Pricing
Receipt Processing · AI extraction

Photograph the receipt. Let AI do the data entry.

Vendor, total, line items, VAT, date, category - extracted in seconds, with confidence scores attached so reviewers know exactly which fields to double-check. JPG, PNG, PDF, HEIC. 25MB per receipt. Native multi-language support.

~3 sec
Average extraction time
94%
Field-level accuracy
4
File formats supported
app.omnipath.ai/expenses/extract
PRET A MANGER
42 King's Road
London SW3 4UD
VAT: GB 232 1849 27
Chicken Avocado£5.95
Filter Coffee Lg£2.85
Kale & Edamame£6.50
Still Water 750ml£1.95
Subtotal£17.25
VAT @ 20%£3.45
TOTAL£20.70
02 May 2026 · 13:24
CARD ****4521 · CONTACTLESS
Receipt #847291
~ Thank you ~
AI Extraction
Done · 2.7s
Vendor
Pret a Manger
99%
Date
2 May 2026
98%
Total
£20.70
99%
VAT
£3.45 @ 20%
97%
VAT no.
GB 232 1849 27
82%
Category
Meals - Lunch
94%
Line items · 4 detected
Chicken Avocado£5.95
Filter Coffee Lg£2.85
Kale & Edamame£6.50
Still Water 750ml£1.95
File formats

Whatever the camera roll or scanner gives you.

No format conversion. No compression. No "please save as PDF". Upload the file the way it left the device.

JPG · JPEG
Phone camera
The default for iOS and Android camera apps. Most common upload type.
PNG
Screenshots
For digital receipts captured from booking confirmations, email screenshots, app receipts.
PDF
Scanned & emailed
Multi-page invoices, scanner output, supplier-emailed receipts. Each page processed separately.
HEIC
iPhone native
Apple's high-efficiency format. Smaller file sizes, no quality loss in transit. Auto-converted server-side.
How it actually works

Three layers of intelligence. One receipt.

Most receipt OCR stops at "the AI got it right" or "the AI got it wrong" - there's no middle ground. OmniPATH treats every extracted field as having a confidence level so reviewers know exactly where to look.

01Per-field confidence scoring

Every field has a confidence score.

Vendor name extracted at 99%? Skip the review. VAT number at 67%? Highlighted in amber so the reviewer's eye goes straight to it. The system tells you exactly where its judgement is shaky - so you spend zero time double-checking the easy fields.

Confidence thresholds are configurable per field type. By default: above 90% auto-accepts, 70–90% flags for review, below 70% requires manual confirmation before submission.

  • Per-field, not per-receipt - so a 99% total + 65% VAT-number flags only the VAT field
  • Three colour-coded bands: green (high), amber (medium), red (low)
  • Configurable thresholds per field (vendor, total, VAT, line items, date)
  • Sub-70% scores blocked from auto-approval - must be confirmed first
Confidence breakdown · Pret receipt
Vendor99%
Pret a Manger · auto-accepted
Date98%
2 May 2026 · auto-accepted
Total99%
£20.70 · auto-accepted
VAT amount97%
£3.45 @ 20% · auto-accepted
VAT number82%
GB 232 1849 27 · review recommended
Receipt no.61%
847291 · manual confirmation required
02Line item detection

Not just the total. Every line.

Most expense tools capture the receipt header - vendor and total - and stop there. OmniPATH reads each line. So when finance needs to ask "what was actually on this £487 dinner?", the answer's already there: 4 main courses, 2 starters, 6 drinks, broken down with prices and per-line VAT.

Line-item detection works on printed receipts, handwritten amendments, multi-column layouts and the awkward thermal-print formats that other OCR engines struggle with.

  • Item description, quantity, unit price, line total - all separated
  • Per-line VAT rates detected (mixed-rate receipts handled cleanly)
  • Sub-totals, service charges, tips and discounts identified separately
  • Currency-symbol detection for international receipts
Line items · The Crown Hotel
8 detected
Item
Qty
VAT
Amount
Steak Frites
2
20%
£68.00
Sea Bass Fillet
1
20%
£26.00
Caesar Salad
1
20%
£14.50
House Red 250ml
3
20%
£28.50
Sparkling Water 1L
2
20%
£8.00
Espresso
3
20%
£10.50
Service charge 12.5%
-
0%
£19.31
Subtotal
£174.50
VAT @ 20%
£34.90
Total
£228.71
03Duplicate detection

Catches the same receipt twice.

The classic expense leak: same lunch claimed by two people, or same receipt submitted in two different months. OmniPATH compares every new submission against the last 90 days of expenses across the workspace - same vendor, similar amount, same date window - and flags potential duplicates before they reach an approver.

Detection runs at submission time, not at month-end audit. The submitter sees the warning, the approver sees the warning and either can resolve it inline.

  • 90-day workspace-wide cross-check at submission time
  • Match on vendor + amount (±£0.50) + date (±7 days)
  • Cross-employee checks - catches the "we both expensed lunch" scenario
  • Inline resolution: confirm legitimate or withdraw with one click
!
Possible duplicate detected
A receipt from Pret a Manger at this amount was submitted 4 days ago by Sarah Johnson. Confirm this is a separate expense or withdraw your submission.
This
Pret a Manger · 2 May 2026
Submitted just now · You
£20.70
Match
Pret a Manger · 28 Apr 2026
Approved · Sarah Johnson
£20.70
Confirm separate expense Withdraw submission
The pipeline

From shutter to submission.

Four stages, every one running server-side after the photograph is taken. Total elapsed time: usually under five seconds.

Stage 01

Image normalisation

Auto-rotate, deskew, perspective-correct. Convert HEIC to standard format. Enhance contrast on faded thermal receipts.

~0.4s
Stage 02

OCR + vision pass

Multi-model OCR plus a vision-language model that reads layout context. Critical for non-tabular receipts and mixed handwriting.

~1.4s
Stage 03

Field extraction + scoring

Identify vendor, totals, dates, VAT, line items. Generate per-field confidence scores. Cross-validate (subtotal + VAT = total).

~0.8s
Stage 04

Categorisation + checks

Suggest expense category from vendor + line items. Run duplicate check against last 90 days. Apply policy validation.

~0.5s
Frequently asked

The technical details.

How does the AI handle handwritten or partially-faded receipts?
The vision-language model reads layout context, not just characters - so it can often infer "this faded number is the total" from position even when the OCR confidence on the digits is low. Confidence scores reflect this honestly: a faded total might come back at 73% with the system asking the user to confirm the figure rather than auto-posting an incorrect amount. Pure handwritten receipts (rare in practice - corner shop, parking attendant, market trader) are extracted at lower confidence and almost always require user confirmation before approval.
What languages and currencies are supported?
English, French, German, Spanish, Italian, Dutch, Portuguese, Polish and Welsh are validated to 90%+ extraction accuracy. Other Latin-script languages work but with lower-confidence outputs. Currencies: GBP, USD, EUR, CHF, NOK, SEK, DKK, AUD, NZD, CAD, ZAR are auto-detected from the receipt format. FX conversion to your workspace base currency happens automatically using the receipt date - historical FX rates are stored alongside the original amount for audit.
What happens if the AI gets a field wrong?
Every field is editable before submission. The submitter sees the extraction with confidence scores, fixes anything wrong and submits. The correction is captured by our model-improvement pipeline (anonymised, aggregated) - your specific corrections train future extractions for your workspace, so the AI gets better at your vendors over time. After ~30 days of use, vendor-recognition accuracy on your top 20 most-frequent vendors typically reaches 99%+.
Can I capture multiple receipts in one photo?
Yes - the receipt-detection layer identifies multiple distinct receipts in a single image (the "I dumped a week's worth on my desk and snapped one photo" workflow) and creates one expense entry per receipt. Works up to about 6 receipts per image before image quality degrades the per-receipt accuracy.
Are receipts stored for audit, or only the extracted data?
Both. The original image is retained alongside the extracted fields for the full statutory retention period (6 years UK, configurable per workspace). The extracted data is fully editable; the original image is immutable. Auditors get full provenance: extracted data, original image, edit history, approver name and timestamp on a single audit trail.
How does mileage / distance-based receipt processing work?
Mileage is a separate flow - no receipt to extract. Submit start + end addresses, OmniPATH calculates the distance via Google Maps and applies your workspace's per-mile rate (HMRC 45p/25p defaults, configurable). Round-trip and multi-stop journeys supported. The calculated mileage record carries the same audit trail and approval workflow as a normal receipt expense.
Bring a receipt

See it process your receipt.

The fastest demo we run. Take a photo of any receipt you've got, drop it in. Watch fields populate with confidence scores in three seconds.

30-min demo · We'll process whatever receipt you bring