.webp)
Docsumo
Docsumo is a Document AI platform whose seven chainable agents collect, classify, extract, verify and decide on business paperwork. It claims 99% field-level accuracy across 250+ document types and 95%+ straight-through processing for operations teams.
What is Docsumo?
Docsumo is an enterprise Document AI platform published by Docsumo Pte. Ltd., a private Singapore company registered under UEN 202003040H at 160 Robinson Road, #20-03, Singapore 068914. Its 2026 positioning is intelligent document processing meeting agentic AI: a stated move from rule-based OCR to LLM-led document processing, packaged as seven specialist agents in three families. Read covers the Email Agent and the Document Collection Agent; Decide covers Classification, Extraction and Verification; Act covers Case Management and Workflow. Each agent can be switched on alone, then chained into a workflow, and every output is meant to stay traceable to its source.
The claimed numbers are precise: 99% field-level accuracy, more than 250 document types, 95%+ straight-through processing and a median of eight seconds end to end, drawing on 100+ pre-trained models built from twenty million documents. Document types cited include invoices, bank statements, ACORD insurance forms, W-2 and W-9 tax forms, 1003 mortgage applications, cheques, bills of lading, utility bills, payslips and identity papers. Accepted formats are PDF, JPG, JPEG, PNG, TIFF, TIF, XLSX, XLS and email attachments, covering scans, native PDFs and handwriting, with complex table extraction: nested tables, tables spanning pages, several tables in one file.
In practice the platform reads a shared mailbox or a folder, classifies and splits packets, extracts fields with a confidence score and a clickable source span, cross-checks values between documents using rules written in plain English, groups everything into a case, returns one decision and publishes it to NetSuite, SAP, Encompass, Epic or Guidewire. Only real exceptions reach a human reviewer. Connectivity is a REST API, webhooks and ready-made connectors, with 30+ APIs mentioned on the About page.
The company started as consulting work on data-capture APIs in 2018, launched the product in 2019 and joined Techstars London in January 2020. It claims 10,000+ customer teams, 20 million documents processed, $100 million in processing savings and 3.4 million hours saved, naming customers such as Arbor Realty Trust, National Debt Relief, Cassena Care and Biagi Bros.
What it does
- Extract fields, tables and signatures across 250+ document types, each value carrying a confidence score and a clickable source span
- Classify every incoming document by type and split bundled packets automatically
- Verify values across several documents with rules written in plain English, escalating only genuine exceptions
- Collect documents from email, SFTP, cloud folders, scanners, vendor portals and the API, and triage a shared mailbox by intent
- Group related documents into a case and return a single decision for that case
- Push decisions into NetSuite, SAP, Encompass, Epic or Guidewire, with every step written to an audit log
- Train a custom model on about 20 sample documents, then export to Excel, JSON, REST API or webhooks
When to use Docsumo / When not to
A quick filter to help you decide if Docsumo is the right fit.
When to use Docsumo
- Operations and technology teams at mid-sized and enterprise companies buried under high volumes of unstructured documents
- Commercial lending, mortgage and banking teams turning bank statements, 1003 packets, financial spreading and KYC onboarding into structured data
- Insurance underwriters and claims handlers working through ACORD forms, certificates of insurance and loss files
- Accounts payable and accounting teams matching invoices to purchase orders and clearing exceptions at scale
- Healthcare, logistics and real estate back offices that must trace every extracted value back to its exact position in the source document for an auditor or a regulator
When not to use Docsumo
- Consumers and hobbyists: article 8 of the terms reserves accounts to individuals or entities acting in a professional capacity, and bot-created accounts are banned
- Anyone looking for a document archive: article 3.1 of the terms states the service includes no storage service, and content is kept 30 calendar days at most by default
- Teams that need a permanent free tier or a published price before contacting sales: only the 14-day trial is free, and Business and Enterprise show a contact button instead of a figure
- Organisations that cannot let a vendor train its models on their documents, since training is declared and no opt-out is documented
- Mobile-first or non-English-speaking teams: there is no mobile app and no browser extension, and the terms limit support to English
How to use Docsumo
A typical end-to-end flow, from setup to results.
- Create an account on app.docsumo.com/signup: self-serve, no credit card, one account per person since sharing is prohibited
- Start the 14-day free trial, whose page quota is announced as 1,000 pages on the pricing card and as 100 pages in the FAQ
- Upload your first documents into the platform, or plug straight into the REST API
- Pick a pre-trained model matching your document type from the 100+ available
- Train a custom model on about 20 sample documents when no pre-trained model fits
- Write validation rules in plain English, column by column or at case level
- Review the flagged fields: each value shows a confidence score and a link to its position on the page
- Build a workflow from a template (AP invoice intake, mortgage packet, insurance FNOL, KYC onboarding) or by describing it in natural language
- Export to Excel or JSON, or wire webhooks and the API into your ERP or CRM
- Roll out with the announced onboarding: a four-to-eight-week success plan, live training sessions and a dedicated automation expert, with support by email, in-app chat and a private Slack or Teams channel
Pros & Cons
Pros
- Traceability by design: every field carries a confidence score and a link to its exact position in the source page, and every workflow step is logged
- Broad, vertical-specific document coverage, from ACORD forms and 1003 mortgage applications to W-2s, W-9s, bills of lading and Medicaid paperwork
- Validation rules written in plain English and workflows generated from a prompt or a ready-made template, with no rules engine to configure
- Announced data residency choice across the United States, United Kingdom, Canada, Australia, Singapore and India
- A published subprocessor list naming each vendor, its service, its processing locations and its legal address
- Ownership of documents and extracted data contractually left with the customer under article 11.2 of the terms
- Self-serve trial with no credit card, plus a REST API, webhooks, ready-made connectors and public API documentation
Cons
- No public price beyond the trial: Business and Enterprise both lead to a Talk to us button, and there is no permanent free plan, only 14 days
- The site contradicts itself on the trial, advertising 1,000 pages and 10 user licences on the pricing card, the structured data and the homepage against 100 pages and one user at a time in the FAQ on that same page
- Plan names do not line up either: Free, Business and Enterprise on the cards, Free, Growth and Enterprise in the pricing FAQ
- Uploaded documents are used to train the AI with no documented opt-out, and no customer DPA is published or announced on request
- SOC 2 Type 2, HIPAA and GDPR are asserted in page copy and by footer badges, with no trust page, no report and no named auditor, so none of it can be verified publicly
- The terms cap liability at fees already paid, exclude any refund including on downgrade, warn that data may be transferred unencrypted, and limit default retention to 30 days since this is not a storage service
- Terms and privacy policy still carry a 4 January 2024 date and never mention the agentic positioning displayed in 2026; support is English-only through privacy@docsumo.com, and there is no mobile app or browser extension
Pricing & Plans
There is no permanent free plan and no published price point in any currency. Access begins with a 14-day free trial requiring no credit card, though the site describes it two ways: the pricing card, the page's structured data and the homepage advertise up to 1,000 pages and 10 user licences, while the FAQ on the same page states 100 pages and one user at a time. Both paid tiers, Business and Enterprise, display a Talk to us button instead of a figure. The FAQ describes usage-based pricing charged per page processed, with the per-page cost falling as volume rises, and Enterprise priced entirely to order. Article 4.1 of the terms states that fees are quoted in euros or in dollars excluding tax; billing is in advance, monthly or annually, with tacit renewal and 30 working days' notice to terminate, and article 4.2 excludes any refund, including for partial months, unused months, upgrades or downgrades. Annual subscriptions are advertised as saving up to 10%. Payment methods are also stated inconsistently: article 4.2 allows credit cards only, namely Visa, Mastercard and American Express, while the FAQ mentions prepaid credit or debit card payments and bank transfer for annual subscriptions.
- a 14-day trial rather than a permanent tier. The pricing card lists up to 1
- 000 pages and 10 user licences
- unlimited pre-trained models
- field and table extraction
- AI-document reviewer
- AI-prompting and refinement
- continuous learning
- ready-made integrations
- everything in Free
- plus unlimited user licences
- master data lookup and basic mapping
- auto-classification and splitting
- AI-transform for validations
- classification and categorisation
- custom document pipelines
- granular permissions
- everything in Business
- plus AI-driven automated workflows
- AI-driven case management
- cross-document validations and real-time document analytics. The FAQ adds custom user and page limits
- a dedicated customer success manager
- SSO and SAML
- priority Slack or Teams support and procurement assistance. Price on request
- the card carries both Most popular and Best value labels.
- a fourth plan named only in the pricing FAQ
- described as 5
- 000 pages per month and free for the first 14 days. It appears on no pricing card
- and the site never reconciles the two plan line-ups.
Data, GDPR & hosting
A consolidated view of how Docsumo handles your data.
GDPR overview
GDPR compliance is claimed explicitly, not merely implied: a GDPR Compliant badge in the footer, an FAQ answer stating the product is GDPR-compliant for European operations, and article 14 of the terms, effective 4 January 2024, listing measures taken under the GDPR, the Data Protection Act 2018, Singapore's PDPA 2012 and California's CCPA 2020. Docsumo positions itself as controller for personal data and processor for imported documents and parsed content, and relies on legitimate interest to train its models. Announced rights are access, rectification, modification, erasure, a copy of the data, withdrawal of consent and restriction of marketing use, all through privacy@docsumo.com, answered within 30 days. The limits are equally clear: no Article 27 EU representative, no DPO, no EU entity or address, no customer DPA published, and a privacy policy governed by Singapore law before Singapore courts.
Who owns the data?
Under the terms of service and the privacy policy, both effective 4 January 2024, Docsumo Pte. Ltd. expressly disclaims ownership of customer material: article 11.1 waives any claim over processed documents and content, and article 11.2 leaves the client owner of its processed documents, content and parsed content. The publisher retains intellectual property only in its own platform and website. Legally, Docsumo acts as data controller for personal data and as data processor for imported documents and parsed content, and states it never sells user data or data extracted from uploaded documents. One caveat sits in the privacy policy: data may be passed to third parties for marketing purposes at the sole discretion of Docsumo employees.
Reuse rights
Because the client keeps ownership of its documents and of the parsed output, it may reuse that output freely, without asking Docsumo: export to Excel or JSON, or push it through the API and webhooks into an ERP or CRM. What Docsumo does with the same material is broader. The privacy policy, effective 4 January 2024, states that uploaded documents are retained and used for further research, development and training of its artificial intelligence before deletion, on the legal basis of Docsumo's legitimate interest under the GDPR, the Data Protection Act 2018, Singapore's PDPA 2012 and California's CCPA 2020. Registration details, billing data and behavioural data also feed internal records, product improvement, promotional email, third-party marketing, automated decision-making and profiling. No opt-out from model training is documented. Note one unresolved contradiction: the FAQ and pricing page describe a stack built entirely in-house, eliminating leaks to third-party models, while the subprocessor page declares Anthropic PBC and OpenAI OpCo LLC as AI model API providers processing in the United States and the EU.
Data retention & training
Hosting summary
Docsumo runs on Amazon Web Services and Google Cloud Platform, per article 6 of the privacy policy, effective 4 January 2024. The FAQ states that documents are stored in AWS across six regions, the United States, United Kingdom, Canada, Australia, Singapore and India, with the actual location depending on the customer's region or project settings. The subprocessor page widens the picture: AWS, Google Cloud and MongoDB Atlas process in the United States, EU, United Kingdom, Australia and India, Cloudflare is listed as Global, Anthropic and OpenAI in the United States and the EU, and Sentry in the United States. Operationally the vendor describes daily full backups, data spread across several disks, servers kept patched, access restricted to authorised employees, hashed passwords, SSL in transit and AES-256 at rest, while article 9 of the terms warns transmissions may be unencrypted. Enterprise plans offer configurable retention. Two caveats: the six storage regions advertised do not include the European Union even though EU processing appears on the subprocessor list, and the resolved IP address is a Cloudflare anycast node, which says nothing about where the origin servers sit.
Things to keep in mind
Risks and trade-offs to weigh before adopting Docsumo.
- Automation complacency: with 95%+ straight-through processing claimed and only exceptions surfaced, reviewers who see almost nothing go wrong for weeks stop reading carefully, and the confidence score quietly becomes a rubber stamp
- Uploaded documents train the vendor's AI on a legitimate-interest basis with no documented opt-out, so whatever a colleague drops in without thinking, a payslip, an ID card, a client contract, goes with it
- The site contradicts itself on third-party models: the FAQ and pricing page claim a stack built entirely in-house that eliminates leaks to third-party models, while the subprocessor page names Anthropic PBC and OpenAI OpCo LLC as AI model API providers processing in the United States and the EU
- Encryption and payment statements conflict too: marketing promises bank-grade SSL and AES-256 at rest while article 9 of the terms warns data may be transferred unencrypted, and the terms accept credit cards only where the FAQ adds debit cards and bank transfer
- The trial is described two ways on the same page, 1,000 pages and 10 licences on the card against 100 pages and one user in the FAQ, and a fourth plan named Growth exists only in that FAQ: budget on either version and you may be wrong
- Content is deleted after 30 calendar days by default and the terms state the service includes no storage, so treat it as a pipeline and keep your own archive rather than discovering later that documents you assumed were filed are gone
- Compliance claims outrun the evidence: SOC 2 Type 2, HIPAA and GDPR are asserted in copy and badges with no trust page, report or auditor, there is no EU representative or DPO, the privacy policy is governed by Singapore law and dated 4 January 2024, and it permits passing data to third parties for marketing at staff discretion
Setup & Integrations
Technical difficulty
Low to moderate. Sign-up is self-serve with no credit card, and the FAQ describes an intuitive interface requiring no coding knowledge: pick one of 100+ pre-trained models, write validation rules in plain English, and build a workflow from a template or by describing it. A custom model needs about 20 sample documents. A technical path exists for deeper work through the REST API, webhooks and connectors, and enterprise deployments add SSO with SAML 2.0 or OAuth 2.0, role-based permissions and a production-like sandbox. Docsumo advertises guided setup: a four-to-eight-week success plan, live training and a dedicated automation expert.
Deployment
Integrations
Supported languages
Behind Docsumo
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Docsumo.
Frequently asked questions
What is Docsumo?
Which document types and file formats does it handle?
Is there a free plan?
How much does Docsumo cost?
Is there an API?
Are my documents used to train Docsumo's AI?
Where is my data hosted?
How long is data kept?
Which certifications does Docsumo hold?
Who publishes Docsumo and what support is available?
Should you pick Docsumo?
Docsumo is a mature product from a funded, documented publisher: $3.5 million of seed money in March 2022, a Singapore entity with a public registration number, a named CEO and CTO, and traction figures it puts in writing. Its real strength is traceability. Every extracted value carries a confidence score and a link back to its exact position in the source page, every workflow step is logged, and the document coverage is genuinely vertical, from ACORD forms and 1003 mortgage packets to bills of lading and Medicaid paperwork, rather than generic OCR. For an operations team in financial services, insurance, real estate, healthcare or logistics that has to justify a figure to an auditor, that combination is the argument.
The reservations are just as concrete. Pricing beyond the trial is entirely opaque: two of the three tiers show a contact button and nothing else, and the site cannot keep its own trial story straight, advertising 1,000 pages on the pricing card and 100 in the FAQ on the same page, under plan names that differ between the two. Uploaded documents are used to train Docsumo's models on a legitimate-interest basis, with no documented opt-out and no published DPA. SOC 2 Type 2, HIPAA and GDPR appear in page copy and footer badges, but no trust page, audit report or auditor is published, so a buyer cannot verify them from the site. And the terms and privacy policy still carry a 4 January 2024 date, describing an extraction SaaS that never mentions the agents the 2026 marketing is built on.
Worth a trial for mid-sized and enterprise operations teams with real document volume, provided procurement goes in expecting a sales conversation on price, a written answer on model training, and evidence behind the compliance badges.
- Choosing a selection results in a full page refresh.
- Opens in a new window.