environments face persistent barriers to effective deployment.
2025-Q4: Methodological innovations advance domain adaptation and data scarcity solutions: HybridoNet-Adapt demonstrates LSTM-Attention-Neural ODE fusion with domain adaptation (152-cycle RMSE reduction for batteries), and Bayesian MCMC frameworks address small-sample RUL prediction on limited failure data. Multi-source fusion models (CNN-BiLSTM-Transformer) reduce RMSE by 7.61% on aero-engine and 16.18% on battery datasets. Market continues scaling to $10.93 billion (22.0% CAGR to $44.0B by 2032). Critical practitioner analysis emerges: STXNext documents real-world project failures driven by problem definition misalignment, data engineering burden (70% of effort), and infrastructure challenges, confirming that organizational and operational barriers—not technical limitations—remain primary adoption constraints. Tutorial-level implementation evidence accumulates (dual-attention model reproduction on NASA C-MAPSS), signaling community engagement. Stratification persists: standardized Fortune 500 manufacturing achieves consistent ROI while data-scarce and heterogeneous environments remain on the frontier despite methodological maturity.
2026-Jan: Siemens Senseye consolidates leadership with updated deployment metrics (50% downtime reduction, 55% productivity gains, 40% cost reduction) while AWS confirms discontinuation of Lookout for Equipment for new customers, solidifying vendor consolidation. Academic methodologies advance in explainability (fuzzy similarity-based RUL for membranes achieving 4.50 MAE) and manufacturing optimization (PSO-BiLSTM models reaching 0.97 R² on benchmarks). Energy sector adoption accelerates: market projected to grow 25% CAGR from $2.81B in 2026 to $8.61B by 2031, with major operators (Chevron, NextEra Energy) reporting quantified outcomes (23% outage reduction, $25M annual savings). Critical vendor analysis reframes maturity: the shift from technology focus to execution focus crystallizes as the core challenge—work management, planning ownership, and response pathway design emerge as larger barriers than algorithmic capability, reinforcing that stratified adoption persists despite technical maturity.
2026-Feb: Academic research methodologies advance with systematic literature reviews synthesizing bearing RUL performance and digital twin-based integrated learning frameworks demonstrating methodological consolidation. Production deployments continue at scale: AWS-based motor plant achieves 35% unplanned downtime reduction and 20% cost reduction in 14-week deployment cycle. Market valuation accelerates: $56.71B projected for automotive predictive technology in 2026 (11.94% CAGR for AI solutions through 2031). Critical implementation barriers surface: automotive sector experiences persistent failure modes including alert fatigue, model drift, and missing contextual reasoning despite technical maturity. Vendor consolidation deepens: Amazon Monitron (hardware-inclusive PdM) discontinues with October 2024 effective date, signaling ecosystem shift away from integrated device solutions toward software-only managed platforms. Adoption stratification persists: Fortune 500 manufacturers achieve documented ROI while mid-market and data-scarce environments face barriers tied to problem definition, data infrastructure, and organizational readiness rather than algorithmic capability.
2026-Q2: Deployment scale and vendor stratification signal maturity consolidation. Audits document RUL adoption across 217 Tier-1 auto suppliers and 89 pharma contract manufacturers; bearing RUL systems achieve 94% accuracy predicting failure 7-14 days in advance. Named case studies expand: food manufacturer uses vibration analysis to predict bearing degradation 6-8 weeks ahead, preventing 85% of undetectable failures and eliminating $45K monthly downtime; pharmaceutical manufacturer deploying RUL on critical equipment (pumps, HVAC) achieves 25-30% downtime reduction in GMP-regulated environment; automotive OEM achieves $3.2M annual savings and 47% downtime reduction across 200+ CNC machines with 85-95% RUL accuracy; US Air Force achieves 35-40% unscheduled maintenance reduction with multi-billion dollar savings across aviation fleets; a pulp mill using Senseye RUL achieved a 27-month asset run versus its 18-month fixed schedule, deferring $94K in capital costs; Siemens Energy's 18-factory IIoT rollout deployed 30 custom RUL use cases following a 6-month 5-site PoC, confirming enterprise-scale implementation pathways. AWS confirmed October 7, 2026 end-of-life for Lookout for Equipment despite active Toyota and Koch deployments, completing the platform's exit from the productized RUL market. A systematic evaluation of 27 PdM vendors found top performers deliver 20-40% downtime reduction with 6-12 month implementation timelines — calibrating realistic expectations against vendor hype. Market outlook accelerates: global predictive maintenance market reaches $17.11B (2025), projected $116.8B by 2034 (24.3% CAGR); manufacturing deployments consistently deliver 30-50% downtime reduction and 300-500% ROI with 3-6 month payback timelines. Research advances: physics-informed graph attention ensemble networks (AMST-GATE) outperform SOTA on three benchmark datasets; data-efficient Transformer-CNN-BiGRU hybrids achieve <3.5% MAPE with only 40% of training cycles; LSTM-GAN approaches address RUL accuracy under partial sensor failure in turbofan engines. Yet critical adoption barriers persist: only 28% of heavy machinery Tier-1 OEMs report full-scale PdM adoption despite 45% projections; 76.4% of manufacturing-specific AI projects fail due to data quality constraints and reliability requirements exceeding consumer standards. Stratification deepens: Fortune 500 standardised manufacturing achieves consistent 40-50% downtime reduction and <18 month payback, while mid-market and data-scarce environments remain on the frontier.