Airlines have been promised that digitization solves their invoicing problem. Scan the invoice, convert it to text, and the hard work is done. But if that were true, the industry wouldn’t be paying an ‘invisible surcharge’ of around $4.6 billion a year in undetected invoicing errors.
Understanding why starts with knowing what OCR actually does, and what it doesn’t do.
What OCR does — and where it stops
Optical Character Recognition (OCR) does one thing: it converts an image of text into machine-readable characters. Feed it a scanned PDF of an invoice and it will produce a string of text. The number “2.71” will appear. The words “BP Air” will appear. The tail number “G-EZBA” will appear. What OCR cannot do is tell you whether any of it is right.
It doesn’t know that 2.71 USD/USG should be 2.68 under your current contract. It doesn’t know that G-EZBA was at Heathrow on that date. It doesn’t know that BP Air’s rate for Jet A-1 at that station follows a different pricing convention to the same supplier at a regional airport in Southeast Asia. It reads characters. That’s it.
The output lands on a clerk’s desk. A human still has to interpret it, check it, and decide whether to approve it. OCR has changed the format but it hasn’t automated the process of verification.
AI Invoice Automation
Reads and verifies every line, of every invoice, for every flight. Download the one-page factsheet:
The errors that get through
This matters because of the volume and complexity of airline invoicing. Our analysis of 31 airlines’ fuel and airport invoices received and processed through the Skymetrix platform in 2024–2025, found that the average number of invoices processed per year was 36,000, and that an average of 40% arrived as unstructured PDFs/paper. To add to the workload, these invoices could be pages long.
Layer that against established benchmarks:
- The Institute of Finance and Management reports that 39% of all invoices contain errors: wrong prices, incorrect quantities, unexpected surcharges.
- Across 25 years of work with airlines, Skymetrix consistently observes that:
- Accounts Payable teams verify only 10–20% of invoices at line-item level. The rest are approved on header total alone.
- Overspend due to invoicing errors equate to 1%+ of direct operating costs. Applied conservatively to the industry’s $455 billion direct operating cost base (IATA, EUROCONTROL), the resulting “invisible surcharge” reaches $4.6 billion a year — roughly an eighth of total industry net profit.
OCR, on its own, does little to close that gap. It makes the invoice readable. It doesn’t make it right.
The four steps that actually solve the problem
Our AI Invoice Automation solution is built on the understanding that reading an invoice and verifying an invoice are entirely different problems. The system runs four stages end-to-end:
Step 1 — Read (OCR)
Every invoice is ingested regardless of format: paper, scanned PDF, email attachment, EDI or XML. For unstructured documents, OCR converts the image to text. For structured EDI and XML, this step is skipped — but critically, the remaining three steps are not. Because while EDI ensures structure, it can’t guarantee truth.
Step 2 — Interpret (semantic model)
The system doesn’t just read the text, it understands it. Aviation-specific entities are recognised and classified: tail numbers, flight numbers, fuel volumes, charge types, station codes, supplier identifiers.
This is semantic understanding, not character matching. A generic OCR tool sees “3,420 USG” as a string. Our solution knows it’s a fuel uplift volume that needs to be reconciled against an operational record.
Step 3 — Normalize
Every entity is matched against live operational data via API connections to flight records, uplift data, and tariff schedules:
- An invoice referencing flight U28344 is matched to the actual departure record.
- A volume figure is cross-referenced against the uplifted quantity logged at the station.
- Unit rates are mapped to the correct pricing index for that supplier, station, and date.
- The invoice is structured correctly before any validation takes place.
Step 4 — Validate
Every field is checked against agreed contract rates.
- Wrong price? Flagged.
- Duplicate charge? Flagged.
- Quantity discrepancy? Flagged.
- Rate that has drifted slowly 0.3% over six months — the kind of subtle change that manual spot-checks never catch? Flagged.
Invoices that pass validation route directly to the journal with human oversight. And genuine exceptions are surfaced for review before payment, not discovered months later in audit.
OCR vs AI Invoice Automation: At a glance
| Plain OCR / generic invoice tool | Skymetrix AI Invoice Automation |
|---|---|
| Reads characters from a page | ✓ Reads characters and understands meaning |
| Treats every number as a number | ✓ Recognises tail numbers, flight numbers, charge types |
| No knowledge of contracts or flights | ✓ Cross-checks live operational data and contract rates |
| Output handed back to a person | ✓ Routes automatically — journal or exception buffer |
| Generic — works on any document type | ✓ Aviation-trained on 25 years of airline invoice data |
| No validation — errors flow through to the ledger | ✓ Flags errors before payment, not months later in audit |
AI Invoice Automation
Reads and verifies every line, of every invoice, for every flight. Download the one-page factsheet:
Why aviation data is the real differentiator
A comparison table makes the capability gap look straightforward. So why can’t a general-purpose AI tool do all of this? The answer is data. Validation is only as good as the reference data behind it.
For example, a generic AI model:
- has never seen an into-plane fuel ticket
- doesn’t know that Jet A-1 pricing at Heathrow follows different conventions to a regional airport in Southeast Asia
- has no way to recognise that a 0.3% drift in a supplier’s unit rate over six months is the sign of contract erosion rather than a pricing update.
Skymetrix has been processing airline cost invoices for 25 years across 135+ airlines, including seven of the ten largest carriers in Europe. Our platform has structured over $100 billion of real airline costs so it understands them in a way that generic AI tools can’t.
The system validates against the data sources airlines actually use: contract rates, fuel pricing indices (Platts, Argus, OPIS), uplift records, flight activity, tariff schedules and FX feeds.
In short, a generic OCR tool can tell you what a number says. Only an aviation-trained system can tell you whether that number is right.
From paper to paid — automatically
OCR digitizes your invoices. Intelligent invoicing systems, like our AI Invoice Automation, does that and verifies them for you.
The difference is whether invoicing errors get caught before payment, or sneak through into your operating costs.
Clean invoices flow directly to the journal. Exceptions are flagged before you pay. Accuracy reaches levels that manual checking cannot match — on every invoice, every time, regardless of format.
