Intelligent document processing

Every field extracted,
scored, and accounted for.

DAN reads invoices, federal forms, contracts, statements, and scanned records, then returns structured data with a confidence score on every field and a reviewer trail on every decision. Built for teams that have to defend the output, not just produce it.

Uncertain values route to a named reviewer, not straight to your system of record.
Every extraction, edit, approval, and export is logged for oversight requests.
Runs in government cloud, your private VPC, or fully on premises.

Request a scoped pilot Model your review load

18+

Document types supported out of the box

0 to 100

Confidence score returned on every field

100%

Of exports written to an immutable audit log

Extraction console

Ready

Extracted fields

Confidence

Illustrative sample data. Field names, values, and confidence figures shown are for demonstration.

Built for

Federal agencies

Shared services, finance offices, program bureaus, and records management.

State and local government

Benefits, permitting, licensing, courts, and county clerks.

Enterprise operations

Accounts payable, procurement, shared service centers, and audit teams.

Systems integrators

Embed extraction inside a modernization program under your own delivery contract.

Certified

SOC 2 Type 2 ISO/IEC 27001:2022 ISO/IEC 42001:2023 ISO 9001:2015 ISO 22301:2019 ISO/IEC 20000-1:2018 CMMI Level 3

ISO/IEC 42001 is the AI management system standard. Most document AI vendors do not hold it, and for a buyer who has to explain how an AI decision was governed, it is the certificate that answers the question.

The backlog

The document is not the problem. The unverifiable copy of it is.

Everything got scanned. What came back was an image and some loose text, and someone still has to retype the number into a finance or case system.

dan.agency.gov/queue

Review queue

14 waiting

INV-40823.pdfInvoice98Posted

grant_q2_close.pdfFFR96Posted

intake_form_p1.tifBenefits85Review

PO-2026-0448.pdfPurchase order99Posted

stmt_jun_2026.pdfStatement88Review

Sorted by confidence, lowest first

Keyed by hand

Vendor

Meridian Supply Co.

Invoice number

INV-40823

Total

13,451.00

Terms

Net 30

Transposed at entry. Caught at audit, four months later.

Read by DAN

VendorMeridian Supply Co.99

Invoice numberINV-4082398

Total13,541.0098

TermsNet 3087

3 posted automatically. 1 sent to a named reviewer with the page attached.

Illustrative. Same document, two ways of getting the number out of it.

What you have today

/Repetitive keying
/Spreadsheet cleanup
/Copy and paste from OCR text
/Verification delays
/Errors found at audit
/No reviewable record

What replaces it

+Field level extraction
+Confidence on every value
+Rules before export
+Side by side review
+Signed action history
+Clean export to the system of record

The reviewer workspace

The person signing this has to explain the output, not the algorithm.

A comptroller, an inspector general, or an oversight committee will ask where a number came from and who accepted it. This screen is the answer: the extracted value, the page it was read from, and the person who signed off.

Traceable

Click a field and the source page highlights where it came from

Attributable

Every approval carries a named person and a timestamp

Contained

Runs inside the boundary you authorize, on your own tenancy

dan.agency.gov/review/doc_9f31c204 a.reyes

Queue14

Review3

Cases9

Schemas

Audit log

Exports

INV-40823.pdf

ap_invoice_v3 · threshold 92

1 to review Approve

Page 1 of 1

INVOICE

Meridian Supply Co. · Columbus OH

Invoice no.

INV-40823

PO-2026-0448

Terms

Net 30

Date

14 Jul 2026

Subtotal12,480.00

Tax1,061.00

Total due13,541.00

Extracted fields · 5 of 8

vendor_name99

Meridian Supply Co.

invoice_number98

INV-40823

po_reference94

PO-2026-0448

total_amount98

13,541.00

payment_terms87

Net 30

payment_terms below threshold. Confirm against the page or correct it.

Document 3 of 47 in queue

Aapprove Jnext

Illustrative interface. Field names and values shown are sample data.

Intake

Pick documents up from wherever they already are

Nobody should have to move files into one place before the work can start. DAN connects to the systems your documents already live in, watches the folders and mailboxes they arrive in, and takes hardcopy straight off a scanner or a phone camera.

Connected sources

Google DriveCloud storage

SharePointCloud storage

GmailEmail intake

OutlookEmail intake

Scanner and MFPHardcopy

Mobile photoHardcopy

Watched folderNetwork share

REST APISystem to system

Arriving now

Queued

FY26_invoice_batch.pdf

Invoice · 14 pages

SourceGoogle Drive

Language detectedEnglish

ContentTyped

PO-2026-0448.pdf

Purchase order · 2 pages

SourceSharePoint

Language detectedEnglish

ContentTyped

facture_avril_2026.pdf

Invoice · 3 pages

SourceGmail

Language detectedFrench

ContentTyped

Rechnung_88213.pdf

Invoice · 1 page

SourceOutlook

Language detectedGerman

ContentTyped

intake_form_p1.tif

Benefits application · 4 pages

SourceScanner and MFP

Language detectedSpanish

ContentHandwritten

receipt_0714.jpg

Receipt · 1 page

SourceMobile photo

Language detectedEnglish

ContentHandwritten

contrato_servicios.pdf

Contract · 22 pages

SourceWatched folder

Language detectedPortuguese

ContentTyped

doc_9f31c204

Delivery note · 2 pages

SourceREST API

Language detectedArabic

ContentTyped

Applied on arrival

+Document type classified
+Language detected
+Extraction schema matched
+Queued with source recorded in the audit log

Illustrative intake stream. Connector availability is confirmed for your deployment during scoping.

Cloud storage

Where the files sit

Point DAN at a Google Drive folder or a SharePoint library and it processes what is there, then keeps watching for what lands next.

Google Drive
SharePoint and OneDrive
Network and watched folders
Object storage buckets

Where they get sent

Give the platform a dedicated intake address, or connect the shared mailbox vendors and applicants already write to. Attachments are pulled and queued automatically.

Gmail and Google Workspace
Outlook and Microsoft 365
Shared and functional mailboxes
Forwarding intake address

Hardcopy

Paper that never got digitized

Scan from the office multifunction device straight into the queue, or photograph a page on a phone at the counter or in the field.

Scanner and MFP
Mobile photo capture
Bulk backfile scanning
Skewed and low quality scans

Systems

Whatever runs upstream

Post documents from a case system, portal, or existing pipeline, and get the structured result back on a signed webhook.

REST API
Batch drop
Signed webhooks
Portal and case system feeds

Handwriting

Forms filled in by hand still count

Handwritten entries on printed forms, signatures, dates, annotations in the margin, and checkbox marks are read alongside the printed content. Handwriting scores lower than clean print, which is the point: those fields surface for a reviewer instead of posting quietly.

Languages

Multilingual documents, one output schema

Language is detected per document, including mixed language pages and right to left and non Latin scripts. Field names in the output stay in your schema regardless of the source language, so a French invoice and an English one land in the same columns of your system of record.

Sequence

Capture, extract, verify, export.

Four stages, in order, each one recorded. Nothing skips verification on its way to your system of record.

invoice.pdf

Capture
Extract
Verify
Export

Capture
Dashboard, watched folder, email intake, batch drop, or API. PDF, PNG, JPG, WEBP, and multi page scans.
Extract
The document is classified, then read against the schema for its type. You get named fields, not a wall of text.
Verify
Values above your threshold pass. Anything below it opens beside the source page for a named reviewer.
Export
Verified records move to your ERP, financial system, or case platform as JSON, CSV, Excel, or a signed webhook.

Cross document checks

A case is not one document. Neither is the decision.

A benefits determination, a grant closeout, a contract file, an invoice payment: each one is a bundle of documents that have to agree with each other. DAN groups related documents into a single case, runs your rules across the whole bundle, and surfaces only the contradictions.

Names, dates, and identifiers checked for agreement across every document in the case.
Arithmetic reconciled across documents, not just within one page.
Required document checklists enforced, so an incomplete case cannot be approved by accident.
One decision recorded for the case, with the supporting documents attached to it.

Case

BEN-2026-14892

2 exceptions

Documents in case BEN-2026-14892 with source and validation status
Document	Source	Required	Status
application_1003.pdf	Portal	Yes	Passed
proof_of_residency.jpg	Mobile photo	Yes	Passed
pay_stub_jun.pdf	Gmail	Yes	Name mismatch
bank_stmt_may.pdf	SharePoint	Yes	Passed
bank_stmt_jun.pdf	SharePoint	Yes	Missing
id_front_back.pdf	Scanner	Yes	Passed

Cross document rules run

Applicant name consistent · Income totals reconcile · Statement months contiguous · Residency address matches application

Sent to caseworker

Name on pay stub differs from application · June statement not received

Illustrative case. Document checklists and cross document rules are defined per program during onboarding.

What changes

The measures your business case will be judged on

Other vendors will quote you percentages from somebody else's program. Those numbers were produced by a different document mix, a different scan quality, and a different threshold. Here is what actually changes mechanically, and where your own figures come from.

Operational measures under a manual process compared with the DAN platform
Measure	Manual or legacy OCR today	With DAN
Straight through processing	No concept of it. Every document is opened by a person, whether it needed attention or not.	Fields above your confidence threshold post without a human touch. You set the threshold and can raise it any time.
Where review time goes	Spread evenly across every document, including the clean ones.	Concentrated on the flagged minority. Reviewers see the value beside the source page instead of hunting for it.
Error discovery	At audit, months later, when correcting it is expensive and visible.	At entry. Low confidence values and failed validation rules are caught before the record reaches your system of record.
Case completeness	Checked by memory and a paper checklist. Missing documents surface late.	Enforced by the case checklist. An incomplete bundle cannot be approved by accident.
Audit response	Reconstructed from email, spreadsheets, and memory when the request arrives.	Already written. Every extraction, edit, approval, and export carries actor, timestamp, and prior value.
Scaling with volume	Linear. More documents means more staff or a longer queue.	Volume scales on the platform. Only the flagged share scales with headcount.
Improvement over time	None. The same mistakes recur with each new hire.	Reviewer corrections are retained against your schemas inside your own tenant, so the flagged share falls as the program runs.

We do not put a headline accuracy percentage on this page. The Phase 1 document assessment runs a representative sample of your own documents and reports the measured field coverage and review rate for your mix. That is the number your contracting officer can rely on, and it is the only one we will put in writing.

Where it goes to work

Wherever paper still sets the pace of a program

Accounts payable

Match invoice, purchase order, and receiving note. Flag totals that do not reconcile before payment.

Grants administration

Award identifiers, reporting periods, and expenditure figures out of progress reports and closeout packages.

Benefits and eligibility

Income, residency, and identity evidence from applicant uploads, so caseworkers make determinations instead of transcribing.

dan.agency.gov/rules

Validation

1 failed

Invoice total = sum of linesPassed

PO reference exists in ERPPassed

Invoice date within periodPassed

No duplicate invoice numberFailed

Rules run before export. A failure holds the record.

Permits and licensing

Applications, site plans, and supporting certificates become structured intake records.

Contract files

Parties, period of performance, ceiling values, option years, and renewal dates from contracts and modifications.

Backfile conversion

Decades of scanned case files turned into indexed metadata for retention schedules and records requests.

dan.agency.gov/insights

Straight through rate

90 days

12 weeks94.6%

Rate climbs as reviewer corrections are retained.

Financial reconciliation

Transactions, balances, and statement periods for reconciliation, grants monitoring, or investigation support.

Hiring and onboarding

Resumes, credentials, and certifications parsed into structured candidate records for classification and verification.

Logistics and receiving

Delivery notes, bills of lading, and shipping records update property and inventory without a second entry pass.

dan.agency.gov/exports

Exports

Live

erp.ap_queuePosted

tyler.permitsPosted

case_mgmt.intakeQueued

warehouse.grantsPosted

Signed webhook fires when a document clears review.

Platform

Built for documents you cannot afford to get wrong

Field level extraction

Named fields with source coordinates, so every value traces back to its exact place on the page.

Headers · Line items · Tables · Signature blocks

Confidence scoring

A score from 0 to 100 on every field. Set the threshold per schema or per program.

Per field scores · Auto approval · Review routing

Custom schemas

Define the output shape once. Every extraction conforms to it, using your downstream field names.

Field mapping · Type constraints · Program specific schemas

Multi page processing

Long PDFs stay one document with continuity across pages, not a pile of images.

Long contracts · Financial reports · Case packets

Human verification

Reviewers see the value and the original page together, edit in place, and sign off.

Side by side view · Editable fields · Named approvals

Validation rules

Arithmetic checks, vendor matching, date logic, and duplicate detection before anything exports.

Total checks · Vendor matching · Duplicate detection

Tables and line items

Every row of an invoice, statement, or cost schedule is captured, including merged cells and tables that continue across pages.

Unlimited rows · Merged cells · Page spanning tables

Learns from your reviewers

Corrections are retained against your schemas, so the same field stops getting flagged. Inside your tenant only, never pooled into a shared model.

Tenant scoped · No shared training · Falling review rate

Nothing invented

A value is returned with the coordinates it was read from, or it is returned empty and flagged. The model is not permitted to fill a gap with a plausible guess.

Source anchored · Empty over invented · Inspectable

Security and governance

Reviewable AI. Encrypted by default.

The security review is usually the longest part of a government purchase. Every action in the platform writes a line your assessor can read.

Audit log Live

09:14:02a.reyesapprovedinvoice_numberINV-40823

09:14:09systemextractedtotal_amount13,541.00

09:14:11systemflaggedpayment_termsconfidence 87

09:15:44a.reyescorrectedpayment_termsNet 45

09:16:03a.reyesapprovedpayment_termsNet 45

09:16:20systemexporteddoc_9f31c204erp.ap_queue

09:18:51m.okaforopeneddoc_9f31c218grant report

09:19:12systemflaggedprogram_officerconfidence 85

09:20:35m.okaforapprovedfederal_expended571,230.00

09:21:07systemvalidatedschema ap_invoice_v33 rules passed

09:14:02a.reyesapprovedinvoice_numberINV-40823

09:14:09systemextractedtotal_amount13,541.00

09:14:11systemflaggedpayment_termsconfidence 87

09:15:44a.reyescorrectedpayment_termsNet 45

09:16:03a.reyesapprovedpayment_termsNet 45

09:16:20systemexporteddoc_9f31c204erp.ap_queue

09:18:51m.okaforopeneddoc_9f31c218grant report

09:19:12systemflaggedprogram_officerconfidence 85

09:20:35m.okaforapprovedfederal_expended571,230.00

09:21:07systemvalidatedschema ap_invoice_v33 rules passed

Request the security package

Audit logging

Append only record of every upload, extraction, edit, approval, and export, with actor, timestamp, and prior value.

Encryption

In transit and at rest, with customer managed keys on private and on premises deployments.

Access control

Role based permissions, separated reviewer and approver duties, and single sign on through SAML or OIDC.

Residency and retention

Pick the region, set purge schedules to match your records policy, keep source documents in your own boundary.

Secure integration

Scoped API keys, token authentication, signed webhook events, and IP allow listing.

Your data trains nothing else

Corrections your reviewers make improve your extraction inside your tenant. They are never pooled into a shared model that other customers benefit from.

Control alignment

SOC 2 Type 2 and ISO/IEC 27001:2022
ISO/IEC 42001:2023 for AI management
HIPAA aligned controls, BAA supported
NIST SP 800 53 aligned implementation
FISMA aligned documentation
FedRAMP and StateRAMP paths with a sponsoring agency
Section 508 and WCAG 2.1 AA developed against

Regulated data

Protected health information, controlled unclassified information, criminal justice information, and tax return data are all handled under a scoped control set rather than a single default posture. Tell us which categories your program touches at the start and the deployment model, retention rules, and reviewer permissions are configured around them before anything is processed.

Alignment statements describe how the platform is built and documented. Specific authorizations and certifications are confirmed in writing for your deployment during procurement.

Deployment

Run it where your data is allowed to live

Extraction engine, review interface, and audit store deploy together inside one boundary. Nothing has to leave it.

Your authorization boundary

Egress blocked

Documents, extracted values, and the audit trail stay inside the perimeter you authorize.

Option A

Government cloud

AWS GovCloud or Azure Government inside your tenancy, with regional residency and your own key management.

US regions only
Customer managed keys
Agency identity provider

Option B

Private cloud or VPC

A dedicated single tenant instance in your own cloud account, peered to the systems it needs and nothing else.

Single tenant
Private networking
No shared processing

Option C

On premises and disconnected

Installed in your data center, including environments with no outbound internet access at all.

Air gapped capable
Offline model updates
Local audit store

For your integration team

One endpoint. Clean JSON. Signed webhooks.

Post a document, receive structured fields with confidence values, subscribe to a webhook for the review outcome.

REST API access
Scoped API keys
Token authentication
Batch upload support
Async processing status
Signed webhook events
Structured JSON responses
Retry and replay controls

Request API documentation

Where verified records land

SAPOracleWorkdayNetSuite Microsoft DynamicsTyler TechnologiesAccela ServiceNowSalesforceInfor SnowflakeDatabricks

Records reach these systems through the API, a signed webhook, or a scheduled file export, so a system that is not on this list is an integration question rather than a blocker. Connector scope is confirmed during the rollout phase.

Submit a document POST

curl -X POST https://api.dan.example.gov/v1/documents \
  -H "Authorization: Bearer $DAN_API_KEY" \
  -F "file=@invoice.pdf" \
  -F "schema=ap_invoice_v3" \
  -F "confidence_threshold=92"

Structured response 200 OK

{
  "document_id": "doc_9f31c204",
  "schema": "ap_invoice_v3",
  "status": "review_required",
  "fields": {
    "vendor_name":   { "value": "Meridian Supply Co.", "confidence": 99 },
    "invoice_number":{ "value": "INV-40823",           "confidence": 98 },
    "po_reference":  { "value": "PO-2026-0448",        "confidence": 94 },
    "total_amount":  { "value": "13541.00",            "confidence": 98 },
    "payment_terms": { "value": "Net 30",              "confidence": 87 }
  },
  "routed_to_review": ["payment_terms"],
  "audit_ref": "aud_2026_07_14_0113"
}

Evaluation

What an evaluation committee usually asks about

Most of the field competes on accuracy percentages. In a government or enterprise evaluation, the questions that actually decide it are where the data sits, who can see it, and whether the output can be defended.

Comparison of manual entry, traditional OCR, generic AI tools, and the DAN platform
Capability	Manual entry	Traditional OCR	Cloud only IDP	DAN platform
Structured field output	None	Partial	Schema enforced	Schema enforced
Per field confidence score	None	None	Every field	Every field
Human review workflow	Is the workflow	External	Built in	Built in
Traceability to source page	None	Limited	Varies	Field coordinates
Audit log for oversight	Spreadsheets	None	Vendor hosted	Append only
Custom extraction schemas	Not applicable	Template bound	Versioned	Versioned
API and webhook automation	None	Rare	REST and events	REST and signed events
Cross document case validation	Manual	None	Varies	Built in
Your corrections stay yours	Not applicable	Not applicable	Often pooled	Tenant scoped
Government cloud deployment	Not applicable	Rare	Rare	GovCloud and Azure Gov
On premises or air gapped option	Not applicable	Sometimes	Typically none	Supported

Buying and engagement

Scoped to your program, priced to your contract

There is no public price list. Volume, schema count, deployment model, and support level all move the number, so we scope against your actual mix and give your contracting officer a firm figure.

Most agencies start with a paid pilot on one document type in one office, then expand once they have seen the review rate in their own environment.

Start a scoping conversation

dan.agency.gov/insights/program

Program metrics

Last 90 days

Documents processed184,220

Straight through rate94.6%

Reviewer hours66 / month

Exceptions resolved99.1%

The figures your contracting officer signs off against, measured on your own documents.

Typical engagement path

Phase 1

Document assessment

We run a representative sample of your real documents and report back on field coverage and expected review rate.

Phase 2

Scoped pilot

One document type, one office, live volume, with schemas configured and reviewers trained on your workflow.

Phase 3

Program rollout

Expansion across document types and offices, with system of record integration and ongoing support.

Contracting

We work with your contracting office to identify a suitable path, which may include a federal wide acquisition or multiple award schedule vehicle, a state term contract or cooperative agreement, a task order under an existing modernization contract, or a subcontract and teaming arrangement with your prime integrator. Current vehicle availability is confirmed in writing before you rely on it.

Standard terms available for legal review
Section 508 accessibility conformance reporting
Security questionnaire and control matrix
References supplied on request under NDA

Questions

The ones that come up in evaluation

Which file types and formats are supported?

PDF, including multi page and scanned, plus PNG, JPG, and WEBP. Handwriting is supported, though cleaner scans mean higher confidence and less review work.

How is this different from the OCR we already own?

Traditional OCR returns text. DAN returns named fields against a schema, each with a confidence score and a page location, plus the review workflow and audit trail an auditor will ask for.

Can our documents stay inside our own environment?

Yes. Government cloud inside your tenancy, a single tenant private cloud instance, or on premises in your own data center, including disconnected environments.

Who is accountable when the model gets a field wrong?

You set the threshold and decide which fields a model may approve on its own. Everything below it routes to a named reviewer, and the log records who accepted or corrected each value. Accountability stays with your staff.

Can we define our own fields for agency specific forms?

Yes. Custom schemas are configured during onboarding using your field names and data types, and they are versioned, so a form change does not silently change the output shape.

How does the data get into our system of record?

Export as JSON, CSV, or Excel, pull through the REST API, or receive a signed webhook when a document clears review. ERP and case system integration is scoped during rollout.

Is the platform accessible for our staff?

The reviewer interface is developed against Section 508 and WCAG 2.1 AA, with keyboard operable review, visible focus, and screen reader labelling. Conformance reporting is provided during procurement.

What does a pilot involve on our side?

A representative sample, one program owner to define target fields, and a small reviewer group. We handle schemas, setup, and training, then report measured review rates against your own documents.

Let’s Talk About Your Product

Get expert guidance on scope, architecture, timelines, and delivery approach so you can move forward with confidence.

Every field extracted,scored, and accounted for.

The document is not the problem. The unverifiable copy of it is.

The person signing this has to explain the output, not the algorithm.

Pick documents up from wherever they already are

Where the files sit

Where they get sent

Paper that never got digitized

Whatever runs upstream

Forms filled in by hand still count

Multilingual documents, one output schema

Capture, extract, verify, export.

Capture

Extract

Verify

Export

A case is not one document. Neither is the decision.

The measures your business case will be judged on

Wherever paper still sets the pace of a program

Accounts payable

Grants administration

Benefits and eligibility

Permits and licensing

Contract files

Backfile conversion

Financial reconciliation

Hiring and onboarding

Logistics and receiving

Built for documents you cannot afford to get wrong

Field level extraction

Confidence scoring

Custom schemas

Multi page processing

Human verification

Validation rules

Tables and line items

Learns from your reviewers

Nothing invented

Reviewable AI. Encrypted by default.

Audit logging

Encryption

Access control

Residency and retention

Secure integration

Your data trains nothing else

Run it where your data is allowed to live

Government cloud

Private cloud or VPC

On premises and disconnected

One endpoint. Clean JSON. Signed webhooks.

What an evaluation committee usually asks about

Scoped to your program, priced to your contract

Document assessment

Scoped pilot

Program rollout

The ones that come up in evaluation

Let’s Talk About Your Product

What happens next?

Every field extracted,
scored, and accounted for.