---
title: "What Does It Cost to Automate Document & Application Processing? 2026 | Croissant & Baguette"
description: "An OCR template is A$89 a month and copies without checking. Cloud document AI reads pages for US$1.50 per 1,000. A system that reads, checks and files applications against your rules is $5,000 to $15,000 in Australia. Five options priced, one real system itemised."
url: "https://croissantbaguette.com.au/guides/automating-document-and-application-processing/"
---

# What does it cost to automate document and application processing?

https://croissantbaguette.com.au/guides/automating-document-and-application-processing/

An OCR template is A$89 a month and copies without checking. Cloud document AI reads pages for US$1.50 per 1,000. A system that reads, checks and files applications against your rules is $5,000 to $15,000 in Australia. Five options priced, one real system itemised.

## What does it cost to automate document and application processing?

Between A$89 a month and $50,000, depending on how much of the job you hand over. A template copies fields off one layout and checks nothing. Cloud document AI reads any page for a fraction of a cent and stops there. A system built around your rules receives, reads, checks, files and chases, and hands the doubtful ones to a person. The five options priced, what breaks in each, what the Privacy Act asks, and one real docket.

**Quick answer**

Every figure on this page excludes GST. Tool prices read on 26 September 2026.

- **Template** A$89–199 a month
- **RPA** A$22–225 a month
- **Cloud AI** US$1.50–50 per 1,000 pages
- **Platform** from US$18,000 a year
- **Built system** $5,000–15,000
- **To run** $20–200 a month

## $5,000 to $15,000 for a system that checks. Less for one that only copies.

In Australia in 2026, $5,000 to $15,000 buys a system that takes each document or application as it arrives, reads it whatever the layout, checks it against your rules and records, files the fields into the systems you run, chases what is missing and queues the doubtful ones for a person; running it is $20 to $200 a month of model and API usage, and approvals, an audit trail or a system with no API take it to $15,000 to $50,000. Everything cheaper does part of the job. An OCR template, A$89 to $199 a month, copies fields off one layout and checks nothing. A software robot, A$22 to $225 a month in licences plus a bot built per process, types what a person would and stops when the screen changes. Cloud document AI reads any page for US$1.50 to $50 per 1,000 pages and does nothing with what it read. A platform, from US$18,000 a year, adds the inbox and a correction screen, not your rules. What decides is what checking costs by hand: ten to twenty minutes an application at a loaded $65 to $75 an hour is $11 to $25 each, two hundred a month is $2,200 to $5,000, and a build at our $8,000 floor pays back in two to four months. Under fifty a month, a template with a person behind it is the honest answer.

## Five ways to stop keying documents by hand.

Cheapest first, in the order people usually try them. Tool prices as published on 26 September 2026, in each vendor's currency; the tools are the ones the AI engines name when asked this question. What we charge is further down.

Tool prices as published on each vendor's price page on 26 September 2026, excluding GST: Parseur and Power Automate in Australian dollars, the rest in US dollars. Build figures exclude GST and model usage. Australian automation price guides published in 2026 put a configured document-processing system at $2,500 to $20,000 and custom development above $35,000; we start at $8,000 a process and most builds land between $10,000 and $30,000.

- **An OCR template** A$0–199 a month You mark the fields on a sample and the tool reads the same spots on every document after it, then posts them to a spreadsheet or an app. Parseur is free for 20 pages a month, then A$89 for 100, A$199 for 1,000 and A$799 for 10,000; Docparser is US$39 a month for 100 documents of up to five pages, US$74 for 250, US$159 for 1,000. — Any document that does not match the sample: a new sender, a second page, a crooked scan. Nothing is checked, only copied, so a wrong number in the right box goes straight through.
- **A software robot, or RPA** A$22–225 a month A bot that clicks through the screens a person would: open the email, save the attachment, type the fields into the portal. Power Automate is A$22.40 a user a month for attended flows and A$224.50 a bot a month unattended, paid yearly; UiPath is free for individuals and from US$25 a month for its Basic cloud plan, higher tiers quoted. The licence is the small part; somebody builds and maintains every bot. — The screen it was recorded on. A moved button, a renamed field or a portal update stops it until someone re-records it, and it reads a scan no better than the OCR under it.
- **Cloud document AI** US$1.50–50 per 1,000 pages Reads any layout and returns the text, fields, tables and checkboxes, each with a confidence score. Google Document AI is US$1.50 per 1,000 pages for OCR after the first 1,000 a month, US$30 per 1,000 for its form parser and custom extractors, US$10 for layout parsing; Amazon Textract is US$1.50 per 1,000 pages for text, US$50 for forms, US$15 for tables and queries, the same rates in Sydney as in the US. Document AI's OCR, form parser and custom extractor run in Sydney too, with limited support. — Nothing, because alone it only reads. The receiving, the checking against your rules, the filing into your systems and the chasing are a build, the band we sit in.
- **A document AI platform** from US$18,000 a year An inbox, extraction, a screen where a person corrects low-confidence fields, and exports to accounting and ERP systems, by subscription. Rossum's Starter plan is from US$18,000 a year on a one-year minimum, unlimited seats, higher tiers quoted; Nanonets meters each step at US$0.02 to $0.30 a run, with a US$100 a month starter plan after US$50 of free credits; Docsumo quotes on volume after a 14-day trial of up to 1,000 pages. Pages are processed where the vendor hosts, Frankfurt or Virginia for Rossum; Australian residency is an enterprise-tier question. — The rules that are yours. A platform validates what its fields say; whether an application meets your criteria, matches a record in your CRM and reaches the right person is configuration at best, a second system at worst.
- **A system built around your rules and your systems** $5,000–15,000 once, then $20–200 Takes the document from email, a web form or an upload, reads it with a cloud service above or a model, checks it against your rules and your own records, files the fields into your systems, asks the sender for what is missing, and queues the exceptions for a person with the failing rule and a drafted reply attached. $15,000 to $50,000 once approvals, an audit trail, a regulated environment or a system with no API come in. — Little, by design: a low-confidence read or a failed rule waits for a person instead of going through. It does need watching as the models and APIs beneath it move, which is the managed plan's job.

## What a document-processing system has to get right.

- **Receive from everywhere.** Email attachments, web forms, portal uploads, phone photos. One intake for every channel and format, or the pile that arrived the other way keeps growing.
- **Read with a confidence score.** OCR or a model reads each field and says how sure it is. Anything under the bar waits for a person, with the page beside it.
- **Check, not just copy.** Completeness, eligibility, and cross-checks against your CRM, the register or the price list. The right number in the wrong box, and the wrong applicant, are caught here.
- **Put a person where it matters.** A queue of exceptions showing the document, the fields, the failed rule and a drafted reply. One click to decide, and the decision logged with a name.
- **Choose where the pages go.** Textract and Document AI run in Sydney; most platforms process in the US or Frankfurt. Under APP 8 an overseas recipient of personal information is your responsibility, so decide the region on purpose and write it down.
- **Say so in the privacy policy.** From 10 December 2026 a privacy policy must name the personal information a computer program uses for decisions that significantly affect a person, and the kinds of decisions. Checking applications is one; the wording ships with the build.
- **Keep only what you need.** APP 11 asks you to protect the personal information you hold and to destroy or de-identify it once it is no longer needed: a retention rule, not a folder that grows for ever.

## One real system, line by line.

Edtripper lists 958 venues for school excursions, and every enquiry a teacher sent through a listing used to be an email somebody had to open, judge and forward. Now each one is read as it arrives and sorted, real or spam; a real enquiry goes straight to the attraction, and when that attraction is not yet a partner the system checks Edtripper's details against the venue's own website and writes the outreach itself. The second job runs the other way: a venue uploads its education pack and a whole programme's questions are drafted from it for an educator to approve.

Published with Edtripper's permission. Excludes GST.

$5,000 sits under our $8,000 floor because the platform already existed: the accounts, the data model, the queues and the deployment were built, and the second job drafts from content the venue uploads anyway. Wired into another business's systems from outside, the same pair is a $10,000 to $15,000 job. Two more of ours without a public price: TTMF, where a trademark application is uploaded, paid for and checked by AI around the clock, and Joris, where a property report is generated, matched to the client's record and emailed from one sentence.

- **Client** Founder-led education business, Sydney
- **Scope** Enquiry triage and outreach, AI programme drafting
- **Built in** 6 days
- **Part of** a $39,000 app and dashboard build

- **AI and automation** $5,000 6 days, on top of a platform we were already building for them

**Costs to run**

The automation itself costs nothing to keep online. The only variable line is model usage.

- **Model usage** PAYG pay as you go, metered per venue
- **Hosting and data** $0 Cloudflare Workers, D1, R2, Queues. Their own account
- **Email** $0 Resend, on our business account at no charge to them

## Who does this in Australia, and what to ask them.

- **Twenty of your worst documents, before the quote.** Real ones, scans and odd layouts included, and ask for the read rate on each field. A demo on a clean PDF proves nothing about your inbox.
- **What happens to a low-confidence field.** If it goes through, the system copies rather than checks. It should wait for a person, with the page shown and the doubtful value highlighted.
- **Where the pages are processed, in writing.** Region, provider, and whether the vendor trains on your documents. Under APP 8 an overseas disclosure is yours to answer for, so get the answer before you sign.
- **One of your rules, running in the demo.** Pick an eligibility rule and ask to see a real application fail it, and what the person then sees. Reading fields is table stakes; applying your rules is the job.
- **The price at your volume, and at double.** Per-page fees, per-run charges, licences and model usage all scale differently. Ask for the monthly figure at what you process now and at twice it.
- **We do.** Croissant & Baguette is a small Sydney studio with a network of Australian specialists. Built in Sydney, never offshored. Edtripper's enquiry triage, TTMF's application checking and Joris's report generation are ours, each running in the client's own accounts with a person holding the final approval, and every build lists in writing which providers see what. — Two weeks of assessment, at a fixed price, tells you which document streams pay back and which are cheaper left with a person.

## What we charge.

Indicative. Fixed price after the assessment.

**Why we read your documents before we quote** Because the documents set the price, not the software. Twenty real samples and the rules you apply to them tell us the read rate, the exception rate and how many systems the fields must reach; the assessment turns that into a ranked list with build prices attached, including the streams that should stay with a person. Start a build within 90 days and the $1,500 comes off it.

What moves a document system within those ranges:

- **Automation Opportunity Assessment** $1,500 fixed (2 weeks · credited to your build)
- **Automation build** from $8,000 (one process end to end, 30 days of hypercare)
- **Managed** from $250/month (monitoring, fixes, monthly report · changes quoted)
- **Internal tool / dashboard** typically $8–20k (on top of the systems doing the work)

- how many layouts arrive. One form is days; scans from fifty senders is weeks
- how many rules, and how many of them need your own records to check
- what it writes to. A clean API, a portal with none, or a spreadsheet
- whether a person approves, and what they need to see to decide quickly
- audit and retention. Who read what, and when it gets deleted
- volume. Fifty a month and five thousand a day are different systems

## Ask away.

Seven things people ask before they stop keying. Still one missing? Ask us.

- **What does it cost to automate document and application processing?** $5,000 to $15,000 to build, for a system that takes each document, reads it whatever the layout, checks it against your rules and records, files the fields into your systems and queues the exceptions for a person, plus $20 to $200 a month to run. Approvals, an audit trail or a system with no API take it to $15,000 to $50,000. Edtripper's two jobs, itemised above, were $5,000 in six days on a platform we were already building.
- **What are other options besides manual data entry for processing applications?** Four, in order of how much of the job they take. An OCR template such as Parseur or Docparser, A$89 to $199 a month, copies fields off a fixed layout. A software robot on Power Automate or UiPath, A$22 to $225 a month plus the bot, types what a person would. Cloud document AI from Google or Amazon reads any page for US$1.50 to $50 per 1,000 pages. A built system does all of that and applies your rules, for $5,000 to $15,000.
- **What's the most recommended way to automate document and application checking?** Read with document AI, check with rules and a model, and keep a person on the exceptions. The engines usually recommend the platforms, Rossum, Nanonets and Docsumo, which suit high-volume invoices with an accounting system on the other end. For applications checked against criteria that are yours, we recommend, and build, a system around those rules and your own records, with every doubtful case waiting for a person.
- **I keep manually checking applications against our rules, how can I automate that?** Write the rules down first; most live in somebody's head. Then a system reads each application as it arrives, runs the rules, cross-checks the answers against your CRM or register, and sorts the result three ways: passes filed, incomplete ones chased with a drafted request, failures and low-confidence reads queued for you with the reason attached. You review a queue instead of a pile. A $5,000 to $15,000 build, and the rule-writing is what the two-week assessment does.
- **Do we still need a person to check them?** Yes, on the exceptions, by design rather than hope. A confidence score on every field and a pass or fail on every rule decide what goes through untouched; Australian accounts-payable guides published in 2026 report 70 to 85 percent of invoices going straight through after a few months of tuning, the rest queued. The person decides in one click, with the page, the fields and the failing rule in front of them, and the decision is logged.
- **Where do our documents go, and is that allowed under the Privacy Act?** Wherever the reading happens, so choose it. Amazon Textract and Google Document AI both run in Sydney; most subscription platforms process in the United States or Europe, with Australian hosting on enterprise tiers if at all. APP 8 allows personal information overseas when you take reasonable steps to ensure the recipient handles it as the APPs require, and you stay accountable for it. We default to Australian-region processing and list the providers in writing.
- **Can it read scans, photos and handwriting?** Printed scans and phone photos, yes: cloud document AI reads them at any angle and returns a confidence score per field, which decides whether a value goes through or waits for a person. Handwriting is read but trusted less, so a handwritten form should route more fields to review. The honest test is your own worst twenty documents, run before anyone quotes, which is why we ask for them.

- **AI automation costs.** What it costs to take one process off your team, with a real system itemised.
- **Enquiries and bookings.** Booking tool, answering service, AI receptionist or a built system, priced side by side.
- **AI Automation.** The five jobs we automate, and how the work is done.
- **Systems in sync.** One home per fact, moved by webhook, from a $0 connector to a built integration.

## Book an assessment.

Two weeks, fixed price. Send twenty of your worst documents and the rules you check them against; you get the read rate, the exception rate and a ranked list of what to automate first, priced. Credited against your build.