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

Document processing & data capture

GOOD PRACTICE
ALSO IN🔄 Operations & Process Automation👁️ Computer Vision & Sensing

AI that extracts data from documents, forms, and handwritten materials using OCR and intelligent processing. Includes template-free extraction and handwriting recognition; distinct from multimodal document understanding which handles complex layouts and diagrams requiring vision-language models. Scope covers ML/AI-powered extraction and recognition; traditional template-based OCR and manual data entry are out of scope.

OVERVIEW

Intelligent document processing has crossed the threshold from promising technology to proven operational capability. ML-powered extraction from documents, forms, and handwritten materials — using OCR, NLP, and increasingly LLM-based reasoning — now runs in production across cloud platforms from Microsoft, Google, and AWS, with GA tooling, competitive pricing, and analyst-validated ROI. Gartner's inaugural Magic Quadrant for IDP (April 2026) names five leaders—ABBYY, Hyperscience, Infrrd, Tungsten Automation, UiPath—confirming ecosystem maturity. An AIIM survey of 600 enterprises found 78% operational with AI document automation, and the IDP market reached $8B in 2024. The practice question has shifted from "does it work" to "how to roll it out" — though that rollout is harder than vendors suggest. The market is now inflecting toward agentic orchestration and decision-acceleration; extraction capabilities are commoditized. Accuracy degrades sharply on handwriting, non-Latin scripts, and edge-case layouts. Production reliability incidents recur across platforms. The gap between pilot success and scaled deployment remains wide: 20-40% of real-world documents fall outside standard templates, requiring HITL architecture and enterprise-grade preprocessing. The frontier is now agentic processing and generative AI integration, but the binding constraint remains the same one it has been since 2017: field-level accuracy under real-world conditions, with silent corruption risk and benchmark-to-production gaps.

CURRENT LANDSCAPE

The market is consolidating around agentic and LLM-powered extraction, displacing the template-based systems that defined the previous generation. Gartner reports 67% of enterprises now evaluating agentic document processing, up from 23% two years prior. Platform vendors are shipping production-grade agentic architectures: AWS published reference architecture for multi-agent orchestration via Bedrock AgentCore (April 2026) with graph-based workflows, dual-path routing (known docs via Textract, complex/handwritten via Bedrock), and serverless deployment. Azure Document Intelligence v4.0 GA shipped searchable PDF output and incremental classification training in February 2026; UiPath's acquisition of WorkFusion signals vendor consolidation toward vertical AI platforms rather than horizontal extraction tools. Everest Group analyst assessment identifies commoditization of core OCR/extraction and market inflection toward orchestration and decision-acceleration, indicating maturity of the extraction layer.

Named deployments continue to deliver strong ROI at increasing scale. A KumoHQ-documented logistics firm cut processing time 87% (40+ hours to 5 hours weekly) at 94% accuracy with LLM-powered extraction and direct ERP integration, achieving 3.5-month payback. AND Digital's IDP Accelerator achieved $2M annual savings for a leading FinTech platform through smart cost routing (Textract for standard, Bedrock for complex). Esprigas processes 27,000 documents monthly with $73,800 in savings; Erewhon processes 20,000 invoices monthly at $45,000 savings; Disney Trucking eliminated 6 FTEs entirely through automation. EY runs a tax processing pipeline at scale with hundreds of extractors mixing OCR, custom models, and generative augmentation. Google uses Document AI internally for sustainability report processing. Tungsten Automation (formerly Kofax) serves 25,000+ customers across 70+ countries, including 8 of the top 10 global banks and 7 of the top 10 insurers, confirming enterprise-scale production deployment breadth. Industry statistics confirm the economics: 60-80% cost reduction per document, 70-90% processing time improvement, 6-18 month payback across lending, insurance, and BPO segments. However, critical accuracy threshold emerges: 96-99% accuracy is required for viable ROI; systems operating at 90-94% accuracy have ROI undermined by manual correction loops.

Progress on handwriting recognition accelerates with frontier MLLMs: latest models (Gemini 3.1, GPT-5.4, Claude Sonnet 4.6) achieve ~85% accuracy on structured handwritten medical forms with 90% weighted F1 scores, indicating viable pathway toward automated handwritten document processing. These successes coexist with deepening recognition of production barriers. Independent testing reveals frontier LLMs (GPT-4o, Claude, Gemini) cluster errors on scanned/low-resolution documents (6-8% failure rates) and multi-currency forms, with dangerous confidence calibration on complex documents—creating silent corruption risk. Practitioner analysis documents benchmark-to-production gap: 55+ percentage point performance variance across document types, and 20-40% of real-world documents fall outside standard templates requiring HITL architecture. Specific failure modes persist: column-order collapse in multi-column layouts, table flattening with merged headers, semantic errors (wrong field extracted but schema-valid), and context truncation across pages. The DOJ and House Oversight Committee released over 3 million PDFs with non-functional OCR, rendering them unsearchable—a reminder that extraction at scale still breaks in ways that matter. Azure Document Intelligence experienced custom classification training jobs stuck at "notStarted" status for weeks. A Vertesia survey of 1,500 IT executives found 96.8% cite ECM vendor roadmaps as significant barriers to AI implementation. Manual correction loops still account for 40% of input management costs, per Parashift's analysis, even as AI reduces error rates from 4% to 0.5%. Enterprise IDP deployments require 3-12 months and $50k-500k+ annual costs with steep learning curves and template maintenance burden, creating accessibility gaps for mid-market and SMB segments. The economics work — but only with sophisticated process redesign, OCR-first architecture (not LLM-only), enterprise-grade preprocessing, confidence scoring, HITL routing, and accuracy-matching strategies most organisations have not yet undertaken.

TIER HISTORY

ResearchJan-2017 → Jan-2017
Bleeding EdgeJan-2017 → Jan-2018
Leading EdgeJan-2018 → Jan-2020
Good PracticeJan-2020 → present

EVIDENCE (136)

— Lleverage case studies showing manufacturing company reducing 4 FTEs to 1 with error rate cut from 7% to 0.5% (€375k annual savings, 375% ROI) and AI-native automation outperforming traditional OCR.

— ABBYY customer outcomes: 99% KYC compliance, 40% efficiency gain, 92% touchless processing, 140+ hours saved monthly. Backed by Gartner Magic Quadrant, Everest Group PEAK, IDC MarketScape recognition.

— Academic benchmark on 7,093 high-difficulty samples across 5 OCR tracks finds state-of-the-art LMMs exhibit substantial performance degradation in production, revealing gap between benchmarks and real-world effectiveness.

— UK government trial of 20,000 civil servants found AI saved ~2 weeks per person annually (26 min/day), with potential £45B public sector savings on 1B citizen transactions, 84% assessed as automatable.

— Everest Group PEAK Matrix 2026 identifies 10 IDP leaders (ABBYY, EdgeVerve, HCL, Hyperscience, Infrrd, Microsoft, Nanonets, Rossum, Tungsten, UiPath) across 32-vendor ecosystem; confirms maturity.

— InduOCRBench research proves high OCR accuracy does not guarantee downstream RAG effectiveness on industrial documents; structural and semantic errors cause retrieval failures despite low character/word error rates.

— Koncile guide with five independent deployment case studies: 12k invoices (65% reduction, €40k savings), 4k payslips (70% time cut, 2 FTE freed), 30k claims (halved reimbursement time).

— Independent evaluation of GPT-4o, Claude, and Gemini on 120 real financial documents reveals error clustering on scanned/low-resolution docs (6-8% failure), multi-currency (6-11 errors), and dangerous confidence calibration, requiring validation layer for production safety.

HISTORY

  • 2017: ABBYY and Kofax released intelligent capture platforms targeting high-value repetitive processes (insurance claims, invoice processing). Market consolidating around two dominant vendors. Gartner estimated 80-90% of new enterprise data was unstructured, creating demand for extraction solutions. Technology still required significant manual configuration and tuning per use case.
  • 2018: Production deployments expanded: major telecom provider achieved 400% productivity gain in invoice automation; Kofax won Ventana award for logistics automation enabling growth without headcount increase. Blue Prism + ABBYY partnership signaled ecosystem integration. Research on OCR accuracy improvements and legal automation ROI (2-3 years) documented both technical progress and practical constraints: field-level accuracy limitations required confidence scoring, and configuration effort remained substantial.
  • 2019: Analyst firms (Everest Group) formalized IDP market validation, recognizing Kofax and ABBYY as Leaders. ABBYY released FlexiCapture 12 with enhanced ML capabilities. Academic research (ICDAR 2019, mobile OCR datasets) demonstrated robust community interest in document processing, but real-world usability challenges in handwriting recognition and mobile scenarios persisted. Adoption remained concentrated in high-volume standardized processes with clear ROI; field-level accuracy constraints and 2-3 year payback periods continued limiting broader deployment.
  • 2020: Everest Group assessed 18 IDP vendors, confirming Kofax #1 for Market Impact; Forrester survey showed 58% of organizations deploying document digitization, signaling broad adoption. ABBYY released FlexiCapture SDK for developer integration, expanding ecosystem. Academic research empirically demonstrated OCR error cascades into downstream NLP tasks, quantifying accuracy-limitation burden. Handwriting recognition (Nebo, web standards proposals) showed real-world progress but remained niche. The vendor ecosystem matured with expanded integrations and adoption frameworks, but field-level accuracy and configuration complexity continued constraining broader deployment beyond high-ROI standardized processes.
  • 2021: Major cloud platform entry: Google Cloud launched Document AI to general availability with use-case-specific solutions (Lending, Procurement, Contract DocAI), signaling mainstream vendor investment. Market projected 55-65% growth with cost reduction as primary driver. Workday integrated Procurement DocAI for multi-language invoice processing, demonstrating cross-vendor ecosystem adoption. Everest Group analysis documented five major IDP adoption pitfalls, and academic research highlighted unsolved problems (table extraction, reliability demands) and implementation gaps. Cloud service reliability issues (Microsoft AI Builder timeouts) emerged in production workflows. Practice consolidated into good-practice tier: proven at scale across vendors, with clear ROI frameworks, but accuracy and configuration barriers limited adoption to high-value standardized documents.
  • 2022-H1: Cloud platforms advanced IDP capabilities: Google Document AI integrated Enterprise Knowledge Graph for entity enrichment; Microsoft Azure experienced custom model training scalability limits. Analyst coverage expanded: Everest Group's 2022 PEAK Matrix assessed 36 vendors with ABBYY as Leader for fourth consecutive year; Gartner cited $1.2B 2020 IDP market. Mortgage industry survey of 200 companies found 38% invested in IDP since 2019, with 87% prioritizing accuracy and 66% eliminating manual procedures, indicating vertical-specific adoption acceleration despite persistent deployment challenges.
  • 2022-H2: Mainstream cloud platform adoption accelerated with GA releases: Microsoft announced Unstructured Document Processing in AI Builder (164-language support); Google Document AI deployed in government (State of Hawaii processing 25,000+ visitor documents daily). Market validation confirmed rapid expansion: analyst forecasts ranged $1.1B→$5.2B (37.5% CAGR) to $2B→$3.5-4B (15.9% CAGR), with adoption driven by cost reduction and digital transformation. However, real-world deployment barriers persisted: ISG reported awareness gaps and compliance challenges; users reported OCR robustness limitations on specific document types (e.g., lottery tickets). Practice solidified in good-practice tier with proven cloud platforms, clear vendor competition, and documented enterprise use cases, balanced against configuration complexity and accuracy constraints limiting broader deployment.
  • 2023-H1: Cloud platform vendors advanced IDP with generative AI: Azure Form Recognizer previewed document classification and Azure OpenAI integration for natural language extraction; AWS demonstrated dialogue-guided IDP with foundation models (Textract + LLM). Government-scale deployment validated production readiness: OPAIDA won IRS competitive selection to modernize 500M+ paper tax returns with ~99% accuracy. Adoption data from ABBYY's 10,000+ customers revealed regional priorities and growing demand for RPA ecosystem connectors. Market forecasts escalated to $18.87B by 2031 at 32% CAGR. Generative AI integration emerged as competitive differentiator, while field-level accuracy and configuration barriers continued limiting deployment breadth.
  • 2023-H2: Cloud platforms matured IDP with GA releases: Microsoft released Azure AI Document Intelligence v3.1 with document classification, prebuilt contract models, and 47-language custom neural model support; Google GA'd generative AI extraction in Document AI Workbench with named enterprise deployments (Deutsche Bank KYC, BBVA complex document handling). ABBYY maintained analyst leadership (Everest Group Leader for fourth year) with 10,000+ customer base. However, production reliability challenges surfaced: Google Document AI experienced outages with HTTP 499/504 errors; Azure API regressions caused accuracy issues in table/entity detection, signaling maturation challenges despite market growth projections ($18.87B by 2031). Practice remained good-practice with proven deployments at scale, yet reliability and accuracy constraints limited adoption breadth to high-value standardized processes.
  • 2024-Q1: Platform maturation continued with ongoing vendor innovation balanced against emerging reliability constraints. Google Document AI custom extractor training UI experienced multi-region outages (January 2024), while regional API limitations persisted for Azure preview features (East US, West US2, West Europe only). Academic research advanced OCR techniques with transformer-based models achieving improved accuracy on mixed handwriting and scene-text recognition. Market projections escalated to 23.7% CAGR through 2031, with cloud-based solutions and BFSI sectors driving adoption; Dociphi launched on Google Cloud Marketplace. Practitioner evaluations showed Azure Document Intelligence at 1.5 cents per page with >99% accuracy on medium-sized text, though accuracy degradation remained severe below 7px character size. Reliability barriers and regional constraints continued limiting enterprise deployment velocity, offsetting strong market demand signals.
  • 2024-Q2: Vendor capability advancement continued with Microsoft adding hierarchical document structure and figure detection to Azure AI Document Intelligence (April 2024); AWS and Google maintained respective platform positioning. Government-scale deployment success: European Patent Office achieved 400K daily patent page processing with <1% OCR error and 5-day→minutes lead time reduction, confirming complex document automation feasibility at scale. Production reliability remained problematic: Azure experienced severe latency issues in East US due to capacity constraints (April 2024), requiring operational workarounds. Everest Group's 2024 market assessment (June 2024) confirmed strong adoption momentum in banking and insurance with market growing at 23.7% CAGR through 2031. Industry analysis acknowledged reality: partial process automation (50-70% coverage) delivered meaningful ROI without requiring complete end-to-end automation. Field-level OCR accuracy degradation below 7px remained unsolved; configuration complexity and accuracy bounds continued limiting breadth beyond high-value standardized processes. Good-practice tier sustained by proven deployments and clear ROI frameworks despite persistent reliability and accuracy constraints.
  • 2024-Q3: Vendor competition intensified with Microsoft cutting custom extraction pricing 40% to $30 per 1,000 pages (July 2024), signaling market-driven adoption incentives. Independent analyst assessments validated IDP market maturity: IDC's MarketScape assessed 16 vendors recognizing leaders including ABBYY, Google Cloud, and Tungsten Automation; market had reached $7B in 2023 at 15% YoY growth with double-digit CAGR through 2028. Generative AI and RAG integration emerged as vendor competitive differentiators (ABBYY, Google, Microsoft). Production reliability challenges intensified: Google Document AI custom model training experienced widespread September 2024 failures requiring vendor fixes; Azure latency constraints persisted in East US region. Field-level accuracy and reliability remained binding adoption barriers. Good-practice tier sustained by proven deployments and competitive vendor landscape, balanced against persistent production reliability constraints and accuracy limitations in real-world deployment.
  • 2024-Q4: Cloud platform vendor roadmaps advanced IDP capabilities with Microsoft releasing v4.0 GA (October 2024) featuring batch API support across all models, custom classifier incremental training, and expanded prebuilt models. Enterprise deployments demonstrated continued strong ROI: case studies showed 20-50% cost reduction in financial services, 50-400% capacity improvement in high-touch processes (RFP response), and 50-75% cycle time reduction in insurance document processing. Market research confirmed growth momentum: Mordor Intelligence forecast IDP market to reach USD 7.18B by 2031 at 17.78% CAGR with cloud deployments capturing 74.10% share. However, production reliability constraints persisted: comparative testing revealed handwriting OCR accuracy disparities (0.9%-23.3% WER) across platforms with Google Document AI exhibiting text ordering failures in handwritten inputs. Critical assessments documented limitations of generative AI approaches for structured extraction: Azure OpenAI struggled with tabular data vs. specialized prebuilt models, suggesting LLM-based IDP remains complementary to specialized extraction models rather than a universal replacement. Technical research (DAML 2024) confirmed ongoing trade-offs in OCR approaches (HMM efficiency, CNN feature extraction, LSTM temporal modeling) without breakthrough solutions to fundamental accuracy constraints. Good-practice tier sustained by proven enterprise adoption and continuous capability advancement, balanced against persistent field-level accuracy limitations and reliability incidents constraining broader deployment beyond high-value standardized processes.
  • 2025-Q1: Cloud platform vendors advanced IDP with AI agent integration: AES deployed AI agents for health and safety audit automation with 99% cost reduction and 14-day→1-hour acceleration on 400-page documents. Deep Analysis analyst report (surveying 57 IDP companies) forecast double-digit market growth through 2028, identifying AI agents and generative AI as disruptive factors. Azure Document Intelligence 4.0 GA released for Power Platform integration. Adoption barriers persisted: Deloitte research documented 70% of enterprises struggling to move beyond 30% of AI experiments to production, with compliance and accuracy constraints limiting deployment breadth. Handwriting recognition remained problematic: University of Zurich abandoned handwriting OCR for exam grading due to stress-induced poor quality and complex formatting. Good-practice tier sustained by continued deployment innovation and vendor capability expansion, balanced against persistent production readiness and accuracy limitations constraining broader adoption.
  • 2025-Q2: Market momentum accelerated with analyst consensus on sustained growth: Everest Group's mid-year comprehensive market analysis forecast continued expansion, while Technavio projected aggressive 46.9% CAGR through 2029 driven by North America and BFSI adoption. Forrester's commissioned ROI study (284% return on investment) validated economic case for enterprise IDP deployment. Technical landscape matured with expanded Vision-Language Model integration alongside traditional OCR approaches, as IntuitionLabs 2025 analysis documented ecosystem evolution. However, production reliability constraints persisted: Azure Document Intelligence custom neural model training failures continued into June 2025, with users reporting unresolved InternalServerError issues affecting advanced deployment scenarios. Critical assessment of handwriting recognition remained sobering: evaluation showed 2025 best-case accuracy of 95%+ on clean Latin text degrading sharply below 80% for non-Latin scripts (Cyrillic 85-91%, Arabic 80-88%, Chinese 75-82%), documenting fundamental technological limitations constraining multilingual document processing deployments. Good-practice tier sustained by strong analyst validation and continued enterprise deployment momentum, offset by unresolved service reliability issues and script-specific accuracy limitations blocking broader geographic and linguistic adoption.
  • 2025-Q3: Enterprise adoption momentum accelerated with AIIM 2025 survey confirming 78% operational deployment across 600 enterprises (US, Germany, Austria, Switzerland), signaling mainstream market penetration despite persistent barriers. AWS and cloud vendors advanced IDP capabilities: AWS released GenAI IDP Accelerator with production case studies showing Competiscan achieving 85% accuracy across 35,000-45,000 daily documents in 8 weeks and Ricoh processing 10,000+ healthcare documents monthly with 1,900 person-hours annual savings potential. However, adoption barriers intensified alongside capability expansion: 61% of IDP workflows still rely on paper documents, 48% expect paper volumes to increase, and critical assessment revealed most enterprises operate rule-based automation rather than true AI intelligence. Fundamental LLM reliability issues documented: Dr. Hardman's experimental analysis showed 80% failure rate (4 of 5 models) on document comparison tasks, with GPT-4o, Gemini, and Copilot hallucinating differences in identical documents. Academic research (systematic review of 1,302 HTR studies) and practitioner assessments documented persistent handwriting recognition barriers: error rates of 3-5% in English, acute challenges for non-Latin scripts, high development costs, and privacy concerns limiting broader adoption. Good-practice tier sustained by proven deployment case studies and vendor momentum, balanced against deepening assessment of adoption barriers, paper persistence, LLM unreliability in document tasks, and fundamental technical limitations in handwriting recognition constraining breadth beyond specialized high-value operations.
  • 2025-Q4: Cloud platform maturity advanced with major GA releases: Google Cloud released generative AI custom extractor with Gemini 2.0/2.5 Flash models (October 2025); Microsoft released Azure Document Intelligence v4.0 with expanded prebuilt models (November 2025). Thoughtworks Technology Radar assessed Azure AI Document Intelligence as 'Assess'-tier with reported reduction in manual data entry and improved accuracy despite latency trade-offs. Handwriting recognition showed experimental improvements (Gemini 3 Pro achieving perfect transcription on historical documents) but critical assessments documented persistent limitations with ~95% best-case accuracy and sharp degradation on cursive and non-Latin scripts. Market data signaled maturity: global IDP demand reached $8B in 2024 at 14.5% growth with 16% CAGR forecast through 2029. Infrastructure gap exposed: Apryse survey of 465 organizations revealed 64.5% have AI in production yet only 38.1% rate document data as 'excellent', indicating widespread deployment but persistent data quality challenges limiting full automation. Good-practice tier sustained by vendor competition and adoption momentum, balanced against unresolved infrastructure readiness gaps and fundamental handwriting accuracy limitations constraining multilingual and complex document deployments.
  • 2026-Jan: Cloud platforms and named-org deployments demonstrated continued adoption momentum alongside emerging critical assessments of pilot failure rates. Market maturity signal: AIIM/Deep Analysis survey confirmed 78% of enterprises operational with AI document automation and 66% of new IDP projects replacing legacy systems. Vendor platform evolution: Tungsten Automation released TotalAgility 2026.1 (January 2026) with LLM-powered classification and on-demand processing. Named-org deployments showed expanding adoption: Google's internal production deployment automated sustainability report processing using NotebookLM/Gemini with claims validation; EY scaled tax processing pipeline to hundreds of extractors mixing OCR, custom models, and generative augmentation with audit traceability; logistics firm achieved 90% time reduction (200→20 hours monthly) and 35% error reduction across 50,000 documents. Handwriting recognition advanced with Learnable.ai production deployment in gaokao exam grading (13M+ students) achieving higher accuracy than human graders. Critical assessment emerged: MIT Sloan analysis documented 95% failure/stall rate of enterprise generative AI pilots, signaling that platform capability advancement has not yet overcome pilot-to-production conversion barriers. Production reliability challenges continued with user-reported failures across Google Document AI and Azure Document Intelligence. Good-practice tier sustained by documented platform maturity and named-org deployment success, balanced against persistent pilot failure rates, production reliability incidents, and unresolved barriers to scaling beyond high-value standardized workflows.
  • 2026-Feb: Agentic processing reached mainstream adoption evaluation (67% of enterprises per Gartner, up from 23% two years prior) with compliance-first design and workflow orchestration emerging as differentiators. Platform maturity advanced: Azure Document Intelligence v4.0 GA released searchable PDF and incremental classification training; Google maintained procurement-focused solutions with 60% cost reduction claims. Named-org deployment continued: KumoHQ case study documented logistics firm achieving 87% time reduction (40+ hours→5 hours weekly) and 94% accuracy via LLM-powered extraction with 3.5-month payback. However, critical failures exposed fundamental reliability limits: DOJ/House Oversight Committee released 3M+ PDFs with non-functional OCR, rendering them unsearchable and contradicting full-automation narratives; Azure Document Intelligence experienced custom classification training failures with jobs stuck at 'notStarted' status since January 30. Critical assessments documented adoption barriers: Vertesia survey (1,500 IT executives) found 96.8% report ECM vendor roadmaps as significant barrier to AI implementation. Parashift analysis quantified hidden costs: manual correction loops represent 40% of input management, though AI reduces IT maintenance by 90% and error rates from 4% to 0.5%, achieving 3.5-month ROI in optimized deployments. Good-practice tier sustained by continued named-org success and platform feature expansion, balanced against emerging evidence of production reliability incidents, fundamental OCR accuracy limits exposed at scale, and structural adoption barriers in enterprise procurement workflows.
  • 2026-Mar: Deployment economics are well-documented but accuracy constraints sharpen as the critical boundary. Named production outcomes show strong ROI at scale — Esprigas processes 27,000 documents/month at $73,800/month savings, Erewhon 20,000 invoices at $45,000/month — with industry benchmarks confirming 60-80% cost reduction and 6-18 month payback across lending, insurance, and BPO. Tungsten Automation (formerly Kofax) serves 8 of the top 10 global banks, confirming enterprise-scale breadth. However, a clear accuracy threshold emerges: 96-99% field-level accuracy is required for viable ROI, while LLM-only approaches fail in production due to fluency-masking errors and layout collapse; handwriting recognition degrades from 3-8% CER on printed text to 15-40% on handwritten inputs, blocking deployment in roughly 30% of regulated-industry workflows.
  • 2026-Apr: Vendor maturity and production deployments accelerated alongside rising critical assessments of accuracy and reliability limits. IDC MarketScape (April 11, 2026) named 8 IDP leaders (ABBYY, Google, Hyland, Hyperscience, Open Text, SER, Tungsten, UiPath), signaling vendor ecosystem consolidation with GenAI and agentic AI as dominant differentiators. AIIM/Deep Analysis independent survey of 600+ organizations found 65% actively ramping document processing initiatives, indicating market inflection toward standard adoption. Real-world production deployments continue delivering strong ROI: Quantiva case studies show financial services firm at 98% accuracy with 5x productivity, film studio with 90% time reduction, music platform cutting distribution from 48 to 30 minutes; Rossum customers report 90% time reduction and 60% straight-through processing; Disney Trucking processes 360k handwritten tickets annually. Deployment economics documented: $2.8B market at 35% CAGR with 6-12 month payback, 93% time reduction, 62% cost reduction; IOFM benchmarking shows 9x performance gap ($2.07-$18.42 per invoice) with IDP in best-in-class tier. However, critical assessments deepen: Bluente benchmark shows state-of-the-art OCR models score below 50/100 on document fidelity tests; LandingAI documents production failure modes (split tables, inconsistent formats, lost context); handwriting OCR benchmark on 5,578 medical prescriptions documents real-world limitations. Good-practice tier sustained by strong vendor competition, proven named-org deployments, and rising adoption rates, balanced against persistent accuracy constraints, production reliability incidents, and documentary evidence of extraction failures at scale.
  • 2026-May: Deployment ROI benchmarks and accuracy constraints sharpen the practice boundaries. Lleverage case studies document manufacturing firm reducing FTE from 4 to 1 with error rate improvement from 7% to 0.5% (€375k annual savings, 375% ROI), while survey data (Koncile, IOFM) confirms 60-80% cost reduction and industry standards of €2.78-12.88 per invoice depending on platform and process maturity. Deployment adoption reached critical mass: Everest Group PEAK Matrix 2026 evaluation identifies 10 leaders across 32-vendor ecosystem, and UK government trial confirms scaling potential (20,000 civil servants, 2 weeks/person annual savings). However, academic research surfaces persistent production barriers: CC-OCR V2 benchmark on 7,093 high-difficulty samples finds state-of-the-art LMMs exhibit substantial performance degradation in real-world conditions; InduOCRBench research proves that high OCR accuracy on conventional benchmarks does not guarantee downstream RAG effectiveness, with structural/semantic errors causing failure despite low character error rates. Accuracy threshold for viable ROI remains 96-99% field-level performance; production accuracy gaps and benchmark-to-deployment variance (55+ percentage points across document types) persist as binding constraints. Good-practice tier sustained by strong vendor competition, documented deployment ROI, and mainstream adoption momentum, balanced against unresolved accuracy limitations, production-environment performance gaps, and evidence that state-of-the-art models fall short of production requirements in real-world document processing.

TOOLS