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

Real-time streaming analytics

GOOD PRACTICE

TRAJECTORY

Plateau

AI applied to streaming data for real-time pattern detection, alerting, and decision-making on live data flows. Includes stream processing with ML models and real-time anomaly detection; distinct from batch analytics which processes historical rather than live data.

OVERVIEW

Real-time streaming analytics has matured into established good practice. The discipline — applying ML models, statistical aggregations, and pattern detection to continuous data flows rather than batch windows — now rests on a stable ecosystem of GA tooling, managed cloud services, and battle-tested deployment patterns. Apache Flink and Kafka have become de facto standards; all three major cloud providers offer managed streaming services (AWS Managed Service for Apache Flink, Microsoft Fabric Real-Time Intelligence, Google Cloud Dataflow); and the industry recognizes data streaming as a formal software category. Production deployments span banking (fraud detection, real-time risk assessment), payments (sub-10ms decision latency), fintech (Toss processing 7-day frequency capping state at 68GB scale), and operational analytics (billions of events daily). Market evidence signals mainstream adoption: analysts project $146.59B market by 2030 (33% CAGR); enterprises report 764% ROI on implementations (Starbucks), 200+ hour annual savings (Arla); and deployments now address cost-optimization as much as capability. The practice has transitioned past the "whether" question to operational "how"—but that transition remains incomplete. The binding constraint is not technical but organisational: the multi-disciplinary expertise in distributed systems, state management, and streaming semantics that production deployments demand. Large enterprises with dedicated data engineering teams absorb the operational complexity that managed services have reduced but not eliminated. Mid-market adoption lags, constrained not by capability but by the specialized expertise and operational maturity required to productionise these systems reliably. By mid-2026, a market-wide recognition has emerged that continuous streaming carries prohibitive operational overhead for many use cases once justified only by latency requirements—vendors are explicitly guiding customers toward micro-batching and warehouse-native architectures when sub-minute freshness suffices, signaling maturation toward pragmatic adoption boundaries rather than universal real-time.

CURRENT LANDSCAPE

Vendor consolidation has solidified around Apache Flink as the stateful processing engine and Apache Kafka as the transport layer. AWS has shifted entirely to Managed Service for Apache Flink (replacing Kinesis Data Analytics), Microsoft Fabric earned Forrester Wave leader recognition (Q4 2025), and Flink 2.2.0 introduced ML_PREDICT and VECTOR_SEARCH — embedding LLM inference and vector similarity directly into streaming pipelines. Apache Kafka 4.2.0 GA introduced Share Groups (enabling per-record acknowledgement) and Streams Rebalance Protocol (delivering faster, more stable rebalances), advancing ecosystem maturity. Commercial distributions advanced with Ververica Platform achieving Forrester Leader status with claims of 100B+ events/day, <10ms latency, and 40% TCO reduction versus open-source Flink. Databricks Structured Streaming now offers Real-Time mode, achieving sub-5ms end-to-end latency for operational workloads alongside the traditional micro-batch model. Financial institutions (Rabobank, ING Bank, Capital One, Nationwide Building Society) now run real-time fraud detection and risk management on event-driven Flink pipelines; Riskified processes $60B in annual transaction volume with sub-10ms fraud detection decisions. Beyond finance, PayTech enterprises (Toss) deploy complex state pipelines (7-day frequency capping with 68GB state), PostNL migrates IoT asset tracking to managed Flink, Intuit operates 200+ Kubernetes clusters handling 5B daily messages, and tech giants at ByteDance maintain one of the world's largest documented deployments—70,000+ Flink jobs, 11 million+ resource slots, processing hundreds of trillions records daily. Uber demonstrates Kappa architecture patterns using Kafka and Spark Streaming for unified batch-stream processing enabling multi-team latency/correctness trade-offs for dynamic pricing. The ecosystem around Flink continues advancing: Kubernetes Operator 1.14.0 (March 2026) adds blue-green deployment capabilities for zero-downtime updates; research (ICDE 2026) targets latency reduction through prefetching and cache optimization; and practitioner tooling has matured with comprehensive production guides documenting exactly-once semantics, RocksDB state backends, and sub-millisecond latency architecture patterns.

Market growth reflects sustained enterprise adoption acceleration. May 2026 forecasts project 33% CAGR with $146.59B market by 2030, driven by IoT adoption, real-time AI integration, and edge computing. Customer deployments demonstrate quantified ROI: Starbucks processes 1B+ monthly rows across 17 countries with 764% ROI; Arla saves 1,200+ manual hours annually harmonizing European operations. Practitioner cost analyses document that infrastructure <30% of streaming system cost; the remaining 70% derives from engineering and configuration complexity — confirming that organizational maturity, not technology, is the binding constraint.

Operational friction persists at the seams, however. Integration barriers emerge when combining best-of-breed tools: Flink's exactly-once guarantees require two-phase commit, but ClickHouse lacks full ACID transaction support, making a native connector impossible and forcing latency/correctness trade-offs that organizations must engineer around. IBM documentation from early 2026 details Kubernetes operator failures — JobManager cleanup deleting HA metadata, Java cipher suite restrictions breaking SSL handshakes — that illustrate the configuration complexity lurking beneath managed-service abstractions. Critical scaling challenges emerge at volume: 200k TPS fraud detection systems require 3.2GB state per second with pod-crash recovery and consistency maintenance, revealing why state management expertise remains a gatekeeping competency. Economics become punitive at scale: Kinesis for transitional 100TB/day workloads costs "high five figures per month," pushing organizations to evaluate alternative architectures when data lifetimes are short. Practitioners report checkpoint overhead, schema evolution failures, and write amplification in lakehouse architectures. Regulated sectors face additional headwinds: healthcare and pharmaceutical organisations find that platform speed outpaces validation frameworks like GAMP 5, creating compliance gaps. These barriers keep mid-market adoption tethered to technology-forward organisations with dedicated streaming expertise, even as large enterprises consolidate streaming as standard infrastructure. By June 2026, cost pressures and operational maturity barriers have shifted market sentiment: vendors (Databricks, MotherDuck) are explicitly recommending micro-batching and warehouse-native ingestion for use cases previously defaulting to streaming, acknowledging that continuous processing overhead—often 70% engineering complexity, 30% infrastructure—is only justified when sub-second latency genuinely drives business outcomes. This pragmatic boundary-setting indicates the practice has matured from "whether to stream" to "when streaming is worth its operational cost."

TIER HISTORY

ResearchJan-2018 → Jan-2018
Bleeding EdgeJan-2018 → Jan-2019
Leading EdgeJan-2019 → Jul-2023
Good PracticeJul-2023 → present

EVIDENCE (157)

— Uber deployed streaming ingestion (Apache Flink) at petabyte scale, replacing batch with 25% compute reduction and hours-to-minutes freshness improvement across Finance, Delivery, Rider organizations.

— Production supply chain streaming (Kafka+Flink): inventory tracking, SLA protection, disruption response with documented ROI. Shows real-world value and organizational complexity barrier (requires platform engineering expertise).

— Databricks GA feature for streaming checkpoint recovery from failure, addressing production reliability challenge. Documents three recovery approaches (full refresh, preserve with backfill, incremental).

— Databricks production best practices for streaming workloads (Lakeflow Jobs, failure restart, autoscaling guidance, RocksDB state, async checkpointing). Shows ecosystem maturity of lakehouse streaming patterns.

— Databricks guidance: streaming adds complexity (stateful operations, out-of-order handling). Recommends by medallion layer: streaming for Bronze ingestion, batch/incremental for Silver/Gold. Authority guidance on pragmatic streaming adoption.

— Uber case study: 120k events/sec, 5M hexagons, real-time ML features for surge pricing. Demonstrates both capability scale and significant operational burden (backpressure, OOM, optimization expertise required).

— CRITICAL NEGATIVE signal: 2026 market shift away from continuous streaming (Flink, Kafka) toward micro-batching and warehouse-native architectures for cost and complexity reasons. Documents adoption barrier.

— Reference architecture for Netflix-scale recommendation streaming (Kafka→Flink/Spark→Feast): 50ms P99 latency SLOs, 23% engagement uplift, 18% revenue lift documented at 8M customer scale.

HISTORY

  • 2018: Apache Flink reached production-grade maturity with exactly-once semantics in Flink 1.4.0; Kafka evolved from message broker to streaming platform with Streams and KSQL; ecosystem adoption accelerated (Kafka Summit 1200+ attendees, Booking/Braze deployments), but operational stability and deployment complexity remained barriers to broader adoption.
  • 2019: Major enterprise deployments demonstrated production maturity: Lyft scaled real-time ML pipelines to 4M events/min; Branch achieved 12B+ events/day with Kubernetes-native architecture; Bloomberg and other enterprises deployed Kafka Streams to production. Adoption surged with stream processing for AI/ML jumping 6x in two years (6% to 33%); Forrester recognized Google Cloud as a leader. Operational challenges persisted with connection failures and HA mode complexity, limiting adoption to organizations with specialized teams.
  • 2020: Alibaba deployed Apache Flink at record scale during Double 11, processing 4 billion records/second and 7TB/second—validating extreme-scale production readiness. Market analysts projected 10.65% CAGR growth, driven by Kubernetes pipelines, regulatory compliance (MiFID III), and 5G telemetry. Enterprise adoption spread (Citi Group, Bazaarvoice). However, critical reliability gaps emerged: Kafka-Flink integration failures, checkpoint scalability limits beyond 50GB state, and version-specific instability in Kubernetes environments continued to restrict adoption to organizations with advanced data engineering expertise.
  • 2021: Apache Flink 1.13 addressed operational barriers with native Kubernetes HA and Reactive Mode elastic scaling, eliminating manual provisioning. Google Cloud Dataflow achieved Forrester Wave leadership with perfect platform scores. Kafka ecosystem standardized on production frameworks (Azkarra), accelerating enterprise deployments. However, stateful workload challenges persisted: GC/checkpointing failures, connection timeouts, and resource tuning complexity continued limiting adoption to organizations with advanced data engineering teams.
  • 2022-H1: Flink ecosystem matured for cloud deployments: Kubernetes Operator reached 1.0.0 production release with automated job management, and major enterprises deployed Flink at scale (Pinterest real-time ad matching and image dedup, Wikimedia event platform). Spark Structured Streaming advanced with asynchronous checkpointing and autoscaling. Industry adoption metrics showed 48% of organizations analyzing streaming data in real-time. However, serialization bugs (Flink 1.14.x) and Kubernetes Operator upgrade issues continued signaling stability challenges, limiting adoption to organizations with advanced data engineering expertise.
  • 2022-H2: Flink Kubernetes Operator advanced to 1.2.0 with standalone mode and improved upgrade flows, reducing operational friction. Enterprise adoption broadened with named deployments (Lumen, Pinterest, Wikimedia). Vendor consolidation accelerated as AWS sunset Kinesis Data Analytics for SQL in favor of managed Flink. Retail industry adoption metrics showed 93% of orgs value real-time data flow. However, peer-reviewed benchmarking identified Kafka Streams instability, and Kubernetes deployment reliability issues (resource leaks, pod orphaning) persisted, indicating operational maturity remained incomplete for edge cases.
  • 2023-H1: Streaming analytics transitioned to mainstream enterprise use with mid-market adoption metrics showing 74% APAC enterprises achieving 2-5x ROI, up from early-adopter percentages. Peer-reviewed benchmarking confirmed framework scalability in cloud but revealed Apache Beam's resource overhead; Flink dominated security (Lacework 14.5 GB/sec) and e-commerce deployments. Release velocity increased (75 bug fixes in Flink 1.17.1) and optimization focus broadened to low-memory deployments (under 500MB) for edge/IoT. Vendor consolidation completed with AWS fully pivoting to managed Flink. However, production reliability gaps persisted: cloud storage failover failures and Kafka source alignment issues, indicating continued barriers for organizations without specialized data engineering expertise.
  • 2023-H2: Ecosystem expansion accelerated with Apache Flink adding three major connectors (DynamoDB, MongoDB, OpenSearch) and new versioning strategy enabling faster vendor ecosystem development. AWS completed Kinesis rebranding to Amazon Managed Service for Apache Flink, formalizing vendor platform consolidation. Industry analyst Forrester established data streaming platforms as a formal software category (Wave Q4 2023), with Kafka adoption reaching 100K+ organizations. Real-time analytics adoption survey (300 engineering orgs) confirmed it as leading use case (71%) and AI/ML as primary growth driver. However, Kubernetes Operator reliability challenges resurfaced with deployment rollback failures requiring manual HA state recovery, indicating persistent operational friction in production cloud deployments—the critical barrier preventing broader adoption beyond specialized teams.
  • 2024-Q1: Production adoption expanded across energy (Uniper), travel (Booking.com), and sports analytics (NHL) sectors with Flink dominating complex stateful pipelines. AWS accelerated managed service adoption through cost optimization guidance, indicating ecosystem maturity. However, peer-reviewed research (Dynatrace) and practitioner case studies documented persistent operational barriers: fault recovery improvements constrained by configuration complexity, weeks required for setup and tuning, and $50K+ costs from 30-minute outages. Critical bugs continued (FLINK-34518: JobManager failover causing state loss). Configuration complexity and operational overhead remained the primary adoption barrier for organizations without specialized data engineering teams.
  • 2024-Q2: Strategic adoption inflection as 79% of IT leaders (Confluent survey, 4,110 respondents) cited streaming platforms as pivotal for agility and 63% for AI/ML development. AWS GA'd Flink 1.19 with expanded state management and cloud integrations; IDC MarketScape named AWS a Leader. However, critical gap emerged between Kafka ubiquity (80% Fortune 100) and actual stream processing adoption—most Kafka users employed it for buffering/decoupling, not streaming analytics. Practitioner reports documented continued operational challenges: disk saturation failures, 75x latency degradation on object storage, serverless debugging complexity, and Kinesis-Kafka incompatibility issues. Configuration complexity and organizational maturity (not technical capability) became the binding constraint for broader adoption.
  • 2024-Q3: Enterprise deployment broadened with new production case studies: PostNL (Dutch postal service) migrated to managed Flink for IoT asset tracking across billions of events; Intuit revealed 200+ Kubernetes cluster deployment processing 5B daily messages with 60M predictions. AWS released Flink 1.20 support; peer-reviewed research confirmed persistent deployment barriers (multi-disciplinary expertise needed, testing complexity, long setup cycles). Managed service integration gaps (Flink SQL limitations, S3 connectivity issues) signaled operational immaturity for mid-market adoption, cementing large-enterprise dominance of the practice.
  • 2024-Q4: Market growth accelerated with analyst projections reaching USD 128.4B by 2030 (28.3% CAGR, Grand View Research). AWS optimized platform economics with per-second billing and new SQS connector, reducing cost barriers for variable workloads. Industry analysis confirmed Apache Kafka as de facto standard (150K+ organizations) and Flink as standard for stream processing, with emerging trends toward real-time AI integration and BYOC deployment models. However, integration challenges persisted: Airflow-Flink-Kubernetes deployment failures documented in public issue queues, underscoring operational friction even as market adoption accelerated. Large enterprises continued to dominate adoption while mid-market constraints (orchestration complexity, configuration overhead) remained binding.
  • 2025-Q1: Market expansion accelerated with quantified adoption evidence showing global streaming analytics market at USD 15.8B in 2024, projected to reach USD 89.3B by 2033 (18.9% CAGR); U.S. market valued at USD 5.3B in 2025, projected to USD 25.6B by 2034 (19% CAGR). Software segment dominated at 65% share, cloud deployments at 60%, with IT/telecom as leading vertical (23.6%) and emerging AI/ML integration driving growth. Enterprise adoption continued broadening across sectors while organizational and operational complexity remained the binding constraint for mid-market.
  • 2025-Q2: Market momentum accelerated with ISG reporting 48% of enterprises deploying streaming in operational processes (up from 44% in analytics). IMARC projects market reaching USD 118.84B by 2033 (22.16% CAGR). Vendor tooling matured: Google Cloud released Ops Agent integration for Flink monitoring, Confluent advanced Flink event tracking. However, critical barriers persisted: UMA Technology analysis documented scalability challenges at 180 zettabytes data velocity, CAP theorem trade-offs limiting consistency, integration complexity, and $50K+ infrastructure costs, confirming operational/organizational maturity—not technology—as the binding constraint on adoption.
  • 2025-Q3: Vendor ecosystem expanded with AWS releasing Managed Flink Studio (interactive SQL/Python notebooks), signaling democratization of streaming analytics for developers. DeltaStream launched serverless stream processing for AI agent context. However, critical adoption barriers persisted: practitioner analysis documented leaky abstractions in Kafka Streams/Flink, inadequate data integration tooling, and configuration complexity limiting adoption to tech-heavy organizations. Market forecasts continued accelerating (360iResearch: USD 87.27B by 2032 at 17.21% CAGR), though large enterprises maintained dominance of production deployments with mid-market constrained by operational overhead.
  • 2025-Q4: Vendor consolidation finalized with AWS completing Kinesis Data Analytics SQL sunset and Microsoft earning Forrester Wave leader recognition. Apache Flink 2.2.0 (December 2025) introduced AI capabilities (ML_PREDICT for LLM inference, VECTOR_SEARCH) signaling real-time AI integration acceleration. Enterprise adoption sentiment reached inflection: 89% of IT leaders cited streaming platforms as critical, 44% reported 5x ROI, and 90% increased investments—confirming mainstream strategic valuation. However, critical barriers persisted: Confluent analysis documented hidden TCO costs beyond implementation; architectural analysis reinforced Kafka/Flink separation patterns; and practitioners continued citing configuration complexity and specialized expertise requirements as binding constraints preventing mid-market adoption despite technology maturity.
  • 2026-Jan: Market growth accelerated with Stratistics MRC forecasting real-time data streaming market reaching $6.11B by 2032 (19.7% CAGR), while aggregate market estimates showed real-time data integration at $15.18B growing to $30.27B by 2030. Apache Flink development continued with January releases adding async Python scalar function support and enterprise integrations (IBM Cloud Pak, Huawei Cloud). Adoption drivers remained strong (72% event-driven architecture adoption, 295% average ROI), but critical barriers persisted: practitioners and analysts documented operational complexity (schema evolution failures, checkpoint overhead), regulatory compliance gaps in highly regulated sectors (healthcare, pharma), and ongoing architectural debates over streaming engine necessity—indicating mainstream adoption constrained by organizational maturity rather than technology capability.
  • 2026-Feb: Ecosystem maturity continued with Apache Flink Kubernetes Operator 1.14.0 incorporating blue-green deployment fixes and active FLIPs addressing adaptive partitioning and performance improvements, signaling ongoing technical refinement. Cloud provider integration broadened through Microsoft Fabric Real-Time Intelligence with 3-8 second end-to-end latencies and practical IoT/finance use cases, and IDC projections forecasting 85% of new enterprise applications on real-time architectures by 2027. However, critical operational barriers remained visible: IBM documentation in February 2026 detailing Kubernetes operator reliability edge cases (JobManager cleanup TTL losing HA metadata, Java cipher suite restrictions blocking SSL handshakes), indicating that despite framework maturity, production Kubernetes deployments continue encountering configuration complexity and stateful recovery challenges. Organizational adoption drivers strengthened through demonstrated ROI (financial institution streamlined fraud detection and customer retention via data product architecture), but deployment complexity and specialized expertise requirements continued constraining mid-market adoption to technology-forward organizations.
  • 2026-Mar: Financial sector adoption matured with peer-reviewed research (IJCA) documenting production Kafka deployments across Rabobank, ING, Capital One, and Nationwide for real-time fraud detection and risk management. AWS Managed Service for Apache Flink FAQs documented canonical use cases (streaming ETL, continuous metrics, responsive analytics), Riskified case study confirmed sub-10ms fraud detection at $60B annual transaction volume with 2-8x scaling during peaks. Production patterns advanced with comprehensive tutorials detailing sub-millisecond ingestion-to-serving latency stacks and exactly-once semantics configuration. Research accelerated latency optimization (ICDE 2026) targeting state I/O decoupling via prefetching. Vendor comparison analysis positioned Flink as standard for stateful processing, managed platforms as adoption accelerators, confirming ecosystem maturity—though organizational readiness (not technology) remained the binding constraint for broader mid-market adoption.
  • 2026-Apr: Deployment adoption accelerated across multiple verticals and scales. Uber published two production case studies: exactly-once ad event processing across Flink/Kafka/Pinot at revenue-critical scale, and 120k events/sec geospatial ML feature pipeline serving demand forecasting across 5M hexagons. Financial sector saw widening adoption: Capital Vanguard Holdings deployed real-time analytics platform replacing spreadsheet workflows (99.8% reduction in data prep time, 500ms update latency); Burton-Taylor analyst report quantified financial market data vendors recording $49.2B revenue with real-time trading >35%. Sector diversification broadened: automotive (Rivian+VW RV Tech, 88% data reduction via Flink), aviation (Etihad Airways, Qantas real-time flight visibility), retail IoT, and telecom migrations documented named production deployments. Payments fraud detection case quantified ROI: streaming-first architecture reduced false positives from 25% to 8%, cut latency 70%, deployed in 8-12 weeks. Ecosystem signals included Apache Kafka 4.2 GA (38 KIPs, 155 contributors), CrowdStrike trillion-events-per-week scale, and market update ($1.37B in 2026 projected to $8.25B by 2034, 25.1% CAGR). TCO analysis quantified the binding constraint: infrastructure <30% of cost; remaining 70% from engineering and configuration complexity—organizational maturity, not technology, limits mid-market adoption.
  • 2026-May: Deployment evidence broadened with Toss (Korean fintech) demonstrating 7-day frequency capping at 68GB live state using Flink+RocksDB, and Uber publishing a Redis/Fargate/Dash system that replaced 1-hour batch latency with real-time dashboards for 10M daily events. Platform ROI quantified: Starbucks 764% ROI on 1B+ monthly rows; Arla 1,200 manual hours saved annually. Ververica and Microsoft Fabric both earned Forrester Wave Leader recognition for streaming platforms. Market forecast updated to $146.59B by 2030 (33% CAGR). Uber's AthenaX case study documented >1 trillion daily Kafka messages with Flink-compiled SQL, compressing deployment cycles from weeks to hours. Practitioner cost analysis confirmed infrastructure is under 30% of streaming system cost—the remaining 70% is engineering and configuration complexity, cementing organizational maturity as the binding adoption constraint rather than technology capability.
  • 2026-Jun: Ecosystem maturity continued with Apache Kafka 4.2.0 GA introducing Share Groups for per-record acknowledgement and Streams Rebalance Protocol for faster application-specific rebalancing. Databricks advanced streaming latency with Structured Streaming Real-Time mode achieving sub-5ms end-to-end processing for operational workloads alongside micro-batch option. ByteDance disclosed 70,000+ Flink jobs, 11 million+ resource slots, hundreds of trillions records daily, demonstrating category-level scale. Uber published Kappa architecture patterns solving multi-team latency/correctness requirements for dynamic pricing, and separately confirmed petabyte-scale Flink streaming ingestion replacing batch with 25% compute reduction and hours-to-minutes freshness improvement across Finance, Delivery, and Rider organizations. However, the market-wide cost correction deepened: MotherDuck and Databricks explicitly recommend micro-batching and warehouse-native ingestion for cases where sub-second latency is not a genuine business requirement—documenting that continuous streaming overhead (70% engineering complexity, 30% infrastructure) is only justified at the latency extremes. Reference architectures demonstrated validated production ROI: Netflix-scale recommendation at 50ms P99 (23% engagement uplift, 18% revenue lift), fraud detection at 50K TPS / 800ms latency with 1B+ events/hour; Goldsky replaced Flink with Rust-based Streamling achieving 30x compute reduction and $1M+/year cost savings across 3,000+ pipelines, a negative signal on Flink's operational overhead for non-hyperscale teams. Databricks GA'd streaming checkpoint recovery with three documented approaches, advancing production reliability patterns.