Document Reconciliation

Only the exceptions reach a person

Invoices, tickets, statements and remittances are read by AI and matched against your bookings, orders or ledger. Clean matches post themselves. Your team sees the handful that need a decision, with the reason attached.

Exceptions · today

1,412 matched · 23 to review
DocumentRecordReasonDiff
INV-88213Booking 51022Amount outside tolerance+ 140.00
INV-88190No candidateReference not found—
TKT-20471Booking 50987Booking amended after issue− 62.50
INV-88102Booking 50811Possible duplicate of INV-880170.00

Illustrative. Rows, references and values are examples.

Selected exception

INV-88213

Invoice total
1,290.00
Booking value
1,150.00
Tolerance
± 2% or 25.00
Extracted from
page 1, lines 3–7

Line 6 is a late-booking surcharge not present on the booking. Likely a valid charge the booking was never updated with.

Resolve

Accept and update booking Query supplier Write off

A team paid to find the one in fifty

Most documents match. The work is in reading all of them to find the few that do not, and the cost of missing one is paid twice: once to the supplier, and again when the customer notices. Volume grows with the business; the team grows with the volume.

1 in 3

firms examined by the SRA were non-compliant on anti-money-laundering controls, with source-of-funds documents "often collected but not analysed". The same pattern, documents gathered and never checked, shows up wherever reconciliation is manual.

SRA AML report, year to April 2025

80%

of homecare providers with public contracts reported being paid late in a 225-provider survey. Proving what is owed starts with matching delivered care to invoices and remittances.

Homecare Association, 2023

From inbox to posted match

  1. 01

    Documents are collected where they land

    Supplier mailboxes, portals, SharePoint folders and scanned post. Suppliers keep sending what they send today.

  2. 02

    Every document is read

    Invoices, tickets, statements, remittance advices and credit notes are read by AI that has been shown your document types, so line items, references, dates and amounts come out as data you can match against.

  3. 03

    Each one is matched against your system of record

    Bookings, purchase orders, the ledger or the CRM, whichever holds the truth for that document. Matching rules are yours: exact references first, then amounts, dates and parties within the tolerances you set.

  4. 04

    Matches post automatically

    A clean match is recorded, attached to the record it belongs to, and written back to your finance or booking system without anyone touching it.

  5. 05

    Only exceptions reach a person

    Mismatched amounts, missing references, duplicates, amendments the supplier never received. Each one arrives in a queue with the document, the candidate matches and the reason it failed, ready to resolve in one click.

  6. See it on your documents

What the reviewer sees

The person who used to read everything now opens a short queue, with the answer most of the way there.

  • —The exception, the document and the best candidate matches side by side
  • —Why it failed: amount outside tolerance, reference not found, possible duplicate, booking amended after the invoice
  • —One action to accept a match, raise a query to the supplier, or write it off with a reason
  • —A dashboard of match rate, exception ageing and value at risk, by supplier and by week
  • —An audit trail of every automatic match and every human decision

Built for a finance team's standards

Runs inside your Microsoft tenant

Built with Microsoft Azure and AI services inside your own subscription. Documents and financial data are processed in the region you choose and never leave your organisation.

Nothing posts without a rule you agreed

Automatic matches follow written rules and tolerances. Anything outside them waits for a person. The rules are documented and can be tightened or loosened by you.

Fits the systems you already run

Booking platforms, finance systems, CRMs and purchase-order tools stay as they are. The pipeline reads from them and writes back through their interfaces or import formats.

Measured from day one

Match rate, exceptions per day and time to resolve are captured from the first week, so the before-and-after is on record.

Built first for a travel retailer

A cruise and package-holiday retailer processing around 500 bookings a day received 1,000 to 2,000 supplier documents daily: invoices, tickets and certificates, each checked against the booking by hand. The pipeline now extracts the document data with an AI model trained on travel documents, matches it to the booking system, flags discrepancies and amendment gaps, and gives the team a dashboard of exceptions.

The same pipeline, other documents

Travel and hospitality

Supplier invoices, tickets and certificates matched to bookings, including amendments made after the supplier was first told.

Charities

Donation platform payouts, finance postings and bank receipts reconciled three ways, net of platform fees.

Care and support providers

Delivered care matched to invoices and to council and NHS remittances, with disputes and aged debt by commissioner.

Finance teams

Ledger entries against the portfolio or booking system at period end, with exceptions surfaced before the close.

Law firms

Source-of-funds evidence checked against the declared source, with mismatches flagged for the fee earner.

How it is sold

Scoped on a sample of your real documents, built as one fixed-price project, kept running under Cover.

Build project

Fixed price, agreed before work starts

  • Extraction set up for your document types, tested on your backlog
  • Matching rules and tolerances agreed with finance
  • Integration with your booking, order or finance system
  • Exceptions queue and dashboard, go-live, 30-day support

Cover

Flat monthly fee, 12-month term

  • Monitoring and fixes
  • Extraction tuned as new supplier formats appear
  • Rule changes within the monthly allowance
  • Next-business-day response

Questions finance teams ask

What document formats does it read?
PDF, scanned and photographed pages, Word and Excel attachments, email bodies, and structured feeds where a supplier offers one. The extraction model is set up for your document types during the build.
How accurate is the matching?
That depends on your documents and your rules, which is why we measure it on your real backlog during testing and agree the tolerances before go-live. Anything the rules cannot match with confidence goes to a person rather than being guessed.
What happens to exceptions?
They go into a queue with the document, the candidate matches and the failure reason. Your team resolves them with one action each, and every decision is logged.
Does it replace our finance or booking system?
No. It sits beside them, reads from them and writes matches back. Your system of record stays the system of record.
Can it handle amendments?
Yes. Bookings or orders changed after a document was issued are a common cause of mismatches, and the pipeline flags them as amendments rather than errors, with the before and after shown to the reviewer.
How is it scoped?
Through the Process & Reporting Review: we take a sample of your documents, measure the current volume and effort, prototype the extraction on real samples, and agree a fixed price before work starts.

Bring a week of documents.

We will show the extraction on your own samples, what the exceptions queue would contain, and give you a fixed price.

Book a readiness assessment