CarTrawler
Nominated Award: Best Application of AI in a Large Enterprise
Website of Company: www.cartrawler.com
CarTrawler is a leading B2B provider of car rental technology solutions to the global travel industry. If you have ever booked a flight with a major airline and they offered you car-rental or ground transportation solutions, that was probably us. We are the booking engine behind car-rental solutions offered by household names such as easyjet and Uber.
The challenge was to make AI and machine learning models a key enabler of the CarTrawler business model. This meant not only doing great science but finding ways to integrate it into the culture and the core business processes.
Over the past two years, CarTrawler analytics professionals utilised their specialist skillset to deliver best in class pricing models, fraud models, partner reporting, CSAT/NPS analytics, supply analysis, traditional survey-based insights and much more. We are thought leaders when it comes to working with modern analytics technologies. We were early adopters of Snowflake and Airflow for data warehouse and ETL workloads. We are almost exclusively Python-based for machine learning and AI models. We have advanced analytics pipelines feeding a growing estate of Tableau dashboards for reporting. These dashboards were originally developed for internal use but we are rolling out access to our partners and suppliers so they can self-serve with queries on their financial and customer performance.
We are global leaders in the application of real-time python machine learning models at scale. Our models score and sort over 1 billion elements every day during peak travel periods. We have developed a full model deployment platform we call ACDC that runs these python models on top of CarTrawler’s Java-based production environment. Many of the processes you experience when using a CarTrawler app or website are now driven by real-time logic supplied by these models.
SORT: We use models to sort hundreds of cars and put the ones you are most likely to want at the top of the page. This results in higher clickthru and higher customer satisfaction.
Pricing: We price cars and insurance algorithmically to reflect local market conditions and claims risk.
Fraud: We apply machine-learning algorithms to detect and flag serveral types of suspicious transactions for further investigation. We have additional model types in development which will be announced in 2022.
Our award-winning team of data scientists, analytics experts and data engineers are active participants in the Irish and international AI ecosystem, speaking at meetups, on podcasts and in events held at CarTrawler HQ. (Interrupted by Covid — hope to resume in 2022)
Reason for Nomination:
CarTrawler is not just “using AI” We have developed a full data-to-algorithm pipeline using robust, scaleable software engineering principles and it generates significantly better outcomes for customers and significantly higher profitability for CarTrawler and our partners.
We automate delivery of insights from data. For example – we have two internally-developed workflow scheduling toolsets, AJB’s and “Action Factory”. These automate data movement and analysis. In 2020-2021, we added our “DQ Framework” which manages data quality on all the ETL feeds to our analytics warehouse.
Our data engineering team, in collaboration with the IT and Developer teams have built ACDC – a Flask platform that deploys machine learning models directly into the CICD production pipelines. Data Scientists in CarTrawler have a very rare opportunity to see their models deployed directly to production. ACDC2 – now nearing completion – further automates the model lifecycle – with model quality monitoring which automatically triggers a model refresh or full retraining when the market environment changes and model performance degrades. This allows us to maintain a large fleet of models covering many segments, geographies, customer types and usecases.
We have fully integrated A/B and multivariate testing into our model deployments. Traffic is split between (for example) a Champion and Challenger model. Once a model shows significant performance advantages, all traffic is switched to the new model.
Snowplow: We’re now in POC with transformative Snowplow web analytics. Increasingly – machine learning models support teams across the entire CarTrawler employee base. Therefore, we believe that CarTrawler is a leader in demonstrating the integration of analytics directly into the business. We would be happy to provide detailed examples at future meetups.
CarTrawler’s Insights and Data Science team have fully embraced scalable MLOps design principles. We leverage the full spectrum of DevOps toolsets to build, test, integrate, deploy and monitor our models. At CarTrawler, our data scientists and data engineers partner with the DX, IT and P&T communities to develop CICD model pipelines using Git, Docker, Kubernetes, Jenkins etc. And we use the full AWS stack to deliver models in production – including low-latency DirectConnect, Kinesis, Fargate, etc. within an EKS cluster. (Apologies for the jargon — but if you’re interested in MLOps and running real-time models at scale, you’ll understand what we’re doing and why we believe it is pretty special)
Our financial results from the models are not public but internally our Finance colleagues have verified that the models deliver ~5% additional EBITDA (profit) vs control. As the number of models grows and the number of use cases widens, we fully expect our models to generate or save millions of euros in measurable contribution. We’re also proud that our models provide better outcomes for our customers when they’re searching for car-rental. This benefits millions of travellers around the world every year.
Even during the COVID period, which decimated revenues in our industry, we have increased investment in analytics headcount and in the technologies that will help us to extend our lead in data-driven decisions within the Travel industry. CarTrawler retained the entire Insights and Data Science team thoughout the downturn. Mission statements about “leadership in analytics” are common. CarTrawler is one of the few companies that has put significant money behind the words.
Additional Information:
For a longer explanation of how CarTrawler uses AI and advanced machine learning at the core of the business – please consider listening to a recent US podcast with the Data Standard.