Document Mining And Analytics Platforms

Document Mining And Analytics Platforms is the fourth name this Forrester evaluation has carried in eight years. AI-Based Text Analytics Platforms in 2018. A document-focused variant of it in 2020. Document-Oriented Text Analytics Platforms in 2022. The current title arrived in 2024, and the Q2 2026 edition is published. Each rename recorded what buyers were paying for that year, and the newest edition adds something none of the earlier ones carried: the price.

The Forrester Wave: Document Mining And Analytics Platforms, Q2 2026 puts the unit economics in writing. Five cents per page at high volume, twenty cents at the low end. A million pages a year at a nickel is fifty thousand dollars in platform fees. That is real money, and it is not the number that decides the budget. The exception queue does. Forrester states that accuracy often starts around sixty percent and improves into the high nineties with tuning, and that human-in-the-loop review remains essential for most production deployments. Hold those two statements together and the category reveals itself: the product being sold is not reading. It is making the reading verifiable and the exceptions cheaper.

The new edition scores eight providers against thirty one criteria, the same field size as the 2018 debut and the largest scorecard in the category's history. Boris Evelson, who has led this evaluation across its renames, is the author. The findings post went up the same day, May 21, 2026.

Eight vendors, thirty one criteria, one published price: The Forrester Wave: Document Mining And Analytics Platforms, Q2 2026

The publicly confirmed placements so far are two Leaders and one Strong Performer.

UiPath is a Leader, on the strength of its Intelligent Xtraction and Processing platform, IXP. The report's own phrases run through it: high marks for agentic AI operations and architecture, agentic AI functionality, and agentic AI integration. Governed agent execution. User-reviewable plans. OpenTelemetry-compliant tracing and auditability. Scores above par for AI model orchestration, hybrid pipelines that route work on confidence thresholds with human-in-the-loop validation, plus composite-document splitting and validation feedback loops. The fit statement is plain: a good choice for organizations that want to standardize on a single agentic automation platform, combining document processing with orchestration, governance, and enterprise integrations. The report also notes UiPath's February 2026 acquisition of WorkFusion, which adds specialized AI agents for AML, KYC, and fraud investigation in banking.

Hyperscience is a Leader too, and the report's Customer Favorite, marked with a double halo for outstanding customer feedback. It took the highest Current Offering score in the field and top marks in eighteen of the thirty one criteria, spanning the full agentic stack: agentic AI functionality and integration, generative AI architecture, AI model lifecycle management, vectorization and hybrid RAG, generative AI guardrails, plus vision, innovation, and roadmap. Forrester calls it the only IDP pure play in the evaluation and frames it for organizations that need a dedicated IDP platform, with a code-first platform built for embeddability. Its reference roster reads like a compliance department's org chart: American Express, Charles Schwab, the U.S. Social Security Administration.

Iron Mountain lands as a Strong Performer on InSight DXP, with the highest possible scores in agentic AI functionality, document labeling and annotation, and data privacy. A records-management company scoring at the top of a document mining evaluation is not an accident. It is the category's history repeating itself. The firms that own the documents have kept pace with the firms that built the algorithms.

The remaining placements sit behind the paywall. Forrester's own findings post describes the market in segments instead: enterprise content management names like Hyland, Iron Mountain, and OpenText; intelligent automation names like Automation Anywhere, EdgeVerve, Rossum, and UiPath; plus search and knowledge platforms and niche specialists. The advice attached to the segmentation: start with the use case, then shortlist within the right segment. Long-form documents and high-volume transactional documents demand different platforms, and no single vendor dominates both.

The word that survived all four names

Analytics is the only word present in every one of this evaluation's four titles. Text died in 2024. Mining entered in 2024. The renames track what the market stopped arguing about.

The 2018 edition, The Forrester Wave: AI-Based Text Analytics Platforms, Q2 2018, evaluated eight vendors on twenty two criteria. Leaders: IBM, SAS, Clarabridge, and Micro Focus. Strong Performers: Expert System, OpenText, Attivio, and EPAM. Its signature finding was that rules still ruled. Rules-based platforms were more accurate out of the box than machine learning-heavy rivals, with less training. The 2020 document-focused edition separated documents from general text analytics. Leaders: Micro Focus, IBM, and OpenText. Strong Performers: Expert System, Google, EPAM, and AntWorks.

By 2022 the report was The Forrester Wave: Document-Oriented Text Analytics Platforms, Q2 2022, twelve providers on twenty six criteria: AWS, AntWorks, EdgeVerve, Expert.ai, Google, Hyperscience, IBM, Micro Focus, Microsoft, OpenText, SAS, and WorkFusion. Micro Focus IDOL led it.

Two facts about that 2022 roster deserve attention. Four hyperscalers sat in the field. None of them appear among the publicly known 2026 placements. And Hyperscience, the 2026 Customer Favorite, was in the evaluation two editions before its current crown. The specialists outlasted the generalists.

The 2024 edition and its six Leaders

The first edition under the current name arrived May 30, 2024: The Forrester Wave: Document Mining And Analytics Platforms, Q2 2024. Fourteen providers, twenty five criteria, drawn from a Landscape of forty one vendors published that quarter. Six were named Leaders. Four of the placements are public.

UiPath earned its first Leader slot with strengths in human-in-the-loop accuracy verification, ML-based AI, generative AI use in document mining, deployment options, and workflow orchestration. Forrester's fit line: "an excellent choice for clients seeking an integrated solution that combines document mining, analytics, and process automation."

OpenText, evaluated on IDOL, took the highest Current Offering score, with top marks in accuracy verification, complex forms and images, security and regulatory compliance, and document-level text mining. Forrester's verdict: equipped to handle any and all document mining and analytics tasks.

expert.ai was cited for hybrid AI that clearly stands out from the crowd, with the highest scores in knowledge-based and symbolic AI, generative AI post-processing, and document labeling for ML training, plus vision, roadmap, and adoption in strategy. Positioned for regulated industries that need explainability: insurance, pharma, financial services.

Hyperscience took the highest possible scores in ten criteria, including accuracy verification, ML-based AI, complex forms and images, deployment options, DevOps, adoption, partner ecosystem, roadmap, and vision. Forrester called it a great choice for customers looking for prebuilt document mining and analytics solutions based on human readable code.

Rossum was a Strong Performer, with the highest possible scores in innovation and generative AI post-processing. Forrester singled out its approach of using a specialized fine-tuned model to validate the results and confidence scores of the foundational LLM, which it said clearly stands out among the competition. Reference customers praised overall business value, the relationship, and the user experience.

In 2024 the scorecard's fresh language was generative AI post-processing, scored for a handful of vendors. In 2026 the Wave devotes its largest block to agentic AI architecture, functionality, integration, guardrails, and model lifecycle. The change in the scorecard is the change in the product.

Why eight is the new fourteen

The field shrank from fourteen to eight while the scorecard grew from twenty five to thirty one criteria. Shrinkage plus a longer test is how a definition gets sharper.

The exclusion clause is the finding. Forrester's blog says legacy OCR pioneers were left out because they either do not focus on this specific enterprise segment or lacked the technical mass required to rank among the top eight. That sentence is the industry's history in miniature. Optical character recognition was the product for thirty years. In 2026 it is an assumed component, and a vendor whose stack stops at recognition does not make the field.

The criteria added between editions point the same way: agentic AI functionality and integration, generative AI architecture, AI model lifecycle management, vectorization and hybrid RAG, generative AI guardrails. The weight of the scorecard moved from how accurately the platform extracts toward how well it acts and whether you can prove what it did.

Five cents a page changes the conversation

Once an analyst publishes the price, procurement conversations change shape. The vendor's discount ladder now has a public anchor. The business case no longer starts from a vendor's spreadsheet. It starts from Forrester's: five cents at millions of pages a year, twenty cents at low volume, with the mix of document types deciding where a deal lands.

The sixty percent starting accuracy is the most honest number in the report. It means the first invoice batch will not be hands-free. The platform's job in month one is to make the exception queue short, sortable, and safe to hand to a person. Accuracy in the high nineties is an endpoint, reached with tuning, on your document mix, not the vendor's demo set. The same post says to expect roughly six months to MVP, not weeks as some vendors claim. The report says so in its own numbers, and the numbers are written down where the vendor cannot rewrite them.

What the new scorecard scores instead of reading

Reading is table stakes. The 2026 criteria that separate vendors describe what happens after extraction, and whether anyone can prove it.

UiPath's user-reviewable plans mean the agent proposes before it acts. OpenTelemetry-compliant tracing means the audit trail exports to the same observability stack the rest of the enterprise uses. Model orchestration means hybrid pipelines route each job to the right model on confidence thresholds, with a human at the boundary. And the routing question Forrester raises is pointed: not every task needs the most expensive LLM. Premium models for cursive handwriting and images, lighter models for summarization, custom machine learning for complex tables. A platform that runs everything through one frontier model is not architecture. It is a bill.

The European angle follows directly. Data privacy was a maximum-score criterion for both Hyperscience and Iron Mountain. UiPath's reference customers explicitly cite document privacy and sovereignty. For regulated and air-gapped deployments, the criterion list doubles as a compliance checklist: model lifecycle management, guardrails, deployment options, privacy. The scorecard is scoring, in part, the paperwork you will inherit in an audit.

The honest gap in the agentic story

Here is the tension the report itself contains. It scores agentic AI functionality at maximum for its Leader, and it states that human-in-the-loop review remains essential for most production deployments. If the agents were autonomous, the humans would be gone. They are not, and the report says so in its own numbers.

The gap between those two statements is where buyer risk lives. Agentic functionality is scored on what the platform can do in a reference architecture. Human-in-the-loop essentialism is what production has required so far. A vendor's five out of five on agentic AI functionality does not tell you what share of your invoices will route end to end untouched. Only your document mix does, and only after tuning.

The two publicly confirmed Leaders demonstrate the point from opposite sides. UiPath leads with the platform: orchestration, governance, integrations, a full automation estate with documents inside it. Hyperscience leads with the documents: a code-first IDP engine built to be embedded. Same tier, opposite gravity. Nothing in the graphic answers which one fits a bank processing millions of mortgage documents a year versus a mid-market CFO automating accounts payable. The graphic ranks. The use case decides.

Three questions before you price the pilot

What is the price per page at your volume, and what does the exception queue cost on top? The nickel is for pages. The labor that reviews the misses, and the tooling that sorts them, is a separate line. Ask the vendor to price both, in the same room.

Does the agent show its plan before it acts, and can you export the trace? User-reviewable plans and OpenTelemetry tracing are scored criteria now. If the vendor demos autonomy without either, you are buying the 2024 product.

Which model runs which job, and what does each call cost? Ask for the routing table: premium models for handwriting and images, light models for summaries, custom machine learning for tables. A vendor that cannot answer is running your documents through one expensive model and will bill you for the privilege.

The category has finished changing names, or so the 2026 edition argues. The word analytics survived all four titles. Reading did not survive as the product. What is being sold now is a governed pipeline from document to decision, priced per page, with humans still in the loop and the audit trail exportable. The Wave ranks the vendors. The spreadsheet prices them. Buy with both open.

That governed pipeline sits downstream of the same discipline covered in Data Management For Analytics Platforms, where the equivalent ingestion-cleansing-governance work is being automated by generative AI on the structured side of the house.

Forrester does not endorse any vendor named here, and tier placement should not be read as a recommendation to buy.