AI invoice data extraction, field by field, and where it stops
Invoices arrive as they always have. We build the extraction in a business day. It costs nothing until it runs.
AI invoice data extraction reads the supplier, the invoice number, the dates, the currency, the totals and the tax lines off an invoice without anybody typing them. The hard part is the line items, where a table can wrap a page break or carry no header. Fabrx builds the extraction free and charges per document.
Fabrx builds the API. Landing it in a given system is done by partners, so the only requirement is a system that accepts an API call.
What moves, and what it becomes
The supplier
A trading name at the top, a legal name in the footer and a bank detail block that sometimes belongs to a factoring company. Which one your ledger keys on is the question.
A matched supplier record
The one that looks easy
The invoice number
Rarely labelled the same way twice, and sitting next to a purchase order number, a delivery note number and an account number that all look like it.
The reference your ledger deduplicates on
The duplicate guard
The dates
Issue, due, supply and period, with two of them usually missing and the rest written in whichever order the sender grew up with.
Due date and posting period
The one that moves money
Currency and totals
Net, tax and gross, where the currency is a symbol in one place and a code in another, and the gross is the only figure anybody checks.
The amount payable
The arithmetic
Tax and VAT lines
Several rates on one invoice, a reverse charge note in the small print, and a registration number that decides how the whole document is treated.
Tax codes per line
The compliance one
Line items
The table itself wraps across a page break, repeats its header halfway down, carries a discount row that is not a line, and sometimes has no header at all.
Rows, each with its own quantity, price and tax
The hard one
What happens to one document
Your documents
An invoice arrives. Your documents
An AP inbox, a supplier portal, a shared drive, or a photograph taken on a site visit.
T+0
Fabrx
Identify what it is. Fabrx
An invoice, a credit note, a statement and a remittance advice often share a template. They are told apart before anything is read.
If A document we cannot identify
Hold for review. Fabrx. Needs a person.
A named person on your side sees the invoice, the field in question and what the extraction read, before anything reaches the ledger.
needs a person
Then it rejoins the path above.
Read the fields. Fabrx
Header fields and the line-item table. Fields are read by what they are, wherever they sit, so a supplier redesigning their template needs no rebuild.
If Lines that do not sum to the total
Hold for review. Fabrx. Needs a person.
A named person on your side sees the invoice, the field in question and what the extraction read, before anything reaches the ledger.
needs a person
Then it rejoins the path above.
Send the payload. Fabrx
A structured record, the same shape every time, with the line items as rows and not as a block of text.
Where we hand over
We return the fields. The match against your purchase orders and goods received is yours, because it needs records that live in your system and never appear on the invoice.
Your accounting system
A draft bill. Your accounting system
Your accounting system, your approval chain, your coding rules. Nobody logs into anything of ours.
What we do, and what stays yours
Fabrx builds and runs
- Reading the invoice, whatever shape it arrives in
- Telling an invoice apart from a credit note or a statement before reading it
- Pulling the header fields and the line items as structured rows
- Checking what the invoice can confirm on its own, such as whether the lines sum
- Flagging anything that needs a person, with the invoice attached
You keep
- The accounting system, its chart of accounts and its approval chain
- Coding each line, which is a decision about your business
- The three-way match against your purchase orders and receipts
- Whether to pay, and when
- Somebody to look at the output and tell us whether it is right
When a document is wrong
A pipeline with no failure path is a diagram of a good day. These are the ones that happen.
The scan is unreadable
The invoice stops before extraction and is never guessed at.
Who sees it. Whoever sent it, by reply, with the page that failed.
- Rescanned and reprocessed
- Keyed in by hand, as it is today
A credit note arrives in the invoice template
It is identified as a credit note at the classification step, so the amount does not reach the ledger with the wrong sign.
Who sees it. Your AP team, in the exception report.
- Posted as a credit
- Returned to the supplier
The line items do not sum to the total
The invoice is held with both figures shown. A rounding difference and a missing line look identical to arithmetic, and only a person can say which.
Who sees it. Your AP team, with the invoice open beside the numbers.
- Released once checked
- Queried with the supplier
A field you need is genuinely not on the invoice
The record arrives with that field empty and marked. Nothing is invented to fill it.
Who sees it. The person who owns the bill in your accounting system.
- Filled from the purchase order
- Chased with the supplier
AI invoice data extraction, four ways to get it
The same document, four approaches, and what each one asks of you.
What you need before you start
Build it yourself
A developer, a cloud AI account, and a sample set to test against.
A vendor platform you configure
A licence, and somebody to configure it for your senders.
A consultancy project
A scope, and a signed statement of work.
Fabrx
A 30-minute call and about twenty of your invoices.
How long to the first real output
Build it yourself
A prototype in an afternoon. The awkward senders take longer.
A vendor platform you configure
As long as the configuration takes.
A consultancy project
The length of the engagement.
Fabrx
Usually one business day from receiving your samples.
What happens to the line items
Build it yourself
The part most prototypes leave until last.
A vendor platform you configure
Depends what the configuration was told to expect.
A consultancy project
Whatever the scope said.
Fabrx
Returned as rows, each with its own quantity, price and tax.
Who maintains it when a supplier changes their template
Build it yourself
You do.
A vendor platform you configure
You do, in their builder.
A consultancy project
A change request, if the engagement is still open.
Fabrx
Nobody needs to. Fields are read by what they are, so a redesigned template is ordinary input.
What it costs to find out whether it reads your invoices
Build it yourself
Your own time, before you know whether it holds up.
A vendor platform you configure
A licence, before you know it reads yours.
A consultancy project
A project fee, agreed before anyone has seen your invoices.
Fabrx
Nothing. The build is free and you pay per document once it runs.
Someone has run this
They built solutions for any situation encountered along the way. For us, this type of partnership matters.
- 300+
- Field agents using it every working day
- 4,000+
- Documents processed every month
- 4 yrs
- In production, on the same workflow
Blitz is a real estate field agency network running identity documents into signed contracts, not invoices. The figures are theirs and describe the pipeline on this page, not an accounts payable deployment.
What a pilot measures
Measured on your documents
Header field accuracy
Supplier, number, dates and totals, scored against your correction.
Line-item accuracy
Scored separately from the header fields, because an invoice can have every header right and the table wrong.
Sample size
How many of your real invoices we ran. You choose which senders.
Exception rate
The share that stopped for a person, and what stopped them.
Build time
Calendar time from your samples reaching us to the pipeline running.
What we ask you for
How long does one invoice take today, from arriving to being posted?
It is the only honest denominator for any claim about time saved.
How many invoices do you process in a month?
It sets the pricing band, because the rate depends on volume.
How many suppliers send them, and do the top ten cover most of it?
A long tail of small senders is a different build from twenty regulars, and the sample we ask for depends on the answer.
Do you need the line items, or only the header and the total?
Header fields alone and a full line-item table are two different builds, and briefs often leave the question out.
What it costs
- 01
We build it
We build the extraction, usually within one business day of receiving your sample documents.
- 02
You check it
It costs you nothing until it runs on your real documents. If the accuracy is not good enough, nothing goes into production and nothing is charged.
- 03
Then you pay
A prepaid package priced per document, valid for a calendar year. The rate depends on the document type and your monthly volume.
- No subscription
- No per-seat charge
- No setup fee at any volume
Questions about AI invoice data extraction
What is AI invoice data extraction?
Reading the values off an invoice and handing them over as structured data, without a person typing them in. The header fields are the supplier, the invoice number, the dates, the currency, the totals and the tax lines. The line items are the table underneath.
Can AI extract data from an invoice PDF?
Yes, and a PDF that was generated rather than scanned is the easier case because the text is already in the file. The harder part is that a PDF carries layout and no structure, so a table can be a set of positioned words that only look like columns.
Can it read line items, or only the header fields?
Line items come back as rows, each with its own quantity, price and tax. It is worth asking any vendor this directly. A table that wraps a page break, repeats its header halfway down, or carries a discount row that is not a line is where invoice extractions differ most from each other.
How accurate is AI invoice data extraction?
Ask for the number on your own invoices, and ask whether it covers the line items or only the header fields. The two are usually very different and a single published figure rarely says which it describes. We measure both separately on your sample before anything goes into production.
What about a scanned or photographed invoice?
Both are ordinary input, including the creased, skewed and badly lit ones. Where it stops is resolution low enough that a person would struggle too, and those come back as unreadable instead of guessed at.
Can I build this myself with a no-code tool?
For a handful of senders with steady templates, often yes, and a prototype can take an afternoon. What takes the time is the long tail: the credit note in the invoice template, the table with no header, and the supplier who redesigns theirs in March. That maintenance is the part worth costing before you start.
Related
- invoice processing
The same documents described by the process around them, and where the hours actually go.
- automatic data extraction
What changes when the thing you are extracting from is a file instead of a page.
- how the build works
The call, the sample files, and what happens in the business day between them.
Send us twenty invoices
We build the extraction and show you what it got right, field by field, before you pay anything.