How to automate data entry from PDFs and emails

The pattern is the same in almost every office. Documents arrive as PDF attachments: invoices from vendors, orders from customers, statements, delivery receipts, applications. Someone opens each one, reads it, and types the same values into an accounting system, an ERP or a spreadsheet. It is slow, it is easy to get wrong, and the person doing it could be doing something better.

Automating it does not mean trusting a machine blindly. Done well, software does the typing and a person checks only what the software was unsure of. This guide shows the flow, how to choose the right tool, and how to set it up with Paige, Ademero's cloud document AI.

Receive, read, check, deliver

Every working setup has the same four steps:

1. 01

   Receive

   Documents land in one place without anyone saving attachments by hand. With Paige, every job has its own email address; you forward the mailbox or set a rule that copies the right messages to it.
2. 02

   Read

   The software separates a PDF that holds several documents, works out what each one is, and pulls the fields and table rows you would have typed.
3. 03

   Check

   Every value is checked. Anything missing, low in confidence or of an uncertain type is held for a person, with the reason, instead of flowing through as a quiet mistake.
4. 04

   Deliver

   Clean data and a searchable copy of the document go to the system that needs them, named and filed the same way every time.

Start with what you retype

Before choosing anything, spend an hour writing down what your team actually keys today. Take one real document and list every value that gets typed. For the sample Bowman Fabrication invoice in our screenshots, that is:

Do the same for each kind of document that arrives, then add three more facts: roughly how many arrive each week, which mailbox they come to, and which system each value ends up in. That list becomes your setup: the document types, the fields, the address to forward, and the destination.

PC or cloud? Choosing between CapturePoint 6 and Paige

Ademero makes two document AI products that both read fields and line items. They suit different starting points, so pick by where your documents begin and where the data has to go.

If most of your volume is paper, start with CapturePoint 6: it has a free trial that starts on first launch. If it arrives by email, read on.

Step-by-step setup with Paige

Plan on an afternoon for the first job. Start with one document stream, usually the one that takes the most typing.

1. 01

   Create a job for one stream

   Sign up, then make a job named for the work, such as “AP invoices”. A job is one stream of documents with its own setup, address and destinations; add more jobs later for orders or statements.
2. 02

   Teach it with real samples

   Upload a folder of recent documents from the mailbox: your biggest senders, a few unusual ones and at least one poor scan. Paige reads them and proposes the document types, fields and tables. Confirm, rename fields to match your inventory, and remove what you do not need.
3. 03

   Point the email at it

   Copy the job’s email address and forward the shared mailbox to it, or add a rule that forwards only messages from vendors. PDF, TIFF, PNG and JPEG attachments become pages, including files inside forwarded emails, ZIP files and Outlook .msg files. Logos, signature images and auto-replies are recognized and left out.
4. 04

   Decide who may send

   Optionally switch on “Only listed senders” and list addresses or whole vendor domains. Paige checks the sender’s address, so treat the list as a filter, not a lock.
5. 05

   Set file naming

   Choose folders and file names per document type from the type, field values, dates and your own words, for example invoice, then month, then vendor, invoice number and date. Sensitive values such as tax ids are kept out of the suggested names.
6. 06

   Choose where results go

   Download is always there. Add an SFTP folder your accounting system imports from, or a webhook if a developer will post the data into your system. Use the Test button on each before you rely on it.
7. 07

   Run it on Human review first

   For the first weeks, every document Paige is unsure of waits for a person. Keep a note of what gets fixed. Each correction teaches the job, so the list of held documents should shrink.
8. 08

   Set retention, then widen

   Choose how long the job keeps documents after export: 7 days, 30 days, a year or forever. Once the first stream runs smoothly, add the next.

Controls that keep the data right

The worry with automation is a wrong number reaching the ledger without anyone noticing. These are the controls that prevent it.

A check on every document

Every reading ends in a check. A document passes only when its values were found and Paige is confident of them and of its type. A document with a missing value, a low-confidence value or an uncertain type is held, and the reason travels with it. A document on which nothing at all was read is always held, whatever the job's settings, so nothing goes out empty.

People review only what was held

On Human review, held documents wait in Paige until a person confirms them. The reviewer sees the page beside the values and the line items below, fixes what is wrong, and confirms. Two reviewers are never handed the same document at once, and confirming needs the reviewer role, so you decide who can release data.

Choose how much runs on its own

Documents that pass the check go straight out in every mode. Many teams keep money documents on Human review and move routine ones, such as delivery receipts, to an automatic mode once they trust them.

Nothing lost on the way

• Email you can audit. Each job lists its recent messages: subject, sender, which files became documents, and any file left out with the reason.
• One bad file costs only itself. A torn PDF in a message does not stop its other attachments.
• Failed deliveries wait. If your SFTP server or endpoint is down, Paige retries, then marks the destination down with the error. Press Retry once it is fixed and everything that failed goes out.
• A record after the files are gone. When retention removes a document's files, its fields and history stay in the job.

Keep the controls that belong downstream

Reading a document correctly is not the same as approving it. Keep duplicate-invoice checks, PO matching and payment approval in your accounting process. The accounts payable automation guide walks through which controls to keep and where.

Common mistakes

• Teaching with perfect samples only

  Include faint scans, multi-page documents and odd senders, or the first bad week will surprise you.

• One job for everything

  Invoices, orders and statements need different fields. Give each stream its own job and address.

• Going fully automatic on day one

  Run on Human review until the held list is short and the fixes are rare.

• Field names nobody downstream recognizes

  Name fields the way your system and your import file name them.

• Forgetting the mailbox rule

  Make sure every way documents arrive, including a second inbox, forwards to the job.

• Never testing the destination

  Use Test on the SFTP folder or webhook before real documents depend on it.

Questions