And here's the kicker! If you've been on a delayed flight that is entitled to compensation, we'll automatically let you know and offer to claim the compensation for you. You'll NEVER miss out on compensation again because AirHelp automatically got your back.
Finding 10-12 flight itineraries among tens of thousands of other emails is more difficult than finding a needle in a haystack. Therefore, the first step of the process is selecting all emails that potentially contain an itinerary. We achieve that by an algorithm that combines some good-old-rules with a machine learning piece based on Extreme Gradient Boosting model. The next step is extracting all flight details from emails. In order to do that effectively, we use a set of our home-grown parsers and partner with 2 vendors to process less popular formats. This way our coverage and accuracy is higher than what any of our competitors can provide, and is paralleled only by Google’s Inbox. Even before the official launch of our email parser, our systems were processing around 500,000 emails every week. It meant that at peak times quite powerful machines were needed to keep up with the demand. Anticipating the official launch, in order to drive down the costs the critical parts of the system got re-written to the Go language, and today’s launch was handled from one of the smallest AWS EC2 instances. Currently we repurposed the tools created for building itinerary parsers to construct software for verifying other documents sent to AirHelp by claimants and extracting from them necessary data.