ContactBook a call

Industries/ Professional services/ Document reading and filing

Documents read and filed on arrival, with doubtful ones sent to a person

Purchase invoices, receipts, bank statements, signed engagement letters and identity documents are read as they land, matched to the right client, job or ledger, and filed. Your team sees only the documents the software is unsure about, with the original beside the details it extracted.

A paralegal, a young Black man in a light grey shirt, at a desk feeding documents into a scanner, a monitor showing files sorting themselves into folders.
Used by
Bookkeepers, client accountants and onboarding staff
Takes the place of
Keying invoices, renaming scans and filing by hand
Connects to
Your accounting software, practice system and document store
Built with
Our document intelligence service

Where it hurts today

A normal week, before it exists

In a bookkeeping or accounts team, a large part of the week goes on opening documents. A client forwards a month of supplier invoices on the last day of the quarter. Another sends a folder of receipt photos. Each one is opened, read, coded to a nominal account, checked for VAT and keyed into the client's ledger, one at a time, by someone who trained to give advice.

Bank statements arrive as PDFs whenever a bank feed breaks, and someone types the lines into a spreadsheet to import them. Signed engagement letters come back by email and need saving to the right client. Identity documents for new clients arrive as phone photos, and someone renames them, files them and copies the expiry date into the onboarding record by hand.

None of it is difficult, which is why it never gets fixed. It is also where mistakes creep in: a transposed digit in a VAT figure, an invoice keyed twice because it came by email and by post, a passport saved under the wrong client. Those mistakes surface weeks later, at review or at the VAT return, when they take longest to unpick.

What it does

Document reading and filing, feature by feature

What happens to each document, from the moment it lands to the moment it is filed.

  1. 01

    Collects from every inbox and upload

    A shared invoices mailbox, the client portal, watched folders and the office scanner all feed one queue. Multi-page scans are split into separate documents, rotated pages are straightened, and each document is labelled by type, such as purchase invoice, receipt, statement or identity document, before anything is read.

  2. 02

    Invoice and receipt data extraction

    For purchase invoices and receipts: supplier, date, invoice number, net, VAT and total, down to line level where you want it. For bank statements, every line with its date, description and amount, with the opening and closing balances checked against the lines in between.

  3. 03

    Matched to the right client and ledger

    Each document is matched to a client by mailbox, sender, VAT number or account details, then to the supplier and nominal code that client usually uses. Codes learned from your team's past choices are suggested, and never applied silently where you have asked for review.

  4. 04

    Checks before anything is posted

    Totals must add up, the supplier must be known or flagged, and the invoice number must not already exist in the ledger. Duplicates sent by both email and post are caught here, along with VAT figures that do not match the rate printed on the page.

  5. 05

    A review queue built for speed

    Doubtful documents appear with the original on one side and the extracted fields on the other, with the uncertain field highlighted where it was found on the page. The reviewer confirms or corrects that one field and moves on, rather than retyping the whole document.

  6. 06

    Engagement letters and ID documents filed

    Signed engagement letters are matched to the client and filed with the date signed. Passports, driving licences and proof of address go to the onboarding record with names and expiry dates captured, so your team can see at a glance what is held for each client and what is missing.

A day with it

How the day runs instead

A normal day in a bookkeeping and accounts team, with documents read as they arrive.

  1. 7.30am
    Bookkeeper

    Before the team arrives, the night's documents have been read. The bookkeeper opens the review queue and finds routine invoices from regular suppliers already waiting in each client's ledger as draft bills. Only the doubtful ones are left for her to look at.

  2. 9.15am
    Bookkeeper

    A receipt photographed in a car park has a blurred VAT figure. She sees the photo beside the extracted fields, zooms in, confirms the amount and moves on. An invoice from a new supplier waits for a nominal code, which she chooses once and the system suggests next time.

  3. 11am
    Client accountant

    A client sends a bank statement as a PDF because their bank feed broke. It is read line by line, the balances check out, and the lines are ready to import. The accountant spots one flagged line, a transfer that matches no invoice, and sends the client a query.

  4. 2pm
    Onboarding administrator

    A new client's passport photo and utility bill arrive through the portal. Both are filed to the onboarding record with names, dates and expiry captured. The administrator compares the details with the application form, then passes the file to the person who signs off identity checks.

  5. Month end
    Practice manager

    The manager looks at the month: how many documents came in, how many went straight through, which fields were corrected most often and which suppliers keep causing trouble. A supplier whose invoices are always queried gets a call from the client asking for a cleaner format.

What it connects to

Works with what you already run

Documents come in from the places they already land and go out to the systems your team already works in.

  • Accounting software

    Draft bills, receipts and bank lines are created in each client's ledger through the accounting package's own interface, for example Xero, QuickBooks or Sage where it offers an API, with the original file attached to the entry it created.

  • Shared mailboxes and the client portal

    Documents arrive from email inboxes, a client portal, watched folders and scanners. Each arrival is logged with where it came from and when, so nobody has to ask whether a client really sent something or when it reached the firm.

  • Practice management system

    Client records, engagement types and who looks after each client come from your practice management system, so documents are matched to the right job and the exceptions reach the person responsible for that client rather than a general queue.

  • Document store

    Every original is filed to the client's folder in your document management system, SharePoint or Google Drive, named by type, date and period, and linked from the ledger entry or onboarding record it created.

The first version

Start small, then grow it

A first version should be useful on day one and small enough to get right. This is where we would usually draw the line, and what tends to follow once people are using it.

In the first version

  • One document type, usually purchase invoices, read and checked for a group of clients
  • Collection from a shared mailbox and the client portal into one queue
  • Duplicate, arithmetic and known-supplier checks before anything is posted
  • A review queue showing the original beside the extracted fields
  • Draft entries created in your accounting software with the original attached

Once it is in use

  • Bank statements read line by line for clients whose feeds fail
  • Engagement letters and identity documents filed to the onboarding record
  • Nominal code suggestions learned from your team's past corrections
  • Monthly reports on volumes, corrections and the suppliers that cause the most queries

Buyer's guide

Before you commission one

Four decisions worth making before you automate any document type.

  1. 01

    Built-in capture or a custom pipeline?

    Cloud accounting packages and add-on capture products often read receipts and supplier invoices for a single business well, and for a firm whose clients all use one package that is usually the cheaper answer. A custom pipeline makes sense when documents arrive by many routes, when you want one review queue across every client, or when the documents include statements, engagement letters and identity papers those tools do not handle.

  2. 02

    What to gather first

    A sample of your real documents, a hundred or so per type, including the bad scans and the supplier everyone complains about. Your chart of accounts and the coding rules for your largest clients. A note of which fields must be perfect, such as bank details and VAT, and which can tolerate an occasional correction. And the routes documents arrive by today.

  3. 03

    Who should own it

    A senior bookkeeper or the head of the accounts team owns the rules: which suppliers post automatically, which fields always go to review, and when a threshold changes. Your money laundering reporting officer owns anything to do with identity documents. A partner signs off what is allowed to post without a person, and revisits that decision every few months.

  4. 04

    How you will know it is working

    Watch how many documents go straight through and how many need a person, which fields are corrected most often, and whether corrections fall over time. Check whether duplicates and VAT errors are caught before review rather than at it. And listen to the team: if they still retype documents from the queue, the review screen needs changing.

Terms worth knowing

The words you will meet

OCR and document AI
Optical character recognition turns an image of text into text, but on its own it cannot tell which number is the total. Document AI, often sold as intelligent document processing, reads layout and tables too, so it can find the supplier, date and VAT on layouts it has never seen. You need both, plus checks against your own records.
Confidence threshold
Each field the software reads comes with a score for how sure it is. The threshold is the score above which a field goes straight through and below which a person checks it. Setting thresholds per field, strict for bank details and looser for a description, keeps automation safe without slowing everything down.
Peppol and e-invoicing
Peppol is a network for exchanging invoices as structured data instead of PDFs. Germany has required businesses to accept e-invoices since 1 January 2025, Belgium since 1 January 2026 and France since 1 September 2026, and the UK plans a mandate from April 2029. Structured invoices are imported rather than read, so the pipeline handles both.

Handled with care

The details that matter here

Documents carry bank details, tax figures and identity information, so these are the details we agree with your partners, compliance lead and data protection officer first.

  • Scored on your own documents first

    Before anything posts automatically, we run the pipeline against a sample of your real documents, including the bad scans, and show you how each field performed. You decide which fields can go straight through and which always go to a person, and you can change that at any time.

  • Where documents are processed

    We tell you which AI provider or reading service each document passes through and which region it is processed in, and show you what its terms say about using your client material for training. Your partners choose, and we write the data flow down for your next client security questionnaire.

  • Identity documents kept apart

    Passports and proof of address are stored apart from bookkeeping files and visible only to onboarding staff. In the UK, regulation 40 of the Money Laundering Regulations 2017 generally requires them to be kept for five years after the relationship ends. Your money laundering reporting officer approves retention and deletion. The software files; a person decides whether a check passes.

  • A trail from figure to source

    Every entry the system creates keeps a link to the original file, the values extracted, the confidence of each field and who confirmed it. When a reviewer or a client queries a figure in the ledger, the source is one click away rather than in someone's memory.

Guides

Worth reading first

Questions

What buyers ask first

Can AI post invoices straight into Xero or QuickBooks?

Only where you have agreed it can. We usually start with everything going through review, then allow routine invoices from known suppliers to create draft bills once the checks have proved themselves on live documents. Anything with a new supplier, a duplicate number or figures that do not add up always waits for a person, and any field can be pulled back to manual review whenever you like.

How accurate is OCR on receipts and bad phone photos?

It depends far more on the documents than on the software. Clean PDFs in familiar layouts read well. Creased, shadowed or handwritten ones read less well and land in the review queue more often. That is why we score a sample of your real documents before proposing a design. Where uploads come through a portal, an unreadable photo can be sent straight back while the client still has the receipt in their hand.

Do we still need document processing if we have bank feeds?

For many clients a bank feed is enough, and we would not read statements where a feed already works. Statements matter when feeds break, when a client banks somewhere without one, or for historic periods at the start of an engagement. Invoices and receipts are a different job, because a feed tells you money moved but not what it was for or how much VAT it carried.

Does it replace the receipt capture in our accounting software?

Not necessarily, and if the built-in capture suits your clients we will say so on the first call. Custom reading earns its place when documents arrive by many routes, when you want one review queue across every client and package, or when the documents are ones those tools do not handle, such as engagement letters, bank statements and identity papers.

Is it safe to send client documents to an AI model?

It can be, with the right choices made first. We tell you which AI provider or reading service each document passes through, where it is processed and what its terms say about using your client material for training, and your partners decide. We write the data flow down so you can answer the next client security questionnaire with it, and confidentiality terms are signed before we see a single real document.

Send us a month of the documents you key in

Book a free first call and tell us which documents your team types up most. We reply within 48 hours with an honest view of which ones are worth automating.

More for professional services

What else we build for professional services