If you’ve ever been in a car accident, you’ll be familiar with the European Accident Form (EAF). It’s a standardized document for noting down all the details of a crash between two parties.
While crucial for insurance companies to properly process a claim, extracting the information from the form also entails quite a lot of administration.
With well over a million accidents on European roads last year, improving and speeding up EAF processing benefits both insurer and client.
The paper challenge
The major hurdle that slows down the extraction of information from an Accident Form is that most of them are still paper documents.
Insurance mailrooms are therefore largely staffed with people whose sole purpose is to transfer the data from these forms to digital systems.
Posting or depositing the paper document and copying the information can take a couple of days, slowing down the entire process. Digitizing this would be ideal, but even then you’re met with some challenges.
Is an app as the ideal solution?
If paper is the problem, a logical fix would be to have a digital version which you can fill out using an app.
This would indeed be the easiest solution, if not simply for the fact that people are creatures of habit, and the shock of the accident often makes them forget they have an app.
Ventures to digitize the form have been less than successful in the past. Moreover, not everyone on the road is well-versed in technology.
Elderly people, for instance, who use a senior-appropriate smartphone or simply don’t even have one, would not be able to fill out the form. And who’s to say your phone will definitely survive the accident?
Stickers and AI:
So no app then. But that doesn’t mean we can’t digitize the forms without human intervention.
Technology powered by artificial intelligence (AI), such as OCR and data extraction, is already quite good at recognizing text from images and labeling it correctly.
But when you’ve just been in an accident, writing legibly is the least of your concerns. And this proves a problem for basic AI algorithms.
Pre-printed stickers with contact details for the people and insurers involved could solve this problem.
This also requires everyone to be on board, however, and could mean even more administrative work as insurers would need to provide these stickers.
While stickers would be the easiest solution when using AI, it’s still not an ideal solution. But this doesn’t mean artificial intelligence is out of the picture.
At Docbyte, we believe handwriting recognition through AI provides the ideal marriage between letting customers work the way they’re used to and digitizing this process for insurers.
To that end, we’ve been working hard to overcome some of the obstacles involved in creating and training the necessary algorithms.
AI can analyze handwriting correctly, but often only in ideal conditions. Deep learning techniques have improved over the years and accuracy is increasing, but mostly for block letters clearly written by one person, not scribbly cursive writing by just about anyone.
It’s not just the handwriting, however; there are other difficulties too. The AI needs to ignore the printed text and take into account the background color on the Accident Form, which indicates whose information it is.
The quality of scanned forms can also vary greatly. An algorithm can theoretically cope with all this, but for that it needs data. Lots of it. And that’s usually where things go awry.
How we’ve overcome the challenges:
We’ve employed two techniques to deal with the challenges of using handwriting recognition on the European Accident Form: object recognition and synthetic data.
The first, object recognition, has been the subject of scientific research, but has not yet found its way into practical applications.
We’ve used it to recognize individual letters instead of entire words, and consequently trained individual models that work this way for some of the fields on the form.
A great approach, but we still need data to train the models for the letters. Gathering all this information would be time-consuming and require a lot of effort.
So instead of filling out countless forms ourselves, we once again used computers to help us. We created synthetic data by having artificial intelligence fill out forms with simulated handwriting.
This enabled us to create a whole heap of completed forms in a matter of minutes, perfectly labeled to speed up the process.
Through the combined use of synthetic data and object recognition, we’ve achieved great results in extracting data from the four most important fields on Accident Forms:
- License plate
- Insurance number
- Insurance company name
We’re not there just yet, but we’re getting close. For example, we’ve achieved almost 80 percent accuracy on the names of insurers.
We’re also making steady progress on the other fields, and this will only improve the more data we feed into our model.
Years of working with insurers and many deep dives into artificial intelligence technology have put us in a perfect position to improve the processing of European Accident Forms with AI.
Want to make the burden of this form a thing of the past? Reach out to our experts who will be happy to help.