What we build

Files in, clean and checked data out.

Software that reads your documents, pulls out the fields you need, checks them and files them.

Last updated

AI document and data extraction reads your files, sorts and names them, pulls out the fields you need and checks them against your rules. Invoices, loan files, contracts and mixed batches suit it. You know the output is right because every field traces to its source file, each step is recorded, and a person signs off on what matters.

What is document and data extraction?

Your people open a file, find the figures or terms that matter, and type them somewhere else. This software does that reading. It takes each file, works out what kind of document it is, pulls out the fields you named, and compares them with your rules. Anything that does not match goes to a person with the page it came from.

Files arriveEmail, upload or shared folder
Read and sortedEach file named and filed
Checked against your rulesBefore anything is confirmed
A person signs offOn outputs that matter
Clean data outRecord written for every step

Is it the right fit?

It fits when

  • The same kinds of document arrive every week or every day, in volume.
  • Someone retypes figures from files into a spreadsheet or system.
  • You can state the rule that says whether a field is right.
  • You need to show where each figure came from.
  • Files arrive mixed, unlabeled or badly named.

It is the wrong tool when

  • The job happens twice a year and takes an afternoon.
  • The format changes every time and no rule holds across files.
  • A person must exercise judgment on every page.
  • An off the shelf tool already does the job at your volume.
  • The source is a clean export, so no reading is needed.

What can you build for a firm like ours?

These are illustrations of document jobs we can build, grouped by industry. They are examples of what is possible, not descriptions of client work. The real work is further down.

CPA and accounting firms

Example

Bank and brokerage statements into a transaction schedule

Today
Staff retype each line of every statement into a workpaper, then foot it against the closing balance.
The build
The build reads each statement, writes every transaction into one schedule with its file and page, and checks that the lines add up to the stated balances.
A person still decides
A named preparer reviews each flagged difference.
Example

K-1s and 1099s into the tax workpapers

Today
Preparers key figures from each form into the workpaper, one box at a time, during the busiest weeks.
The build
The build identifies the form type, reads the boxes and places each figure in the right workpaper line. It flags missing forms against last year's list.
A person still decides
The preparer decides how each item is treated.

Investment advisers and asset managers

Example

Client onboarding forms into the CRM record

Today
Operations staff read each signed form and type names, accounts and contact details into the client record.
The build
The build reads the form, fills the record fields and checks for blanks, mismatched names and missing signatures. Incomplete forms go back with a list of what is missing.
A person still decides
A named person approves the record before it is used.
Example

Custodial statements checked against the portfolio system

Today
An analyst compares positions and balances on each custodian statement with the portfolio system, line by line.
The build
The build reads each statement, matches positions to the portfolio export and lists every difference with the statement page. Matching lines are marked and need no review.
A person still decides
The analyst decides which side is correct for each difference.

Law firms

Example

Discovery productions sorted and indexed

Today
Paralegals open each file in a production, name it, and add it to an index by hand.
The build
The build reads every file, assigns a type, a date and a name from your convention, and writes the index. Files it cannot place go to a review list.
A person still decides
An attorney decides what is relevant or privileged.
Example

Contract terms pulled into a clause table

Today
A lawyer reads each agreement and copies parties, dates, renewal and termination terms into a spreadsheet.
The build
The build reads each agreement and fills a clause table, with the page of each term. It marks clauses that differ from your standard wording.
A person still decides
A lawyer reads every marked clause and decides what it means.

Mortgage and lending

Example

Loan file documents checked against the conditions list

Today
A processor goes through the file folder by hand to see which conditions have a document and which do not.
The build
The build sorts each document in the file, matches it to a condition on the list and shows what is outstanding. It notes expired or unsigned documents.
A person still decides
The underwriter decides whether a condition is satisfied.
Example

Income documents into a qualification worksheet

Today
Staff read pay stubs, W-2s and tax returns and key the income figures into a worksheet.
The build
The build reads each income document, enters the figures in the worksheet with their source pages and flags figures that disagree between documents.
A person still decides
A loan officer or underwriter confirms the income used.

Operations-heavy businesses

Example

Delivery tickets and scale tickets into invoices

Today
Office staff match paper delivery and scale tickets to orders, then key quantities and weights into each invoice.
The build
The build reads each ticket photo or scan, matches it to the order and drafts the invoice lines. It flags tickets with no order and weights outside the order.
A person still decides
A billing clerk approves each invoice before it goes out.
Example

Vendor invoices coded to the ledger

Today
An accounts payable clerk reads each invoice, picks the account and cost center, and keys it into the ledger.
The build
The build reads the invoice, proposes the coding from your past entries and checks the total against the purchase order. Duplicates are held back.
A person still decides
The clerk approves coding that falls outside your rules.

Have a pile of documents someone retypes by hand?

Send us the kind of file and the rule that says it is right. We will say what a build would take.

Built and running

You can trace any field back to the file it came from.

Built and running

A forensic accounting practice

Before
Document intake was manual and ate billable hours.
What it does
Intake automation reads, names and files case documents and checks statements. It checks before it confirms, keeps an append-only record of every step, and runs in a private cloud environment.
A person still decides
The expert analysis stays with the people.
Result
100% of automated document steps are captured in the record. It is used daily by every role at the practice, from staff to partners.

One client's intake went from five days of manual sorting per case to about an hour.

Case Manager by the numbers

  • 18,748 documents

    Read, named and filed in 30 days for one CPA firm, about 160,900 pages across 50 active cases.

  • 85% filed unaided

    85% of documents were filed with no question to staff.

  • About an hour

    A typical delivery is fully classified within about an hour of arrival.

  • 6,570 documents in 3.8 hours

    Its largest single delivery. The manual sort it replaced took five or more days per case.

Read the story See Case Manager on the Solutions page

Anonymized at client request. Reference calls available under NDA.

How does a build work?

Step 1ScopeOne working session. A plan and a price.
Step 2BuildOn your own documents, shown early and often.
Step 3LaunchYour people trained on their real work.
Step 4RunHosted, watched, improved.

The plan sets the timeline. For larger builds there is an optional scoping sprint: two weeks and a working prototype on your own documents.

  1. Collect sample documents, including the awkward ones.
  2. Agree the fields to pull out and the rules that say a field is right.
  3. Build and test against your own files.
  4. Review the exceptions together and adjust the rules.
  5. Train the people who will work with the output.
  6. Switch on with the record running, then watch and improve.

What do you get at handover?

  • Working software in your environment

    Running on your own documents.

  • The record of actions

    An append-only log of every step.

  • Written rules

    What the software may do and what it checks.

  • Training

    For the people who use it, on their real work.

  • A named security review

    Done before launch.

  • Someone to call

    DoubleChecked runs it: hosted, watched, improved.

How is our data handled?

The software runs in a private, client-controlled environment, and client data does not go to public AI services. DoubleChecked contracts directly with Anthropic under commercial terms and a data processing agreement, with no reseller in between. Zero data retention is available through the API, not the chat interface. A named security review of each build happens before launch. Read how we handle security

Who can get inSingle sign-on and roles
Where it runsA private environment you control
What it may doChecks before it confirms an action
Who signs offA person at your company
What it didAn append-only record

What does it connect to?

  • Accounting and general ledger
  • Document management
  • Email and calendar
  • CRM
  • Spreadsheets
  • File shares and line of business systems

Questions buyers ask

How long does it take?

The plan sets the timeline. After one working session you have a plan and a fixed price. You then see working software early and often during the build, so you are never waiting for a single reveal.

What does it cost?

Every project gets a fixed price after one working session, so you know the number before we build. A scoping sprint is an optional first step for larger builds: two weeks and a working prototype on your own documents.

Where does our data go?

The software runs in a private, client-controlled environment. Client data does not go to public AI services. DoubleChecked contracts directly with Anthropic under commercial terms and a data processing agreement, with no reseller in between.

Who checks the output?

The software checks before it confirms an action and keeps an append-only record of what it did. A person at your firm reviews and signs off on outputs that matter. Each build gets a security review before launch.

What happens when it is wrong?

Fields that fail a rule go to a person with the file and page they came from. The correction is recorded, and the rules are adjusted so the same miss is caught next time.

What do you need from us to start?

One working session, 45 minutes on one idea from your company, and some sample documents. DoubleChecked signs an NDA before that session and replies within one business day.

Can it read scans and photos, not just clean PDFs?

Scans and photos are part of the job. We test on your own files, including the poor ones, during the build. Documents it cannot read with confidence go to a person rather than being guessed.

Tell us what you want built.

A working session is 45 minutes on one idea from your company: what to build, what it needs, and what it would cost.

  • A fixed price before we build
  • Your data stays in your environment
  • A record of every action
  • Reference calls under NDA
  • NDA before the first session
  • Since 2019
Start a project