In September 2021, Air Canada was tasked with verifying the COVID-19 vaccination status of thousands of worldwide employees in only two months. After realizing manual verification would be too costly and complex within the time constraint, Air Canada turned to its internal AI team for an automated solution. Working with Microsoft, the AI team used Form Recognizer, an Azure Applied AI service, to roll out a fully functional, accurate solution within weeks, meeting the government mandate on time and saving thousands of hours of manual work.
A new team within a global company
Founded in 1936, Air Canada is the fifth largest airline in North America, and before the onset of the COVID-19 pandemic, carried more than 50 million customers safely to their destination every year. The company has over 30,000 employees across the world, including a small AI team formed in 2019. Created to explore how AI and emerging technologies can benefit a global airline, the team has initiatives ranging from revenue management to capturing passenger demand to predicting the price of cargo shipments.
A daunting task (and timeline)
The ongoing COVID-19 pandemic presented major challenges across industries and nations. To fight the pandemic, the Canadian government mandated in September 2021 that all employees of federally regulated companies had to be vaccinated against COVID-19. Suddenly, Air Canada was tasked with verifying thousands of employee vaccination forms in only two months.
Air Canada’s first instinct was to do the work manually, but that idea was quickly grounded. “Not only was it not feasible within the timeframe, but it's very costly to train staff to recognize all the different types of forms we receive,” explains Cindy Yao, a data scientist on the AI team.
AI offers agility at scale
Air Canada then asked a daring question: would AI make a suitable solution? Could it handle the massive scale? And could it be implemented fast enough?
The AI team took up the challenge — a huge one from a technical perspective. Each Canadian province and territory had its own set of forms, totaling 27 different formats. On top of that, forms came in two languages (English and French), and the uploaded documents were often low-quality photographs.
An additional challenge was security. Air Canada needed a solution they could trust, as each form contained sensitive personal information.
The team went in search of a service that could harness AI to meet the government’s mandate while protecting the privacy of Air Canada’s employees.
A responsible AI solution
Knowing that time was of the essence, Air Canada and Microsoft teamed up to find a solution. After exploring a few potential services, Air Canada chose Form Recognizer, an AI-powered document processing service that applies machine learning to extract text, key-value pairs, tables, and structures from documents.
With Form Recognizer, Air Canada was able to achieve the privacy and security required for medical information. “Using AI responsibly was so important to us, and Microsoft gave us guidance with best practices on sensitive data,” says Air Canada data scientist Yassine Parakh. Another benefit of Form Recognizer was that Air Canada already used Microsoft Azure for its tech stack, so Form Recognizer would integrate naturally with other Azure services like Databricks.
From idea to implementation in only 4 weeks
Air Canada spent the first two weeks on environment setup, connecting Azure services, and training Form Recognizer on an initial set of 15 form types.
“Form Recognizer was able to predict the form within seconds. For us, that was really great due to the volume of forms that had to go through validation.”
Cindy Yao, Data Scientist, Air Canada
The next two weeks were focused on maintaining business and compliance rules for sensitive information. And in just a month, the team had produced a working prototype that could handle more than half of the form submissions. “It was a small team. We were two data scientists, one data engineer, and one system integrator. To be able to deliver such a complex project within a short amount of time was a great accomplishment,” says Parakh.
For each document, Form Recognizer verified three key pieces of data. First, was the document a valid form type? Second, did the form contain valid vaccine dose information? And third, did the name on the form match the name of an employee in Air Canada’s database? If Form Recognizer could successfully validate all three elements, no human intervention was needed. Any document that Form Recognizer couldn’t validate was flagged for human review.
“Form Recognizer provided a very efficient solution within a very short period of time.”
Cindy Yao, Data Scientist, Air Canada
The team continued to train Form Recognizer on more document types, and within two months, Air Canada had fully met the government mandate. Form Recognizer handled 75 percent of vaccination forms without human intervention.
The future of AI at Air Canada
Getting a solution off the ground so quickly showed the value of using AI to solve real-world challenges. It also earned the AI team a lot of positive attention and recognition from stakeholders throughout the company.
The team plans to build on its success by initiating new AI projects around operations research and building an API wrapper to help Air Canada use Form Recognizer in even more ways. “Air Canada is a company where there's a good amount of diverse data and the application of AI is really wide,” says Parakh. “We always have to be more creative. And that is the kind of challenge that any data scientist would look forward to.”
“Using Form Recognizer to analyze and go through all of these submissions put us in a much better position to meet the deadline.”
Kendra Dacko, Screen Program & OHS Initiatives Manager, Air Canada
Follow Microsoft