
ABBY
Gartner MQ Leader 2025; consistent recognition 10+ years
Global enterprise finance, insurance, healthcare (specific names not confirmed this cycle
90% out-of-box; up to 95%+ with FastML continuous learning
No formal named guarantee found
90%+ from day one; up to 95% achievable
High STP achievable; no distinct named zero-touch tier found
90% out-of-box from 150+ pre-trained skills
Yes — DeepML + FastML; no templates
150+ pre-trained skills: invoices, IDs, contracts, tax forms, and more
Yes — 37yr ICR (Intelligent Character Recognition) heritage; core capability
Yes — multi-document pipeline; LLM via RAG
Long IP development history; count not publicly disclosed
DeepML + FastML + LLM via RAG processing modes; no named buyer-facing tiers
Yes — low-code/no-code platform (Vantage meta description confirmed)
Yes — FastML learns from outlier samples
Yes — HITL review stage with routing built into process skill flow
Yes — cloud-native; AWS, Azure, Google Cloud
Yes — strong on-prem (FlexiCapture); Vantage also supports on-prem containers
Yes — REST APIs confirmed
BPM, ERP, ECM, CRM connectors; specific list not confirmed this research cycle
Minimal — DeepML pre-trained; FastML from few samples
Yes — enterprise sales process; free trial/sandbox not confirmed
Finance, insurance, healthcare, manufacturing, legal, tax
Avg contract ~$150K/yr (Vendr); quote-based enterprise; no public price list

Docsumo
G2 Leader in IDP (badge confirmed on pricing page)
~7 years (founded ~2019; Sequoia-backed)
Hitachi Payments, PayU, National Debt Relief, Arbor Realty Trust, Biagi Bros, Carbon Direct
>99% claimed (meta); >95% post-training; >90% with pre-built models
No formal named guarantee found
95%+ STP (National Debt Relief case study; pricing page stat)
Yes — Straight-Through Processing is a named product feature
>90% with 30+ pre-built models (plug and play, no training needed)
Yes — 30+ pre-built models; no templates
30+ models: bank statements, ACORD, invoices, checks, utility bills, W9, pay slips
Yes — cross-document validations (Enterprise tier); master data matching
In-house tech confirmed ('no data leaks to third-party models'); count not disclosed
Free/Business/Enterprise product tiers; STP is a named feature within all plans
Yes — 30+ plug-and-play pre-built models; go live within days
Yes — Continuous learning listed in all pricing tiers (including Free)
Yes — HITL routes failed extractions; validation checks; approval workflows
Yes — SaaS (cloud-native)
Yes — webhooks, APIs, Excel downloads (all tiers including Free)
Salesforce, Zapier (500+ apps), Google Sheets, Yardi, QuickBooks, Xero, Zendesk, OneDrive
Zero for 30+ pre-built models; 20 samples for custom models
Yes — 14-day free trial, 1,000 pages (no credit card required)
Yes — Dedicated AM & Solutions Engineer (Business plan, confirmed on pricing page)
Commercial Lending, Financial Services, Healthcare, Commercial Banking; Lending DocAI product
Free (14-day trial) → Business (custom) → Enterprise (custom); setup fees apply

Hyperscience
Gartner MQ Leader 2025; Forrester Wave Leader Q2 2024; IDC MarketScape Leader 2024; GigaOm Radar Leader 2025
HMRC, US Social Security Administration, American Express, Charles Schwab
99.5% claimed in production with named enterprise clients
Yes — Accuracy Harness: buyer defines SLA target; platform autonomously meets it
98% automation claimed; Hyperscience prefers Average Handling Time over STP% as metric
Yes — via Accuracy Harness (98% automation supports lights-out processing)
ORCA VLM zero-shot; 99.5% day-1 accuracy for supported production workflows
FedRAMP High (via Palantir FedSTART)
HIPAA; data masking; full audit trails
SNAP, freight pay, mortgage, invoices, paystubs; count not disclosed
Yes — 98% accuracy on handwriting/cursive in low-quality scans (IDC MarketScape cited)
Yes — Document Mining; LLM-ready output; long-form cross-document context
ORCA VLM is proprietary; specific patent count not publicly disclosed
ORCA VLM + AI-in-the-Loop + HITL; automation governed by Accuracy SLA setting
Yes — no data scientists required; AI-guided co-pilot manages model lifecycle
Yes — continuous learning through human feedback
Yes — intelligent routing; Accuracy SLA-driven orchestration
Yes — SaaS on AWS and Google Cloud
Yes — full on-prem + air-gapped via Google Distributed Cloud and HPE
AWS S3, Azure Blob, Google Cloud, SAP, Salesforce, Microsoft 365, IBM FileNet, UiPath, MuleSoft
Zero — ORCA VLM zero-shot; trainable models also available
Yes — 'Book a Meeting' / demo library; expert consultations
Energy, Financial Services, Healthcare, Insurance, Legal, Manufacturing, Public Sector, Retail, Transportation & Logistics
Volume-based per-page; decreasing unit cost at scale; no public price list

Infrrd
Gartner MQ Leader 2025; Everest PEAK Star Performer (4th year); IDC Leader 2023; GigaOm Leader & Outperformer 2025
9 years (founded 2016, San Jose CA)
Fortune 500 mortgage lenders, top-10 US insurers, global logistics carriers
95%+ automated mode; 100% with HITL-backed Accuracy Guarantee Program
Yes — Accuracy Guarantee Program: Basic / HITL / NTP tiers (unique in market)
70%+ consistent STP in standard deployments
Yes — No-Touch Processing (NTP) mode: named buyer-facing tier
95%+ on supported document types before training
Yes — visual AI; no templates required
Mortgage (ACORD), insurance, logistics, finance, healthcare + Lending DocAI product
12 patents in visual AI and computer vision (unique across all 9 vendors in this comparison)
3 tiers: Basic / HITL / No-Touch Processing (NTP) — unique named buyer-facing model
Yes — self-service model setup in natural language; no data scientists required
Mortgage platforms, DMS, SAP, Salesforce, major ERPs
Minimal — self-service setup from document samples
Business hours + enterprise 24/7 options
Mortgage/lending, insurance, logistics, financial services, healthcare
Custom enterprise pricing; ROI calculator available at infrrd.ai

Nanonets
#1 on IDP Leaderboard ahead of GPT-5, Gemini, Claude. No Gartner/Forrester/IDC found.
~9 years (founded 2017, San Francisco CA)
Volkswagen, Schneider Electric, Ryanair, Mondelez, Roche, P&G, Juniper Networks
#1 on IDP Leaderboard; 94% of processes close without human
No formal named guarantee found
94% of complex processes close without human (homepage)
Yes — agents default to full automation; 94% complete without human
High — OCR-3 is zero-shot model (#1 IDP Leaderboard)
Yes — OCR-3; 'No templates' confirmed on homepage
AP, Order Mgmt, Logistics, Healthcare RCM, Contracts, Vendor Onboarding; count not disclosed
Yes — 3-way PO matching; agents operate across systems of record
OCR-3 proprietary; open-source models on HuggingFace; count not disclosed
Starter/Growth/Enterprise pricing tiers; agents default to full automation
Yes — no-code/low-code agent workflow builder
Yes — agents learn from configuration and feedback over time
Yes — configurable approval gates; route edge cases to Slack/Teams/email
Yes — cloud-native; private cloud/on-prem also in Enterprise
Yes — private cloud / on-prem deployment (Enterprise tier)
Yes — API from Starter; Agentic Data Extraction API; LangChain/LlamaIndex
SAP, QuickBooks, Xero, Sage, Salesforce, HubSpot, Google Drive, Slack, Teams, Snowflake, Jira
Zero — OCR-3 zero-shot model
Yes — free trial with $200 credits (no credit card required)
Yes — dedicated support (Enterprise tier confirmed on pricing page)
Custom SLAs (Enterprise tier confirmed on pricing page)
Accounts Payable, Order Management, Logistics, Healthcare RCM; enterprise-broad
Pay-per-block: $0.02 (simple) to $0.30 (complex AI); $200 free credits; 40% volume discount

Ocrolus
Analyst-recognized leader in AI financial document automation (Everest Group, IDC, HousingWire)
~12 years (founded 2014, New York NY)
Enova, LendingClub, Kapitus, PayPal, SoFi, Zillow, Better.com, Brex, CrossCountry Mortgage
99%+ for financial documents (bank statements, pay stubs, tax forms)
No formal named guarantee found
the platform maintains a 99% classification accuracy across 1,700 document types
Implied via high automation; no named zero-touch mode found
High for financial docs; pre-trained on lending documents from day 1
Yes — ML-based; no template configuration required
Financial docs: bank statements, pay stubs, W-2, 1040, 1099, K-1, business financials
Yes — fraud detection across docs; cash flow analysis across multiple bank statements
Robust portfolio of proprietary AI technology and domain-specific intellectual property
No named tier model; configurable confidence thresholds
Partial — API-first developer design; dashboard also available for business users
Yes — model continuously improves; HITL corrections feed back
Yes — documents below confidence threshold routed for human review
Yes — cloud-native on AWS
Yes — API-first; docs at docs.ocrolus.com
LOS (Loan Origination System) integrations; bank data providers; lending analytics platforms
Zero — pre-trained on financial lending documents
Yes — free account creation at app.ocrolus.com/signup
RTO 4 hours confirmed (security portal); specific SLA not published
SMB Funding, Mortgage, Auto Finance, Consumer Lending, Legal, Medicaid, Tax, Tenant Screening
Custom; outcome-based unit economics; no public pricing

Rossum
G2 Leader Fall 2024; Forrester Strong Performer Q2 2024; IDC Leader 2023-24; Everest Star Performer 2024. Note: Acquired by Coupa 2026.
~9 years (founded 2017; acquired by Coupa 2026)
Panasonic, Siemens, Bosch, Celonis, Wolt, Master Trust Bank of Japan, Adyen, Kingfisher
93% overall; 94–97% on invoices and transactional documents
No formal named guarantee found
Variable; Wolt case study: 60% STP. 82% time saved on validation used as metric.
No named zero-touch mode; platform minimises human review over time
High for invoices/POs via Aurora LLM; less certain for non-transactional types
Yes — Aurora LLM; 100% template-free
Aurora LLM for transactional: invoices, POs, bills of lading, customs
98% accuracy achieved on handwritten and degraded text
Yes — duplicate detection, fraud detection, AI Agents for multi-document
In-house models confirmed; count not disclosed
Single platform; automation increases as Aurora learns; no named tiers
Yes — managed by business users; low-code extensions
Yes — learns from every human hover, keystroke, mouseclick
Yes — flexible queuing system; trigger-based communications
Yes — cloud-native SaaS on AWS (EU/US/Japan regions)
Yes — API & SFTP access (all plans including Starter)
SAP, Coupa, NetSuite, Workday, Microsoft Dynamics; 200+ integrations/partners
Minimal — Aurora pre-trained; learns from corrections in real-time
Yes — 14-day free trial publicly available
Yes — Signature onboarding & success plans (Enterprise); dedicated project teams
Custom SLAs (Enterprise tier); community support (Starter)
Wholesale Distribution, Manufacturing, Construction, Logistics, Retail & CPG, Technology, Shared Service Centers
Tiered subscription: Starter from $18K/yr (confirmed); Business/Enterprise custom

Textract
Forrester: 73% ROI report cited. Not ranked in standalone IDP analyst reports (extraction API, not full IDP).
~8 years as product (Textract GA 2018; AWS founded 2006)
AWS customers broadly; specific Textract users not publicised
Returns confidence scores 0–100 per element; no single % claimed
No — confidence scores returned; buyer builds own thresholds
Textract does not have a fixed, built-in Straight-Through Processing (STP) rate
High for supported APIs (Expense, ID, Lending); no training required
Yes — ML-based; 'without templates or configuration' stated for Expense/ID/Lending APIs
5 APIs: Detect Text, Analyze Document, Analyze Expense, Analyze ID, Analyze Lending
Yes — 'automatically extracts printed text, handwriting, layout elements' (product headline)
No — single-document API; cross-document correlation must be built by developer
Amazon holds thousands of AWS patents; Textract-specific count not disclosed
No — requires developer implementation for all integrations and workflows
Partial — Custom Queries adapters can be retrained; standard models don't retrain from corrections
No — must be built by developer using Lambda/Step Functions
Yes — AWS-managed service (15+ global regions)
No — fully managed AWS cloud service; GovCloud available for US federal
Yes — AWS SDK and REST API; core API-first architecture
AWS ecosystem (S3, Lambda, Comprehend); no pre-built ERP/CRM connectors
Zero for standard APIs; Custom Queries adapters need labelled training data
Yes — AWS Free Tier: 3 months; 1,000 pages/month for Detect Text
Yes — AWS Pricing Calculator (calculator.aws) — cost/page calculator (cost, not business ROI)
No — no dedicated AM for Textract; TAM covers full AWS account at additional cost
AWS Support: Basic (no SLA) → Enterprise ($15K/mo, 15min critical response)
Financial Services, Healthcare & Life Sciences, Public Sector
Transparent per-page: $0.0015–$0.07/page by API; AWS Pricing Calculator available

True.ai
Lacks major independent analyst rankings
~9 years (founded 2017 as SoftWorks AI; rebranded TRUE 2022)
Fairway (400+ locations), MGIC, Arch MI, Arvest, First Continental (Rocket Mortgage historical)
Enables enterprise-grade, 97.5%+ accurate data extraction for lending documents, reducing manual review.
No formal named guarantee found
90%+ STP confirmed on homepage (primary headline metric)
Yes — 90%+ STP; 'background automation that works behind the scenes'
High for mortgage docs; fully pre-trained on mortgage workflow
Yes — pre-trained on mortgage documents; no templates
Mortgage-only: loan applications, income docs, title, appraisal, insurance, closing
Yes — condition matching and clearing across full loan file documents
Academic AI lab origin; count not disclosed
No named tier model; 90%+ STP is the standard operating state
Yes — 'works behind the scenes; no workflow disruption'; business users operate
Yes — condition matching; auto-cleared or routed; mortgage SLA management core
Yes — pre-built, cloud-delivered
Candor, Tavant partners confirmed; LOS integrations; out-of-box connectors
Zero — pre-trained on full mortgage workflow
Yes — demo request (sales-led; no self-service free trial found)
Yes — SVP Technical Account Management is a named C-suite leadership role
Mortgage ONLY — Lending Originators, Wholesalers & Correspondents, Insurers
Custom; demo-led; volume-based implied; no public pricing page

Tungsten Automation
Gartner MQ Leader 2025; 25,000+ customers in 32 countries.
~41 years (founded as Kofax 1985; rebranded Tungsten Automation ~2023)
ASN Bank, Toshiba, Marion Superior Court; 25,000+ organisations in 32 countries
Yes, consistently achieve 90% to over 99% accuracy on various document types,
No formal named guarantee found
Achieves a 95% straight-through processing (STP) rate for enterprise document processing and validation
Yes — STP achievable; no named zero-touch tier found
Tungsten AI delivers 97%–99% Day 1 accuracy, with 95%+ baseline precision for Australian enterprise documents.
Yes — AI-powered document intelligence; no templates
3,000+ pre-trained models for ready-to-use extraction across global documents like invoices, tax forms, and receipts.
Yes — Kofax ICR is a 40yr legacy capability; confirmed through heritage
Yes — Knowledge Discovery module; document correlation within workflow
41yr IP portfolio (Kofax + Tungsten); count not publicly disclosed
Tiered plans with flexible processing modes
Yes — low-code process orchestration; demo centre at tungstendemocenter.com
Yes — ML retraining implied through Gartner MQ recognition and mature platform
Yes — intelligent routing; eliminates bottlenecks (confirmed on product page)
Yes — AWS, Azure, Google Cloud
Yes — strong on-prem heritage from 41yr Kofax history
Yes — TotalAgility REST APIs; API-first integration architecture
SAP, Microsoft 365, SharePoint, MuleSoft, Azure; broad ERP/ECM through 41yr partner network
Yes — demo centre at tungstendemocenter.com (sales-led)
Yes — global partner/support network; dedicated enterprise sales & technical engagement
Banking/Financial Services, Supply Chain, Government, Healthcare, Insurance; 32 countries
Custom enterprise; quote-based; no public pricing