Automatic Recognition of Receipts

Background information

When you capture a receipt using the camera on your smartphone or tablet, the Cleemy mobile app automatically creates an expense.

When adding the receipt, Cleemy applies character recognition to extract the amount, date, taxes and expense type. When information is detected, the expense created after adding the receipt has these pieces of data filled in.

Activating and configuring the feature

An administrator has to enable this feature from the ‘Parameters’ page in the Settings menu. Using the last setting on this page, you can choose between OCR provided by Abbyy or our new partner Mindee.

Mindee is a faster and more accurate OCR engine. However, it currently only works on all Europe countries, United States and Canada. If most of your expenses occur on non-Roman alphabet, we recommend you use Abbyy.

If the feature is not enabled, receipts will be automatically converted into expenses, only character recognition is disabled. The expense will be invalid and will require the employee to add information.

After enabling the service, your employees must also update their mobile app’s configuration (in its menu) to immediately use the feature. If they do not do this, they will still have access to the feature a little later: the configuration automatically updates every 2 days. The nature correlation table is used to assign expenses identified by automatic recognition to an expense nature.

What does the feature include?

The ‘camera’ button on the mobile app’s homepage allows the employee to take a picture of a receipt, so an expense can be created.

When you do this, the image is automatically enhanced (brightness, contrast, converted to black and white) and cropped. This enhanced image will now be used as an overview of the receipt, making it much easier for controllers to read.

After taking its picture, the receipt is sent to the Cleemy server, which will organise extraction of the image’s data:

  • The feature requires Internet access. You can capture new receipts in offline mode, but they will not be processed until the mobile app is connected to the Internet again.
  • It takes about 10 seconds to process the receipts after they are uploaded. Receipts being processed are shown in an expense with the title ‘Receipt processing’:

mceclip0.png

This feature is particularly well-suited to users who capture their receipts as they receive them and complete their expense report later as well as users who create their expense report all at once, at the end of the month, for example.

Whether or not it is successful, the receipt is always automatically converted into an expense. The user simply needs to fill it out based on what elements were or were not reported by the recognition (for example, by entering attendees) before submitting their expense report.

If an expense with the same characteristics already exists (date and amount) but has no receipt linked to it, Cleemy will merge them and attach the receipt read to the expense, possibly adjusting its VAT.

If the image quality is insufficient for character recognition to work (image too dark, very crumpled receipt or hard to read), the image will still be linked to the expense but it will be invalid while awaiting additional information.

Optimising the recognition rate

The technology for capturing data from receipts gets better every day, but it is still unable to read all receipts.

  • We focused attention on restaurant, fuel, toll and parking receipts. These represent 70% of all expenses. PFD receipts from your merchant sites and transport companies are also handled well. 
  • The feature is officially available for receipts in Europe and North America. The recognition rate for receipts issued in other countries may be significantly lower. Currently, only receipts using the Roman alphabet are supported.
  • Handwritten receipts are not supported.
  • Very crumpled or folded receipts have lower read rates: read your receipts before capturing them.
  • Image quality has a significant influence on the rate of success: try to take photos in a sufficiently well lit location and make sure your camera focuses on the text.

Accuracy of results using the Mindee OCR engine is:

  • 95.3% for the total amount,
  • 97.9% for the date,
  • 90% for detecting the expense nature,
  • 96% for the currency
  • 92.7% for the VAT

 

Page content

Was this article helpful?
2 out of 4 found this helpful