-
Expedia was an early adopter of AI technology and has built more than 300 ML models
-
It uses AI/ML to power both internal and customer-facing experiences
-
An Expedia exec told us she expects AI will likely be able to tackle more complex tasks in the future that could make travel smoother - even when there are delays and disruptions
She couldn’t have known it at the time, but the neuroscience degree Shiyi Pickrell pursued in the early 2000s turned out to be surprisingly applicable to her more recent work on artificial intelligence (AI) and data science. That’s come in handy for travel company Expedia, which is working hard to apply AI to improve both internal and customer-facing processes.
Pickrell told Silverlinings that before becoming a data scientist and eventually ending up as Expedia’s SVP of Data and AI, she got her Ph.D. in neuroscience and worked in biosciences on high-throughput DNA genome sequencing.
“Years later it just hit me that the Hadoop thing, the parallel processing of data, is the same fundamental as the high-throughput DNA sequencing. You break it into millions of little pieces, you’re simultaneously sequencing it or processing it and then you bring it together,” she explained. “It’s just fascinating to me that across different industries, across different fields the fundamentals are similar.”
She added the way deep neural networks function echoes what she studied about how the brain works.
These similarities were part of what made it easy for her to make the transition from bioscientist to data scientist more than a decade ago. Before joining Expedia in late 2021, she spent nearly eight years at Microsoft, finishing her tenure there as Director of Data Science and Analytics. Pickrell subsequently did a short stint at an ag-tech startup before assuming her current position leading the AI charge at Expedia.
AI adventures
AI and machine learning might not be the first things you think of when you hear the name Expedia. But maybe they should be.
Pickrell manages a team of 800 which works to turn the company’s data into actionable insights and customer recommendations. Thus far, she said it has built 300 (!) active machine learning prediction models to aid internal productivity and power both personalized experiences for customers.
“If you think about it, in our company probably the largest workforce is developers writing code. We want them to write better code faster so they can get more done,” she said of one internal use case. Pickrell said Expedia also uses Generative AI (GenAI) to help its customer service agents summarize calls faster, which in turn helps them move on to the next case and reduce customer wait times.
On the customer front, Expedia is using GenAI to help recommend itineraries for travelers based on what their dream vacation is.
But Pickrell said getting to this point was a process.
Expedia began using Amazon Web Services (AWS) way back in 2013 and in 2017 announced it was putting 80% of its mission-critical workloads and data into the cloud. But while it had primarily one cloud provider, Pickrell said when she first joined the company two and a half years ago it had 10 different machine learning platforms thanks in part to a series of acquisitions.
So, over the past several years, Expedia went through a process to consolidate these to a single platform. This, she said, helps the company work faster and has helped cut the time needed to build a new model from months to weeks.
The platform Expedia uses today is a combination of assets it built itself – mostly core components and company-specific features – as well as open source and commercial components (like ChatGPT) for more generic tools.
Making AI usable
Pickrell, who cited computer programming pioneer Grace Hopper as one of her role models, said even with the new platform, there’s plenty of work to be done to make AI that’s both usable and user friendly.
To ensure a given AI model works correctly, Pickrell said you have to be careful to ensure the data going into the system is fresh, clean and accurate. It’s kind of like weeding a garden – which also happens to be one of Pickrell’s hobbies. (She prefers flowers to vegetable gardening, in case you were wondering, and sometimes brings fresh cut flowers into the office.)
“I think one of my lessons learned is you’ve got to have the data piece all lined up, clean, high-quality data first and then you can tackle those fancier machine learning models. If you don’t get the data nailed and you jump straight into the model piece, the results won’t be gratifying. You’re doomed to stumble,” she said.
Then there’s the work of making AI that’s user friendly.
Pickrell noted using new technology like AI tools can be confusing, especially for older generations. She used her 70-something-year-old mom as an example, explaining that for her “using some of this new technology is not natural.”
“It’s still a learning curve for them to adopt or get familiar or trust those new features,” Pickrell explained. Thus, Expedia’s technologists and “product people” have to be careful to ensure their AI tools are designed in a way that’s inclusive and accessible.
What’s next
Gartner recently predicted that the percentage of new applications which use AI to drive personalized adaptive user interfaces will jump from less than 5% today to roughly 30% by 2026.
Jim Hare, Gartner Distinguished VP Analyst, added in a statement “intelligent applications recommend or automate actions instead of just providing analysis, so they can drive improvements — including better personalization, more efficient use of resources, improved accuracy, increased automation, more finely grained responses and decision support.
Indeed, Pickrell said these are exactly the kind of tasks she expects AI to be able to accomplish in the future.
Today, large language models generally focus on completing a single task. But asked to look into her crystal ball so to speak, Pickrell said models of the future will likely be able to tackle more complex action.
For instance, she noted travel plans are often disrupted – a flight is canceled and customers are left to sort out rebooking, overnight accommodations and transportation to a hotel. In the not-too-distant future, AI could take care of all of those things, she said, and potentially even offer to help customers reschedule personal appointments impacted by the travel delay.
Pickrell’s vision is similar to one espoused by Oracle SVP of AI and Data Management Greg Pavlik, who recently told Silverlinings the next frontier for AI involves chain of thought and iterative reasoning.
It’s not entirely clear how long it will take to get there given AI is developing at a nearly unprecedented pace.
But some new tools are already on the horizon. Pickrell said Expedia is planning to unveil some new GenAI features at its upcoming Expedia Explore event in May.