Perly Consulting │ Beck Eco

The State of Play

A living index of AI adoption across industries — where established practice meets the bleeding edge
UPDATED DAILY

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.

The Daily Dispatch

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.

AI Maturity by Domain

Each dot marks the weighted maturity of practices within a domain — hover for a brief summary, click for more detail

DOMAIN
BLEEDING EDGEESTABLISHED

Incident triage & root cause analysis

GOOD PRACTICE

TRAJECTORY

Stalled

AI that classifies, routes, and prioritises incidents while automating root cause diagnosis. Includes intelligent ticket routing and automated fault tree analysis; distinct from automated remediation which takes corrective action rather than diagnosing.

OVERVIEW

AI-driven incident triage and root cause analysis is a proven practice with a mature vendor ecosystem, GA tooling from multiple platforms, and documented ROI at enterprise scale. The core value proposition — classifying and routing incidents automatically while diagnosing why failures occur, not merely that they occurred — has been validated through years of production deployments at organisations ranging from Fortune 500 to Tier-1 service providers. The question for most teams is no longer whether these tools work, but how to implement them without drowning in integration complexity and alert noise. That distinction matters: despite broad vendor capability and strong MTTR reduction evidence, a persistent adoption paradox has emerged. Organisations overwhelmingly invest in AIOps platforms yet struggle to operationalise AI-assisted triage beyond initial pilots, particularly in the mid-market. LLM-assisted diagnosis is accelerating vendor roadmaps, but research reveals systematic reliability gaps that prompt engineering alone cannot resolve. The practice is mature and accessible; the barrier is execution, not technology.

CURRENT LANDSCAPE

The vendor ecosystem is broad and GA-ready. Splunk ITSI, BigPanda, Moogsoft (Dell), IBM Instana, and Logz.io all ship LLM-assisted triage and root cause suggestion as production features; BigPanda's AI Incident Assistant and Microsoft's RCA Agent via Copilot Studio represent the latest wave of generative-AI-native releases. Named deployments continue to demonstrate measurable impact: ServiceNow's NBA Workplace Service Delivery achieved 51% annual ROI with 30-50% MTTR reduction and 99.2% noise suppression; Thoughtworks reports RCA cycles compressed from hours to minutes across 16+ client engagements, with L1/L2 ticket volume down 35-40%. Incident.io documents 37% faster MTTR through AI-automated post-mortems, saving teams roughly 75 minutes per incident. Türk Telekom achieved 49% improvement in service outages and Vodafone reduced alarm noise by over 70%.

These results coexist with a striking adoption gap. A Sumo Logic survey of 500+ security leaders found that 90% consider AI important for security purchases, yet only 9% have deployed it for incident triage. Operational toil rose 30% in 2025 despite 51% of companies deploying AI tools, and 73% of organisations experienced outages from ignored alerts. Causal AI adoption grew 40% in 2026 but pilot failure rate stands at 95%, driven by data quality issues. The gap is not capability but integration: mid-market RCA projects face severe cost overruns — one insurance deployment reached $4.7M against a $1.2M budget — and 94% of IT leaders cite vendor lock-in as a concern. On the technical side, LLM-based RCA shows systematic failures: a February 2026 study found hallucinated data interpretation and incomplete exploration persist across all model tiers regardless of capability level, requiring human review. Five critical data quality barriers block deployment: incomplete work order history, inconsistent asset naming, missing failure classifications, data silos between observability systems, and inconsistent technician data entry—costing organisations $12.9M annually. Security governance adds friction: 98% of CISOs in one survey report delaying AI agent deployments due to insufficient controls. The tooling works; the organisational and technical scaffolding to deploy it reliably is what most teams still lack.

As of May 2026, agentic RCA platforms reach production scale with multi-vendor GA: Splunk Observability Cloud AI Troubleshooting Agent correlates metrics, logs, traces into evidence-backed root cause summaries, shifting on-call engineers from data gathering to decision-making; BigPanda's AI Detection and Response (ADR) with AI Incident Assistant delivers real-time triage and root cause analysis integrated with ServiceNow workflows; Kentik Network Intelligence Platform introduces AI Advisor with automated diagnostics (on-demand connectivity, config context, device access) reducing manual troubleshooting steps. Research validation advances causal approaches: peer-reviewed frameworks at ICML 2026 and arXiv validate methods (LATS-RCA achieving high diagnostic accuracy on production microservices despite polyglot stacks; Bayesian Root Cause Discovery with statistical consistency guarantees; PRIM meta-learning achieving zero-shot inference in 17ms for systems with 100+ variables) while identifying real-world implementation constraints. Real-world deployment evidence strengthens: American Express reduced MTTR 32% and achieved 82% RCA accuracy using Traversal; SquareOps managing 50+ production Kubernetes clusters reports 40-70% MTTR reduction via LLM triage and RCA; Microsoft Triangle demonstrated 91% reduction in Time-to-Engage in Azure production. Yet the governance and trust barriers remain acute: a 1,000+ IT leader survey found 61% adoption of AI for accelerated RCA but 71% still manually verify outputs and 62% struggle trusting recommendations, capturing an implementation gap where teams receive AI insights but lack confidence or governance frameworks to act on them directly.

TIER HISTORY

ResearchJan-2018 → Jan-2019
Bleeding EdgeJan-2019 → Jul-2022
Leading EdgeJul-2022 → Jan-2023
Good PracticeJan-2023 → present

EVIDENCE (123)

— SRE consulting firm managing 50+ production Kubernetes clusters reports 40-70% MTTR reduction via LLM triage and RCA; demonstrates safe agentic remediation architecture with policy engine guardrails.

— Comprehensive analysis proposing 6-tier agentic RCA capability ladder (L0-L5) with named vendor examples; Traversal achieves 32% MTTR reduction and 82% RCA accuracy at American Express.

Release Notes - BigPanda DocsProduct Launches

— BigPanda AI Detection and Response GA includes AI Incident Assistant with real-time incident triage, root cause analysis generation, and ServiceNow integration for L4-L5 agentic investigation.

— Recent peer-reviewed research on causal meta-learning for RCA; achieves zero-shot inference in 17ms for systems with 100+ variables, advancing methodological boundaries of RCA under structural uncertainty.

— Splunk Observability Cloud AI Troubleshooting Agent GA correlates metrics, logs, traces for evidence-backed root cause summaries; shifts on-call role from data gathering to decision-making.

— Operational analysis quantifying MTTR composition (15min context, 20min troubleshooting, 13min docs); Datadog Bits AI SRE achieves 95% MTTR reduction by eliminating context-assembly coordination tax.

— Peer-reviewed LLM multi-agent RCA framework (LATS-RCA) achieving high diagnostic accuracy on production microservices while identifying real-world challenges like polyglot stacks and inconsistent logging.

— ICML 2026 research on causal inference RCA with statistical consistency guarantees, evaluated across three microservice systems with state-of-the-art top-k accuracy in failure diagnosis.

HISTORY

  • 2018: IBM/Instana deploys AI-based RCA using Dynamic Graphs; Applitools launches web app RCA tool; academic research advances Bayesian and algorithmic approaches to production failure diagnosis. Evidence remains sparse, confined to early vendor announcements and research papers rather than widespread enterprise deployments.
  • 2019: Production RCA frameworks reach large-scale infrastructure (LinkedIn/Microsoft deployed dimensional analysis for millions of log entities). Splunk ITSI and Moogsoft gain traction in enterprise triage (LAX airports, Allied Irish Banks). Academic work extends RCA to causal discovery in ML models and CI/CD testing. Negative signal emerges: healthcare critique documents RCA cost ($8,000+/incident) and scalability challenges, tempering optimism about practice maturity.
  • 2020: Cloud-native RCA offerings proliferate (Moogsoft Observability Cloud launch, BigPanda 14 new integrations). Industry surveys show 44% AIOps adoption consideration and 12-hour P1 RCA times as median, indicating pain-driven market expansion. Academic research advances (AURORA automated crash diagnosis), but methodological debates persist about RCA implementation effectiveness and systemic vs. blame-focused approaches. Adoption remains concentrated at large enterprises with dedicated observability teams.
  • 2021: Market consolidates around AIOps platforms with triage/enrichment capabilities. Vendors iterate on existing offerings (BigPanda automatic triage, Moogsoft APEX updates) but deployment evidence remains limited to large enterprises. Academic discussion continues around RCA methodology and systemic challenges. Evidence of mainstream RCA adoption beyond Fortune 500 remains sparse.
  • 2022-H1: RCA adoption accelerates market-wide: 90%+ of organizations investing in AIOps, with RCA cited as top critical MSP capability (48%) and monitoring challenge (46%). BigPanda achieves $1.2B valuation (155% YoY ARR growth) with deployments at Cisco, Sony, Autodesk. Concrete outcomes documented (AmerisourceBergen 2/3 alert reduction, Wiley 50%+ false positive cut and 37% MTTR improvement) alongside critical signals—failed Splunk ITSI and New Relic pilots due to tuning burden. Adoption broadens beyond Fortune 500 but implementation barriers and tool complexity remain significant.
  • 2022-H2: Vendor platform maturity advances (Moogsoft v9.0 GA, BigPanda Series E extension at $1.2B valuation). Deployment evidence spans federal agencies (HHS/Splunk ITSI), media/tech (BBC Studios achieving 33% cloud cost reduction), and financial services (Wells Fargo, UBS via BigPanda). IBM Instana customers report 50% MTTR improvement and 75% reduction in debugging time. Industry data quantifies pain point: average IT outage costs $12,913/minute across 300 surveyed businesses. Adoption plateaus outside well-resourced enterprises; implementation complexity and false positive burden remain primary barriers.
  • 2023-H1: Moogsoft deployment case study demonstrates MTTD reduction of 75% and incident reduction of 30%, confirming triage platform effectiveness in communications infrastructure. Microsoft Research advances LLM-assisted incident management methodologies (ICSE 2023). BigPanda gains Forrester Wave recognition as Strong Performer in process-centric AIOps evaluation, validating multi-vendor market consolidation. Academic and vendor momentum supports leading-edge positioning, though adoption beyond well-resourced enterprises remains constrained by implementation complexity.
  • 2023-H2: Vendor landscape consolidates via Dell's acquisition of Moogsoft, signaling market maturity and investor confidence in RCA/incident triage as core IT infrastructure capability. BigPanda launches generative AI for automated incident analysis, advancing root cause suggestion and impact estimation capabilities. Open-source ecosystem expands with PyRCA ML library. Adoption barriers persist: 74% of ITOps professionals report tool workload struggles despite broad 90%+ AIOps investment intent.
  • 2024-Q1: Real-world RCA effectiveness continues in production: NCTA technical paper reports MSO networks achieving 99% alarm suppression and 80% first-recommendation accuracy. BigPanda customer deployments show measurable outcomes (Autodesk 69% incident reduction, IHG 99.8% availability). Splunk ITSI adoption extends to legacy mainframe infrastructure. However, market signals reflect infrastructure strain: PagerDuty survey documents 16% rise in enterprise incidents and warns that rapid AI deployment may be overwhelming monitoring capabilities. Analyst opinion cautions against hasty RCA tool adoption without governance, drawing parallel to cloud migration overruns (75% budget overages). Deployment effectiveness proven at scale, but adoption barriers remain related to implementation complexity and integration burden.
  • 2024-Q2: LLM-assisted RCA reaches production at major tech companies: Meta deploys fine-tuned Llama 2 (7B) achieving 42% accuracy for web infrastructure incidents, while Microsoft's RCACopilot achieves 0.766 accuracy on production cloud incident dataset after 4+ years of integration. Vendor platforms mature with IBM Instana launching Probable Root Cause feature and Splunk ITSI v4.19 adding Service Impact Analysis. However, critical assessment documents persistent causality-vs.-correlation challenges in observability tools, emphasizing that correlation-based alerting does not equal true root cause diagnosis. Wipro reports MTTR gains with Splunk ITSI but notes end-to-end visibility limitations. Market demonstrates LLM-driven RCA viability at scale, balanced against methodological constraints and visibility gaps in current platforms.
  • 2024-Q3: Production RCA deployments consolidate as mainstream practice: Chipotle achieved 50% MTTR reduction with BigPanda AI-driven incident triage; Moogsoft (Dell APEX) demonstrates practical alert correlation and root cause identification in multi-source distributed infrastructure. Academic survey (arXiv) validates RCA methodologies across microservices while documenting persistent fault localization challenges and real-world outage prevalence. Splunk's Gartner Leader positioning confirms analyst validation of observability platform maturity for RCA capabilities. Practitioner guidance emphasizes AI-enhanced RCA integration with existing tools and claims of 70%+ MTTR gains, signaling broader adoption in DevOps/Kubernetes environments. RCA practice demonstrates established production viability with named deployments, analyst recognition, and methodological guidance for integration.
  • 2024-Q4: RCA platforms enter mature steady state with incremental feature evolution: IBM Instana introduces Probable Root Cause using causal AI, and Logz.io integrates AI-driven RCA agents—demonstrating sustained vendor investment in automation. However, critical practitioner assessment documents persistent operational challenges: fragmented dashboards, alert noise, and cross-team coordination delays continue limiting RCA effectiveness despite technological maturity, highlighting implementation barriers rather than capability gaps. RCA adoption remains mainstream in well-resourced enterprises, with competitive pressure driving feature iteration but operational friction points constraining broader mid-market expansion.
  • 2025-Q1: RCA deployment evidence broadens beyond software platforms into global service infrastructure: CMC Networks (Tier-1 service provider) achieves 38% MTTR reduction and 74% faster issue resolution across 62 African and Middle East countries using BigPanda/NetBrain event correlation and intelligent diagnostics. Atlassian survey finds 79% of teams exploring AI for incident trending, signaling continued organizational adoption momentum. Splunk ITSI deployment in critical infrastructure (electrical utility) confirms RCA effectiveness in regulated environments with strict compliance requirements. BigPanda customer testimonial (Zayo) documents faster root cause diagnosis improving MTTR. Applied research (HCL Technologies patent) advances RCA methodology by addressing false positive challenges in automated analysis, reflecting ongoing technical refinement. Practitioner analysis emphasizes AI's potential to overcome traditional RCA limitations (human bias, incomplete data, time-consuming investigation) while acknowledging complexity of data integration requirements. Evidence portfolio shows deployment scale increasing beyond Fortune 500 to include MSPs, service providers, and regulated infrastructure, with AI-enhanced triage and diagnosis capabilities becoming standard expectation in enterprise ITOps platforms.
  • 2025-Q2: LLM-assisted RCA accelerates with major vendor GA releases: Microsoft launches RCA Agent via Copilot Studio for automated root cause identification; BigPanda deploys generative AI for incident analysis with LLM-generated titles, summaries, and root cause suggestions; Moogsoft (Dell) releases Probable Root Cause feature for correlation and feedback. Academic validation strengthens: Microsoft researchers (eARCO) demonstrate 21% accuracy improvement over RAG-based LLMs on 180K+ historical incidents using prompt optimization. Alert fatigue remains persistent pain point: studies cite 51% of SOC teams overwhelmed by volume; vendor claims of 95% noise reduction with AI-driven correlation highlight market's focus on triage burden. Practitioner assessment emphasizes RCA's continued challenge in moving from reactive firefighting to proactive anomaly-driven incident management, though LLM integration accelerates practical adoption of automated root cause suggestion at scale.
  • 2025-Q3: AI-driven RCA adoption signals strengthen in enterprises: ETR survey of 1,700 IT decision makers across 23 countries reports AI-assisted root cause analysis and troubleshooting as top impactful capabilities, with 54% AI monitoring adoption (up from 42% in 2024) and full-stack observability cutting downtime costs in half ($2M/hour median impact). Vendor case studies document production deployments achieving 65-75% faster processing times (healthcare prior authorization 8 days→2.5 days, loan processing 12 days→5 days with $2.3M annual savings). Global AI RCA market reaches $1.7B valuation, projected to grow 18.2% CAGR through 2033 across manufacturing, healthcare, telecom, and financial sectors. However, critical research from MIT NANDA initiative reveals 95% of GenAI investments yield zero measurable business returns due to tools unable to adapt to dynamic workflows and data foundation deficiencies. Industry signals document rising AI initiative abandonment: 42% of AI projects abandoned before production deployment (up from 17% prior year), indicating real-world implementation challenges, vendor lock-in risks, and execution barriers constraining broader RCA adoption despite technology maturity and vendor investment.
  • 2025-Q4: RCA consolidation reveals gap between vendor capability maturity and real-world deployment ROI. Academic survey (arXiv, October 2025) analyzes 135 RCA papers and identifies systematic methodological gaps in goal-driven classification. Production deployments document continued success at scale: telecom alert noise reduction 90% with improved NPS; industry reports show AIOps RCA market growth to $1.7B. However, critical assessments surface severe adoption barriers: consultancy analysis documents 68% of AI projects miss ROI targets within 2 years; detailed case study of mid-market insurance RCA reveals $4.7M actual cost vs. $1.2M budget with integration and change management costs 10-15x underestimated; Zillow RCA system failure demonstrates data completeness risk ($500M impact from incomplete data reliance). Practitioner assessment: 68% of operations teams report alert fatigue, 45% of MTTR consumed by data gathering, implementation complexity and cost overruns limiting mid-market adoption. RCA proven at enterprise scale with mature observability investment, but ROI challenges and vendor lock-in risks constrain broader deployment.
  • 2026-Jan: RCA adoption paradox widens: academic advances accelerate (multimodal frameworks achieving 48.75% diagnostic accuracy), vendor platforms mature with production deployments (Splunk ITSI retail chain 40% storage savings, 64% ROI improvement), and consultant field reports confirm effectiveness (Thoughtworks: 35-40% L1/L2 reduction, 65-75% faster processing). Yet adoption barriers deepen: Runframe synthesis shows operational toil rose 30% in 2025 despite 51% AI deployment, 73% of orgs experienced outages from ignored alerts; Sumo Logic survey reveals 90% say AI important for security purchases but only 9% deploy for incident triage—documenting widening gap between vendor capabilities and real-world organizational ROI. RCA remains effective at enterprise scale but mid-market complexity and vendor lock-in risks intensify barriers to broader adoption.
  • 2026-Feb: Fundamental RCA limitations and governance concerns surface: BigPanda launches GA AI Incident Assistant for automated incident analysis and root cause suggestion, signaling continued vendor investment; incident.io case study reports 37% MTTR improvement via AI-automated post-mortems. Yet critical academic research (1,675 benchmarked LLM RCA runs) identifies 12 systematic failure types that persist across all model tiers, showing prompt engineering cannot resolve core reliability issues. Security and governance barriers tighten adoption: 98% of CISOs report delaying AI agent deployments due to insufficient security controls; Vectra's survey shows 76% deploy AI for SOC but gains lag; 94% of IT leaders concerned about vendor lock-in. RCA proven at enterprise scale but technical limitations and governance gaps constrain broader adoption momentum.
  • 2026-Apr: Adoption paradox sharpens with new survey data: PagerDuty's 2026 State of AI-First Operations survey of 1,000 leaders documents that 63% of more-resilient organizations actively use AI in incident response versus 53% of less-resilient peers, while financial exposure reaches $1M+/hour for 8% of firms. Causal AI adoption grew 40% in 2026 with 80% time reduction in defect investigation, yet pilot failure rate stands at 95%, driven by the same five data quality barriers costing $12.9M annually per Gartner. New peer-reviewed research (CausalRCA) demonstrates 35% accuracy improvement and 28% MTTR reduction over correlation-based methods in Kubernetes environments using structural causal models; Cisco's unified SOC/NOC deployment with Splunk confirms production-scale MTTR gains; Microsoft published formal guidance on AI-specific triage challenges including root cause ambiguity from non-determinism. NeuBird AI survey of 1,039 SRE/DevOps/IT Ops professionals identifies automated RCA as the leading AI use case in incident management, while Gartner-cited data shows diagnosis time compressing from 8 hours to 2 hours at organizations where AI RCA is embedded.
  • 2026-May: Platform ecosystem consolidates with LLM-native RCA capabilities: Splunk Observability Cloud AI Assistant GA (April 2026) supports automated incident analysis and error root cause identification across observability data; BigPanda AI Detection and Response (ADR) delivers AI Incident Analysis with plain-language summaries and root cause suggestions plus ServiceNow integration; BMC Helix AIOps Root Cause Analyzer GA feature demonstrates autonomous agent RCA with deployment-specific MTTR gains. Academic advancement includes PRIM meta-learned Bayesian RCA achieving 17ms zero-shot inference for systems with 100+ variables, and the Arvo AI 6-tier agentic RCA capability ladder (L0–L5) documenting named vendor outcomes including Traversal's 32% MTTR reduction and 82% RCA accuracy at American Express. Real-world deployment cases confirm ongoing maturity: SquareOps managing 50+ production Kubernetes clusters reports 40-70% MTTR reduction via LLM triage; Cisco IT achieved 25% incident reduction compressing diagnosis from hours to minutes; NTT Data global SOC achieved 50-70% per-incident effort reduction with 90% auto-close on false positives. However, critical assessment persists: deterministic graph-based RCA reveals LLM-based AIOps failure modes; SolarWinds survey of 1,000+ IT leaders documents 61% RCA adoption but 71% manual verification and 62% trust gaps, capturing implementation barriers where governance and confidence constraints limit direct action on AI recommendations. RCA platform maturity proven at scale, but organizational readiness and methodological limitations constrain broader mid-market deployment momentum.

TOOLS