Following a research note I published this week on the GA release of Box Extract, I wanted to examine how other vendors are enabling organizations to optimize enterprise content management and unlock value using agentic AI. The major bottleneck isn’t just technical — it’s organizational and governance-based. Enterprises possess vast repositories of unstructured content, but lack both the tools to extract actionable meaning from it and the operational frameworks to govern and route that extracted intelligence reliably through their AI systems. This is why vendors like Box are treating content management as foundational infrastructure, not just storage, for AI-driven operations. (Note that Box and some of its competitors are advisory clients of my firm, Moor Insights & Strategy.)
This shift is important because failing to adequately address unstructured data creates a compounding problem for enterprises trying to implement AI effectively. When unstructured data remains inaccessible to AI agents, organizations cannot operationalize their largest pool of information assets. The resulting friction cascades: teams continue manual processing, AI agents operate on incomplete context and organizations fail to capture the competitive advantage that AI promises.
The question isn’t whether enterprises have enough unstructured data to make use of (they do, in overwhelming volumes) or whether that data should be harnessed for AI (there is broad consensus that it can be highly valuable). The question is whether they can make it AI-ready in a way that integrates with their governance and operational workflows.
The Core Problem: Harnessing Unstructured Content For Agentic AI
Enterprises possess vast amounts of unstructured data, but this material — from documents, e-mails, service logs, call transcripts, and so on — remains largely inaccessible to AI agents. The core challenge, and the central argument of this analysis, is that making this unstructured data usable is critical for effective enterprise AI. Komprise’s 2026 State of Unstructured Data Management survey shows that 74% of IT and data storage leaders now manage at least 5 petabytes of unstructured content — a 57% increase over 2024.
Traditional optical character recognition tools digitize unstructured text effectively, but they fail to extract meaning. So, for instance, OCR might make a contract searchable, but an AI agent still cannot discern whether a specific date in the middle of a paragraph represents a signing date, an expiration date or a renewal date — distinctions that matter enormously in legal and financial operations.
To take the example I explained in more detail in my research note, Box Extract addresses this challenge directly using AI. Rather than simply converting images to text, it applies agentic reasoning — powered by models including Google’s Gemini, Anthropic’s Claude and OpenAI’s GPT — to understand context, semantic relationships and document hierarchy. Thus the system can recognize when “trust” in a document indicates a specific legal entity rather than a bit of marketing language. Building on OCR technology, it can also interpret handwritten text alongside printed content, and extract meaning from documents whose layouts have shifted over time. The extracted metadata turns into structured, governed content that agents can reliably act upon. This is precisely what enterprises need: a way to transform their vast repositories of unstructured documents into AI-ready information assets.
Naturally, Box’s customers are not the only ones facing this problem. Across the industry, content intelligence vendors are responding to the same customer reality: enterprise AI initiatives stall when organizations cannot structure their existing content at scale. A 2025 Fivetran survey found that 42% of enterprises report that more than half of their AI projects have been delayed, underperformed or failed due to data readiness issues.
There is reason for hope, however, and Box CEO Aaron Levie explained in a LinkedIn post why this moment is different: “AI is getting insanely good at structuring unstructured data … We could always do this with our structured data — the information that goes into a database, ERP system or CRM system — but it was never possible to pull this off for all the unstructured data, or content, that makes up the vast majority of the data estate of any enterprise. You either had to use [robotic process automation] systems, which couldn’t adapt easily to new data types, use custom-trained [machine learning] models that only worked for a subset of data, or had to spend an inordinate amount of time manually reading each document and pulling out the relevant information.”
In response, more enterprises are asking fundamental questions: How do we prepare our document repositories for agents? How do we scale extraction across billions of documents? How do we ensure that agents have access to accurate, governed information? The vendors best positioned to answer these questions will define how enterprises move from AI pilots to production-scale operations.
What Are Vendors Doing? Four Approaches To The Extraction Challenge
The competitive response to this challenge has split into several approaches, each reflecting different assumptions about what matters most in enterprise environments. These distinctions shape the vendor landscape discussed below.
Box’s integrated platform strategy positions content extraction as one layer within a comprehensive intelligent content infrastructure designed for AI operations. Box Extract is not a standalone tool; it sits alongside Box Shield Pro, which applies AI agents to security and compliance operations, and integrates deeply with Box Relay automation. Extracted metadata flows into Box-native workflows, syncs to external data warehouses and integrates with downstream systems via APIs or connectors to Slack, Salesforce, ServiceNow, Workday and others. Box’s reasoning is clear: extraction matters only if the content it governs reaches the workflows and systems where agents operate. The value proposition combines extraction precision, platform integration and operational scale. For organizations already using Box for content management, the tool is native, reducing data movement overhead and accelerating time-to-value.
Hyland positions itself as an enterprise platform that combines content services, intelligent document processing and workflow orchestration, with deep domain expertise. In June 2025, it announced agentic document processing that requires no training data. Its Content Innovation Cloud lets enterprises connect multiple content sources and apply consistent extraction and classification across hybrid environments. Hyland’s main strengths are breadth and domain depth. Its platform offers document capture, classification, extraction and low-code workflow design, and it addresses domain-specific challenges in regulated sectors. However, large organizations may require integration support due to the complexity of implementation.
Microsoft and Salesforce have each embedded extraction directly into their dominant ecosystems. Microsoft’s Azure Document Intelligence and Salesforce’s Document AI operate on cloud platforms that many enterprises already use for identity, data and business processes. Microsoft deeply integrates extraction with Copilot and Microsoft 365, while Salesforce focuses on field mapping and workflow automation in its Sales Cloud and Service Cloud. Both leverage proprietary model access and deep ecosystem integration. Neither position extraction as a standalone product; instead, both treat it as a feature accelerating existing use cases. Google Cloud Document Intelligence takes a different approach, offering pay-per-page pricing and handwriting recognition across 50 languages, making it attractive for organizations processing scanned documents at scale. An ecosystem-centric model like this works best for companies that want document intelligence tightly coupled to a specific cloud or CRM platform. Meanwhile, organizations with more specialized or cross-platform needs may look to content-centric platforms such as Box or Hyland, or to neutral IDP vendors.
OpenText and Workiva both take a horizontal data platform approach. OpenText File Content Extraction supports 2,300-plus file formats, positioning extraction as infrastructure for data pipelines, analytics and search. Workiva, focused on financial reporting and compliance, extracts terms and clauses from documents and grounds analysis in company-specific context. Both providers see extraction as foundational for analytics and governance, not just departmental workflow automation.
The Decision Tree: How Organizations Choose
Vendor selection in this area should reflect how enterprises value extraction. Salesforce users will find Document AI convenient, but Box also offers deep Salesforce integration for routing extracted content. Financial firms with variable documents may value Hyland’s specialization. Tech companies embedding extraction may opt for OpenText or niche players offering APIs.
Governance is another vital consideration, and the market agrees: extraction without adequate governance at agent speed is risky. As agents can now query millions of documents instantly, errors are no longer isolated — they propagate rapidly, risking compliance and business processes. Box’s integration of extraction with Shield Pro and Hyland’s focus on policy-driven classification underscore the importance of governance. Vendors that see extraction as a narrow technical task, without governance and auditability, will likely fall behind.
What’s Shaping The Market In 2026: From Pilots To Production Scale
The shift from 2025 to 2026 marks a transition from experimentation to operational maturity. Enterprise adoption of agentic AI is accelerating rapidly, moving from early pilots toward broader production deployments. This acceleration is not primarily driven by model improvements; rather, it is driven by organizations finally solving the content-readiness problem through platforms and tools like those that Box, Hyland and others are bringing to market.
As vendors compete to capture this opportunity, four dynamics will reshape how extraction capabilities are built and deployed:
- Multi-agent orchestration will require specialized extraction agents. Organizations are moving away from monolithic AI systems toward coordinated teams of specialized agents. This means extraction will not be a one-time data preparation problem but a continuous, ongoing process in which different agents apply distinct extraction logic to different document classes. Vendors that can support domain-specific agents for contracts, claims and compliance filings will outperform generalist platforms.
- Governance and auditability will become competitive assets. As extraction-driven agents make business decisions at scale, organizations will demand traceable, defensible extractions. Which model was used? What was the confidence score? Which data was masked for privacy? What prompted the agent to take a particular action? Vendors offering transparent, auditable extraction pipelines will win over those treating extraction as a black box.
- Synthetic parsing pipelines will emerge as the architectural norm. Rather than routing entire documents to a single model, documents will be decomposed into constituent parts such as titles, paragraphs, tables and images and be routed to the model or agent best suited to understand it. This requires extraction vendors to offer flexible routing logic, not just extraction capability. Some incumbents have distinct advantages in this regard: Box’s API architecture and integration model position it well for this evolution; Hyland’s low-code workflow studio enables the needed flexibility; and the platform nature of Salesforce and Microsoft also enables this. By contrast, specialized extraction vendors will need to evolve or risk commoditization.
- The gap between feature release and operational scale will become the critical differentiator. To stick with Box as an example, the general availability of Box Extract is important, but the real measure of success will be how many Box customers have extraction-powered agents running reliably in production by the end of 2026. Vendors that invest in implementation frameworks, reference architectures, prescriptive guidance and industry-specific templates will accelerate customer time-to-value and scale. By contrast, vendors that simply release features and assume that customers will figure out the rest will increasingly lose to competitors offering comprehensive operational support and success frameworks.
The Governance Bottleneck Meets The Autonomy Imperative
The solutions mentioned here address one of the most critical challenges facing enterprises implementing AI at scale: making unstructured content accessible and useful to intelligent agents. This is not a marginal problem; it is central to whether enterprises can move from AI pilots to production deployments. In my view, Box is leading this movement by integrating extraction, governance, security and workflow automation into a cohesive platform designed for AI-powered operations. That said, the company faces sophisticated competition from established players like Hyland and technology giants such as Microsoft and Salesforce, each bringing different strengths and perspectives to address the same fundamental problem.
As touched on above, the real test in 2026 will not be which vendor has the most sophisticated extraction model. It will be whether enterprises can operationalize content governance at the speed their agents demand. Research from Deloitte and others reveals that 65% of organizations are struggling to achieve AI success, not because the technology is immature, but because organizations deployed agents faster than they could control them.
The vendors that win will be those helping customers move beyond governance-as-policy to governance-as-architecture, where controls execute at machine speed and auditability is built into the system itself rather than bolted on afterward. The market will ultimately reward those that execute this transition most effectively at scale.
Moor Insights & Strategy provides or has provided paid services to technology companies, like all tech industry research and analyst firms. These services include research, analysis, advising, consulting, benchmarking, acquisition matchmaking and video and speaking sponsorships. Of the companies mentioned in this article, Moor Insights & Strategy currently has (or has had) a paid business relationship with Box, Google, Microsoft, Salesforce and ServiceNow.
