Skip to content
Vertical AI Eats the Stack: Acquisitions, Agentic Agents, and the End of Horizontal IDP?
NEWS 17 min read

Vertical AI Eats the Stack: Acquisitions, Agentic Agents, and the End of Horizontal IDP?

The intelligent document processing market crossed a structural threshold in early 2026: vertical AI platforms are absorbing horizontal IDP vendors faster than those vendors can respond. UiPath's acquisition of WorkFusion on February 6, absorbing pre-built financial crime compliance agents backed by $395 million in cumulative venture funding, was the most consequential IDP deal so far this year. It caps an accelerating consolidation wave: since 2023, at least seven other horizontal IDP vendors have been absorbed into vertical software stacks: Infinia ML into Aspirion (September 2023), Metamaze into Duco (February 2024), Novodoc into Archive-IT (February 2024), Eigen Technologies into Sirion (June 2024), Semantha into Aleph Alpha (April 2025), AntWorks into GTT Data Solutions (August 2025), and Insiders Technologies into Proalpha (September 2025). Each confirms the same pattern: standalone document processing platforms are becoming features inside vertical software stacks, not products in their own right.

Every section of this report reinforces that thesis. The winners are platforms with deep vertical workflows, proprietary data moats, and agentic orchestration. The losers are general-purpose extraction tools competing on accuracy alone.

Welcome New Vendors

We welcome Cambrion to our vendor directory, a Munich-based agentic AI platform founded in 2024 that uses vision-language models for zero-shot document processing without OCR, currently closing its pre-seed round. Also new: Google's LangExtract, an open-source Python library for LLM-powered structured information extraction with precise source grounding; LlamaParse, LlamaIndex's GenAI-native document parsing platform that has processed 500M+ documents with multimodal capabilities across 90+ formats; and PaperQA Nemotron, Future House's open-source scientific document processing platform combining PaperQA2's RAG capabilities with NVIDIA Nemotron models for superhuman performance in research tasks.

IDP Consolidation Wave: Eight Deals in Two Years

UiPath's February 6 acquisition of WorkFusion (covered in detail below) is the latest in an accelerating pattern of vertical platforms absorbing horizontal IDP vendors. Eight acquisitions over the past two years trace the same arc: a sub-scale horizontal IDP vendor exits to a vertical software platform that needs document intelligence as a feature, not a product.

The timeline tells the story. Infinia ML sold to healthcare RCM provider Aspirion in September 2023, redirecting its general-purpose ML extraction entirely to healthcare revenue cycle management. Metamaze sold to Nordic Capital-backed Duco in February 2024, merging its no-code unstructured data extraction with Duco's proprietary matching engine for post-trade financial operations. Michael Chin, CEO of Duco, said: "They can avoid point solutions and automate the biggest time wastes, such as manual data entry, manual data validation and reconciliation." (Fintech Futures). Novodoc was absorbed into Archive-IT the same month, consolidating German archiving and digitization.

The highest-stakes deal came in June 2024: Eigen Technologies joined Sirion, pushing Sirion's valuation to $1 billion with TPG Capital and Warburg Pincus in preliminary talks for a $500M+ controlling stake. The combined entity manages 3.5 million contracts across 100+ countries. "This combination creates the world's largest labelled contract dataset," said Ajay Agrawal, Founder and CEO of Sirion (Finance Director Europe). The contract analysis market, valued at $3.61 billion in 2024 and projected to $11.95 billion by 2033, is the prize.

The pace quickened in 2025. Semantha, the German semantic document platform deployed in automotive, finance, and public administration, was acquired by Aleph Alpha in April 2025 and folded into PhariaAI. Aleph Alpha's second acquisition after Lengoo (enterprise translation) signals a deliberate full-stack European document intelligence assembly: translation plus semantic understanding, unified under GDPR-sovereign infrastructure. AntWorks sold to GTT Data Solutions in an all-stock deal in August 2025, adding approximately $2.4M ARR, with the completed integration announced in February 2026. Insiders Technologies, a 27-year DFKI spin-off with 6,000+ customers, was absorbed into Proalpha's Industrial AI Platform in September 2025, following the same logic: ERP vendors acquiring document AI rather than building it.

The accounts payable automation and legal document automation markets are the next consolidation targets. Watch for CLM and ERP vendors acquiring remaining independent specialists in those categories.

Who wins: Vertical software platforms (ERP, CLM, ECM, RCM) that acquire rather than build; enterprise buyers who get document AI embedded in existing workflows without integration projects; PE firms with portfolio companies in regulated verticals.

Who loses: Horizontal IDP vendors at sub-$20M funding competing as standalone products; buyers who shortlisted any of the acquired vendors as best-of-breed point solutions; IDP vendors whose differentiation rests on accuracy metrics alone, without vertical workflow depth.

UiPath's Vertical Bet: Financial Crime and Healthcare in the Same Month

UiPath executed the most consequential vertical expansion of the period, acquiring WorkFusion on February 6 and launching three healthcare agentic AI solutions on February 23, just seventeen days apart.

The WorkFusion acquisition brings pre-built AI agent libraries for AML alert triage, KYC operations, sanctions screening, adverse media monitoring, and transaction monitoring investigations, with confirmed enterprise customers including BMO, Deutsche Bank, and Raymond James. "Financial institutions need intelligent solutions to combat sophisticated financial crimes," said Daniel Dines, CEO of UiPath (UiPath Newsroom). WorkFusion had raised $395 million in total venture funding across 13 rounds, including a $45 million Series G just five months before the exit, yet still sold to a larger platform for undisclosed terms. That sequence is the clearest evidence yet that financial crime compliance AI cannot survive as a standalone category.

The healthcare launch at ViVE 2026 introduced Medical Records Summarization, Claim Denial Prevention and Resolution, and Prior Authorization. These three workflows represent the highest-friction document bottlenecks in U.S. healthcare. The Medical Records Summarization outcome is the most specific metric in this report period: medlitix reports average summary review time dropped from 70 minutes to 6 minutes, a 90% reduction (UiPath IR). The Prior Authorization solution partners with Genzeon, one of six CMS-selected WISeR Model vendors covering 100+ healthcare clients and 30+ disease-specific clinical models, a credentialing moat that pure-play IDP vendors cannot quickly replicate.

For buyers evaluating medical document processing or KYC document verification, UiPath now offers pre-built vertical agents rather than configurable extraction templates. That distinction between agents and templates is the competitive frame that matters in 2026.

Who wins: UiPath; banks and healthcare systems seeking pre-built compliant AI agents; Genzeon (enterprise distribution via UiPath's platform).

Who loses: Independent compliance AI vendors; horizontal IDP platforms without vertical agent libraries; vendors competing for healthcare revenue cycle budgets without CMS-credentialed partners.

Vertical Funding Validates Domain-Specific AI Economics

Three funding events this period confirm that domain-specific document AI commands growth-equity valuations that horizontal platforms cannot match.

Reducto AI, which closed a $75 million Series B led by Andreessen Horowitz in October 2025 (bringing total funding to $108 million), made its enterprise distribution play in February 2026: an AWS Marketplace listing enabling committed-spend purchasing and Enterprise Discount Programs, positioning Reducto directly against Amazon Textract inside AWS's own procurement channel. The platform has processed 1 billion+ pages, with named customers including Harvey, Mercor, Rogo, Scale AI, an unnamed Fortune 10 company, and a Global Top 5 Hedge Fund. For buyers evaluating OCR technology or AI data extraction within AWS environments, Reducto's committed-spend eligibility changes the procurement calculus.

mea Platform closed €42.2 million ($50 million) in minority growth equity from Scottish Equity Partners, its first external capital after four consecutive profitable years since founding in 2021. The platform is live in 21 countries, processing $400 billion+ in gross written premium, with named customers including AXIS Capital, CNA Financial, The Hartford, Markel, SCOR, and Lloyd's of London. The architecture, a domain-specific language model plus insurance knowledge graph, is the template for what vertical AI looks like at production scale. "Our opportunity to improve client combined ratios and margin is built on years of developing and deploying insurance-specific AI at global scale," said Martin Henley, Founder and CEO (Pulse2). Horizontal IDP vendors competing for insurance claims processing wallet share face a platform that has been profitable for four years without external capital.

Alkymi launched Alkymi Private Credit on February 4, extracting structured data from Loan Agent Notices, Compliance Certificates, and Financial Statements with real-time covenant breach detection, then announced a strategic partnership with FINBOURNE Technology on February 24 for integrated credit risk monitoring. The platform serves investment managers representing $20 trillion+ AUA, including Northwestern Mutual, PGGM, Strategic Investment Group, HESTA, and SimCorp. Morgan Stanley projects the private credit market grows from $3 trillion to $5 trillion by 2029. "Private credit is scaling faster than the infrastructure that supports it," said Harald Collet, CEO of Alkymi (PR Newswire). The covenant-specific data validation logic and FINBOURNE data integration represent a moat that horizontal IDP vendors cannot quickly replicate.

Who wins: Vertical AI platforms with proprietary domain data and workflow integration; growth-equity investors in profitable domain-specific AI; enterprise buyers in private credit, insurance, and financial services.

Who loses: Horizontal IDP vendors (ABBYY, Hyperscience) competing in financial services without domain-specific logic; general-purpose LLM wrapper vendors without workflow depth; Amazon Textract facing Reducto's committed-spend positioning inside its own marketplace.

Platform Giants Embed IDP Into Existing Enterprise Relationships

While vertical specialists raise capital and acquire, platform incumbents are embedding document intelligence into workflows where enterprise documents already live, eliminating the integration project that standalone IDP vendors require.

Box reached general availability for Box Extract on January 15, 2026, for Enterprise Advanced plan customers. Powered by customer-selectable models from Google Gemini, Anthropic Claude, and OpenAI, Box Extract offers two agent tiers, Standard and Enhanced, with custom agents that auto-trigger on document arrival per folder. Production deployments include RWS Global (contract processing reduced from 20 minutes to under 2 minutes; a 200-hire batch reduced from 8.5 workdays to 5 hours) and Texas DMV (forms and public records). "With Box Extract, that information is now unlocked and can transform how businesses analyze information and make decisions," said Aaron Levie, Co-founder and CEO of Box (BusinessWire). Analyst Alan Pelz-Sharpe of Deep Analysis called it "a good first step" and positioned it against Tungsten Automation, ABBYY, and Hyland. It marks the first time a content management vendor has been explicitly benchmarked against the IDP category's established leaders.

Hyland delivered the most comprehensive IDP product push of the period: a February 24 platform update across six products (Hyland IDP, Hyland Automate, Content Federation Service, OnBase, Alfresco, Nuxeo), self-reporting a 220% quarter-over-quarter surge in agentic content services adoption in Q4 2025 (no baseline or third-party verification disclosed). The Content Federation Service connects OnBase, Alfresco, Nuxeo, and SharePoint 365 without data migration, reframing legacy repositories as federated nodes in a multi-platform AI content graph rather than migration targets. IDC named Hyland a Leader in its Worldwide IDP Software 2025–2026 MarketScape. "Hyland's Content Innovation Cloud and its federation layer represent a major leap forward in intelligent document processing," said Amy Machado, Senior Research Manager at IDC (Hyland Newsroom). The healthcare vertical bet, Intelligent MedRecords and Intelligent Correspondence for Revenue Cycle debuted at HIMSS 2026, putting Hyland in direct competition with UiPath's healthcare AI launch in the same month.

SAP Document AI reached GA on four capabilities in Q4 2025: multimodal vision extraction (text plus images including hazard pictograms, stamps, and signatures), email attachment processing, multi-step document workflows, and SAP Cloud Transport Management integration for schema deployment across dev/QA/prod environments. The workflow engine GA is the strategic inflection: SAP Document AI is no longer a point extraction tool but a document orchestration layer embedded across 32 SAP business processes, serving 30,000+ customers. The SuccessFactors Onboarding embedded edition claims 15% onboarding acceleration and 30% validation accuracy improvement. For any enterprise running S/4HANA, Concur, Fieldglass, or SuccessFactors, the integration cost of a competing IDP vendor just increased materially.

Microsoft was ranked second Overall Leader in ISG's February 2026 IDP Buyers Guide (behind Appian, ahead of ServiceNow), with Exemplary ratings across three evaluation categories. Azure AI Foundry added mistral-document-ai-2512, a compound model pairing Mistral's OCR engine with mistral-small-2506, claiming 95.9% OCR accuracy on scanned documents versus 89–91% for unnamed competitors (vendor-reported; no independent validation published). The ARGUS open-source pipeline now supports runtime switching between Azure Document Intelligence and Mistral Document AI 2512 via three environment variables. "Integrated, AI-enabled software platforms are becoming foundational to enterprise operations," said David Menninger, Executive Director of Software Research at ISG (BusinessWire).

Who wins: Box (Enterprise Advanced upsell); Hyland (IDC Leader, healthcare vertical); SAP (distribution moat across 30,000+ customers); Microsoft (ISG ranking closes perception gap); enterprises already on these platforms who gain IDP without integration projects.

Who loses: Pure-play IDP vendors (ABBYY, Tungsten Automation, Hyperscience) competing for accounts where Box, Hyland, SAP, or Microsoft are already deployed; standalone document extraction point solutions without platform distribution.

Open-Source Infrastructure Raises the Floor for Every IDP Vendor

IBM's Docling is no longer a parsing utility. It is enterprise AI pipeline infrastructure, and its February 2026 releases set a new baseline that every IDP vendor must now compete against.

IBM released Granite-Docling-258M on Hugging Face under Apache 2.0, replacing SmolDocling-256M-preview. The model uses a Granite 3 language backbone plus SigLIP2 visual encoder, capturing charts, tables, forms, code, equations, footnotes, and captions in a single-pass DocTags format with output in Markdown, JSON, and HTML. Model string: ibm-granite/granite-docling-258M. Docling was simultaneously donated to the Linux Foundation under the Agentic AI Foundation alongside BeeAI and Data Prep Kit. A Docling OpenShift Operator launched with Red Hat targets banks specifically. NVIDIA RTX GPU acceleration delivers up to 6x speedup over CPU-only PDF processing, with hardware tiers ranging from RTX 5090 (batch 64–128) to RTX 5070 (batch 16–32). The Anyscale Ray Data plus Docling architecture scales across 10–100 Kubernetes nodes via KubeRay. The project has 37,000+ GitHub stars and an 8-repository ecosystem including docling-mcp for agentic workflows.

"It's not just conversion anymore. We're thinking through it. We're generating and manipulating documents," said Peter Staar, Principal Research Staff Member at IBM Research Zurich and Chair of Technical Steering for Docling at the Linux Foundation (IBM Think).

The Apache 2.0 license, Linux Foundation governance, OpenShift operator, and MCP tooling collectively lower adoption barriers while locking in IBM's distribution channels. For buyers evaluating document processing for RAG or self-hosted document processing, Docling provides a free, GPU-accelerated, enterprise-governed baseline. No head-to-head benchmarks against Mistral OCR, GPT-4o Vision, or Google Document AI have been published. That gap is the next credibility test.

AYR released a materially upgraded Intelligent Document Simulator on February 14, adding structural layout manipulation: users input one sample document and generate variants by swapping table columns and shuffling sections horizontally and vertically, enabling model training without access to sensitive live data. Deep Analysis named AYR alongside Salesforce, Celonis, UiPath, and Xceptor in its Vendor Vignettes series, calling its multimodal fusion AI "one of the most efficient and flexible IDP systems we have seen." The synthetic document generation approach directly attacks the two biggest IDP deployment bottlenecks, data sensitivity and layout variability, potentially compressing go-live timelines from months to days for regulated-industry buyers evaluating IDP implementation.

Who wins: IBM/Red Hat (enterprise distribution via OpenShift); NVIDIA (RTX acceleration validates GPU stack for document AI); regulated industries gaining on-premises sovereign AI document processing; developers building RAG pipelines.

Who loses: Proprietary document parsing vendors without open-source equivalents; multi-stage ensemble pipeline vendors that Docling's single-pass DocTags architecture directly counters; IDP vendors without GPU acceleration paths.

Security, Sovereignty, and Compliance Reshape Procurement Criteria

Three developments this period confirm that security posture and data sovereignty are becoming first-order procurement criteria, not afterthoughts.

Novee Security disclosed 16 zero-day vulnerabilities across Apryse WebViewer SDK and Foxit PDF cloud services combined, discovered using a hybrid human-agent LLM system trained on manually identified vulnerability patterns. Of the 16, three CVEs were assigned specifically to Apryse: CVE-2025-70402 (Critical: DOM XSS via malicious uiConfig JSON, enables full account takeover in one click), CVE-2025-70401 (High: stored DOM XSS in annotation author fields, triggers on single keystroke), and CVE-2025-70400 (High: SSRF via iFrame rendering, enables internal network reconnaissance). All three were patched before public disclosure via coordinated disclosure. "The issues referenced in Novee's upcoming research were responsibly reported and have been addressed through product updates, documentation improvements, and strengthened default configurations," said Stan Kornacki, VP of IT and CISO at Apryse (SecurityWeek). The attack method, LLM agents trained on vulnerability patterns and deployed autonomously against obfuscated code, signals that every IDP vendor with an embeddable document SDK faces an elevated and accelerating threat surface. For buyers embedding document SDKs in authenticated enterprise applications, patch verification cadence is now a vendor evaluation criterion. See the document processing security guide for evaluation frameworks.

Objective Corporation reported H1 FY2026 results that include a landmark efficiency proof point: 4 billion tokens processed at an aggregate cost of AUD 4,000. "We recently processed 4 billion tokens at an aggregate cost of AUD 4,000," said Tony Walls, CEO (Investing.com). Combined with IRAP certification (opening Australian federal pipeline) and explicit Canadian government interest in non-US software alternatives, Objective is positioned to capture a geopolitically-driven wave of public sector demand for sovereign, cost-efficient document AI. Revenue reached AUD 66.7 million (+8.8% YoY), ARR AUD 120 million (+12%), and the Rule of 40 score is approximately 52. These metrics validate the sovereign AI positioning as commercially durable, not just politically convenient. Buyers evaluating government document processing should treat Objective's IRAP certification as a procurement shortlist criterion.

Conduent reported FY2025 adjusted revenue of $3.04 billion (−4.2% YoY) with a new CEO less than 30 days in role at earnings. Then a data breach affecting 25 million+ people was reported by TechCrunch on February 24, with no management response on record. The Commercial segment ($1.5 billion, −5.9% YoY) is currently leaderless. "Ladies and gentlemen, this is a turnaround story," said Harsha Agadi, CEO (MarketBeat). The breach, combined with a leaderless commercial segment and declining revenue, creates compounding credibility risk that competitors can exploit in government and healthcare verticals, Conduent's highest-margin segments.

Who wins: Objective Corporation (sovereign AI positioning, IRAP certification); vendors with strong internal security review cadences; enterprises that patched Apryse promptly; IDP vendors with FedRAMP or IRAP credentials.

Who loses: Apryse (reputational risk despite responsible disclosure); Conduent (breach liability unquantified, commercial leadership vacuum); US-headquartered cloud ECM vendors facing data sovereignty headwinds in Commonwealth markets; IDP vendors with large SDK surfaces and infrequent security audits.

MCP Becomes the Integration Standard for Document AI Agents

The Model Context Protocol is emerging as the connective tissue between document intelligence platforms and AI agent runtimes. The vendors moving fastest to implement it are building distribution moats that slower adopters will struggle to close.

DocuSign launched its IAM integration inside Anthropic's Cowork via MCP connector on February 24, enabling five concrete workflows: contract drafting from MSA templates, 90-day expiry surfacing by clause type, AI redline review plus vendor routing, summary reports by clause, and customer onboarding with identity verification. "What Docusign brings to agentic experiences like Cowork is deep context across all business agreements," said Allan Thygesen, CEO of DocuSign (PRNewswire). The integration repositions contract data as first-class infrastructure for AI agents, and any MCP-compatible runtime can now execute agreement workflows. The beta is English-only, limiting multinational procurement use cases near-term.

iManage reached GA on January 29 with Ask iManage expanded to platform-wide natural-language search across entire iManage Work repositories, with Model Context Protocol exposing its governed layer to third-party tools. The platform is powered by Microsoft Azure OpenAI Services inside the customer's own tenant, with US data center processing for US customers and permission boundaries enforced at the AI layer. iManage's concurrent Knowledge Work 2026 Benchmark Report (3,185 decision-makers, 26 countries) found that 85% of firms are piloting or using AI, but only 17% have fully embedded it in daily operations, while 72% are planning new document or knowledge management platform investment within two years (iManage Benchmark Report). That 72% figure is the addressable market signal for every IDP and DMS vendor in this report.

Coveo launched a hosted MCP Server on February 10, connecting AI agents to Coveo's unified content index without custom integrations, with 10 customers already in production and pricing included within existing subscriptions. "We are already working closely with 10 customers leveraging our hosted MCP server to enhance leading models such as Anthropic's Claude and OpenAI's ChatGPT," said Laurent Simoneau, Co-Founder and CEO (PRNewswire). Coveo's Q3 FY2026 results showed record new business bookings, with generative AI solutions exceeding 25% of new bookings, validating the pivot from enterprise search destination to foundational AI retrieval infrastructure.

IBM's docling-mcp repository directly competes with Docugami's MCP Server in the same period, and MuleSoft reached GA on Agent Fabric, automatically discovering, cataloging, and governing AI agents across Agentforce, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Copilot Studio. The MCP protocol is no longer an emerging standard; it is the integration layer that enterprise document AI platforms must support to remain relevant in agentic workflows. See the agentic document processing capability page and the agentic document processing guide for implementation frameworks.

Who wins: DocuSign (distribution via Anthropic's ecosystem); iManage (72% platform investment signal converts to pipeline); Coveo (record bookings, zero-incremental MCP pricing); IBM (docling-mcp positions Docling as agentic infrastructure); MuleSoft/Salesforce (agent control plane positioning).

Who loses: CLM vendors without MCP integrations (Icertis faces a DocuSign-Anthropic combined workflow); standalone enterprise search vendors without agentic integration strategies; IDP vendors that have not published MCP tooling.

The Verdict

The February 2026 IDP market delivered a verdict that buyers and vendors must act on immediately: by Q4 2026, no horizontal IDP vendor below $50M ARR without a vertical workflow anchor or platform partnership will survive as an independent product. They will be acquired as features or exit the market. Eight acquisitions since 2023, culminating in UiPath's absorption of WorkFusion in February 2026, plus three vertical funding rounds above $42 million, and platform giants embedding IDP into existing enterprise relationships collectively confirm that document intelligence is becoming a feature of vertical software stacks, not a product category in its own right.

For buyers, the implication is concrete: evaluate IDP vendors not on extraction accuracy alone, but on vertical workflow depth, agentic orchestration capability, MCP integration status, and security posture. The 72% of firms planning new document or knowledge management platform investment within two years (iManage Benchmark Report) will find a market where the best standalone IDP vendors have already been acquired, and the survivors are either vertical specialists with defensible moats or platform giants with distribution advantages. Use the IDP vendor evaluation framework to stress-test shortlists against these criteria before mid-2026.

For vendors, the consolidation window is closing. Sub-scale horizontal platforms that have not established vertical depth or platform partnerships by mid-2026 face the same exit path as AntWorks, Metamaze, Insiders Technologies, and Infinia ML, absorbed as features, not acquired as platforms. The premium valuation goes to domain-specific AI with production-grade deployment, as mea Platform's €42.2 million growth-equity round at four years of profitability demonstrates.

The decisive claim sharpens: agentic AI systems that make decisions, orchestrate workflows, and handle exceptions autonomously represent the new competitive standard. As Daniel Dines, CEO of UiPath, framed the WorkFusion acquisition: "Financial institutions need intelligent solutions to combat sophisticated financial crimes and navigate evolving compliance requirements." The same logic applies to every regulated, document-heavy industry. Platforms that deliver autonomous compliance, not just accurate extraction, will define the next competitive tier.