Automated data capture: technologies, costs and how to choose software

Automated data capture is software that reads documents, forms and images. It turns what it reads into data your systems can use, with no one typing. Some people call it automatic data capture. Simple jobs use barcode readers, OCR or ICR. Business documents with varied layouts need AI. This AI approach is called IDP.

OCR is optical character recognition, and ICR is intelligent character recognition for hand-printed text. IDP is intelligent document processing. Below are the technologies, a worked example of IDP, what a system costs, and how to choose and roll one out.

What is data capture?#

Data capture, or data capturing, is getting information from its source into a system. Done by hand, people read each document and type the values in. When it's automated, software reads the document and passes the values on, and people review only the exceptions.

Data capture technologies#

• Barcodes and OMR

  Read data encoded in advance, such as barcodes, QR codes, filled bubbles and checkboxes. They can't read free text.

• OCR

  Reads printed text from scans and images. It returns text, not fields, and struggles with poor scans.

• ICR

  Reads hand-printed characters, usually one per box on a form. Messy or cursive handwriting trips it up.

• Zonal (template) OCR

  Reads text in set areas of a page whose layout you already know. It fails when the layout changes.

• Intelligent document processing (IDP)

  Finds fields and tables in any layout, printed or handwritten, and checks them. Accuracy varies by document, so prove it with a pilot on yours.

Natural language processing (NLP) finds values by meaning in free text, such as a lease's renewal date. IDP uses NLP inside extraction. Robotic process automation (RPA) doesn't read documents. RPA copies captured data into systems that have no API, after capture. An API is a direct connection for other software. More on ICR, zonal OCR and IDP vs OCR.

How automated data capture works#

IDP-based capture is also called cognitive data capture or cognitive capture. It works out what each value is from the words around it, not from where it sits on the page. It handles layouts it hasn't seen before.

Take a mortgage application that arrives by email with a bank statement, a pay stub and a tax return attached. The system first classifies each file, so the statement is not read as a stub. It then extracts the fields and tables from each one, whatever the lender's or employer's layout. Every value comes with a confidence score.

Next comes validation, which means checking the values against rules. Rules check each document on its own, for example that the statement's rows add up to the statement's total. Rules can also compare documents, such as the income on the stub against the deposits on the statement. Cross-document checks like these are part of Docsumo's Enterprise plan.

Last, only uncertain or failed values reach a reviewer, shown next to the source page. If every check passes, the file goes straight to the loan origination system. If one deposit does not match, the reviewer sees only that deposit.

Manual vs automated data capture#

On Docsumo, the difference looks like this:

• <5 min<5 minper document, down from 2+ hours by hand
• 95%+95%+of documents processed straight through, without manual review
• $15$15saved per processed document

What does an automated data capture system cost?#

Budget for more than the license:

• Software. Priced per page, per document, by subscription or by enterprise license.
• Setup. Document types, fields, validation rules and any custom models.
• Maintenance. Updating templates or models as layouts change. Template-based systems need the most.
• Review time. People still check exceptions, so a higher straight-through rate lowers this cost.
• Integration. Connecting the output to your ERP, loan origination system or CRM.

Measure the return in cost per document and hours saved.

Industries that use automated data capture#

Document data capture reads business documents. It pays off wherever teams retype values from paperwork.

• Lending

  Bank statements, pay stubs, tax returns and rent rolls for income checks and underwriting. See IDP for lending and CRE underwriting.

• Accounts payable

  Invoices, receipts and statements, captured and checked before they're posted. See invoice extraction.

• Insurance

  ACORD forms, claims and certificates of insurance, with each value checked against the policy or contract. See insurance automation.

• Logistics

  Bills of lading, delivery receipts and customs forms for tracking and billing. See bill of lading extraction.

• Healthcare

  Intake forms, referrals and claims, captured into patient and billing systems. See IDP for healthcare.

How to choose automated data capture software#

• Document coveragePre-trained models for your main documents, and a way to add your own.
• ValidationConfigurable rules, plus checks across the documents in one file.
• Review screenExceptions shown next to the source, with the reason each was flagged.
• IntegrationsAn API and webhooks into your ERP, loan system or CRM. See integrations.
• SecuritySOC 2 Type 2, HIPAA and GDPR where your data needs them. See security.
• PricingCost at your expected volume, and what the free trial covers.

For a side-by-side of vendors, see the best data capture software.

How to roll out automated document data capture#

1. Map your workflowsPick the document-heavy processes, such as invoice approval or claims intake, and record today's volume, time per document and error rate.
2. Define the fieldsList the fields each process uses and their formats. You rarely need every value on a page.
3. Set input standardsAgree a minimum scan quality. Blurry photos and low-resolution faxes lower accuracy with any tool.
4. Pilot one document typeRun it alongside the current process, bad scans included, and compare field-level accuracy and straight-through rate.
5. Train the team on reviewStaff move from typing to checking exceptions, so involve them early.
6. Measure and expandTrack error rate, straight-through rate and cost per document, then add the next document type.

Does the whole loan file agree?#

Our take. We'd judge a capture system by the whole loan or claim file, not the single page. Lending and insurance decisions depend on several documents agreeing with each other. The checks between documents matter more than reading one page perfectly.

Automated capture replaces typing with checking. Match the technology to the input, and use IDP for varied business documents. Then compare systems on accuracy with your documents, plus review and integration.

Book a demo with a few of your own documents, or start a free trial.

Frequently asked questions#