Feb 22, 2023
Data science is the art of analysing complex data and extracting actionable insight out of it, says head of data science at CarTrawler, Mike Wilson. “The way we achieve it is through using sophisticated mathematical and statistical methods and also machine learning (ML) and artificial intelligence (AI).
Wilson says the key to what’s important for excellent data science is understanding the business domain. “Data science can be applied to any data in any industry but designing good data science solutions relies heavily on understanding your business domain.”
“That’s something that we value in CarTrawler – we want to understand how our market operates and that feeds into how we build our solutions.”
Data science enables businesses to get more granular in how they operate and transition away from averages to propensities. “Rather than having a one size fits all approach, machine learning allows you to tailor your offering at a greater level of granularity than say a human-developed solution would be able to provide,” says Wilson.
This allows the business to achieve greater uplifts by offering the right product to the right customer at the right time. For example, a human user may be able to apply an average price strategy to a single market that works well overall. But as with any average approach it will perform well in some pockets and perform poorly in others. A machine learning algorithm, however, can analyse and understand all the underlying trends within that market, be they univariate or multi-variate, enabling it to apply multiple price strategies to optimise performance even further. Wilson explains that you can monitor all the univariate charts and graph you want but there are trends and behaviours that you’ll just never be able to see. You need very sophisticated techniques to pull out and utlise the multivariate trends and it’s in these where the greatest value lies.
“It’s the unknown unknown – it’s something you don’t even know exists, because you don’t even have the tools and capabilities to pull this out and see it. This is where data science comes in and helps. We have the tools and can extract that information out in an efficient and understandable manner.”
Data science can also help improve workplace efficiency. “Many of the menial jobs we tend to do within a business, we can transfer over to an algorithm or a machine to perform,” says Wilson. For example, every day an analyst might check performance of key metrics in their particular market to identify areas of under/over-performance. This takes time to perform and ultimately a lot of that time is spent checking metrics that are completely fine. This is the type of job that can be replaced by either an automated process or an algorithm of some sort. We’ve done both here at CarTrawler, we have automated processes that check key metrics and algorithms that identify when those metrics have deviated from the norm. This is performed in a fraction of a second and allows the analyst to focus their time and efforts on more value-add work.
“First and foremost, CarTrawler uses data science to optimise performance for partners and suppliers.”
The Data Science team sits within the Revenue function and work closely with trading colleagues to optimise performance across many of our trading levers, says Wilson. “We’ve built a whole suite of algorithms and tools that our trading team use on a daily basis to improve performance. We support them with that while also developing further applications and upgrades.”
Broadly speaking, the data science team is split into two sub-team – one team is focused on developing the tools and infrastructure which enables us to build, deploy and maintain all our Data Science solutions. They’re called the ML Ops team.
The other team is focused on building data science solutions through machine learning, predictive models, forecasting, optimisation algorithms, etc. They’re called the Modelling team.
Some of the main responsibilities of the Data Science team include automating pricing across our insurance and car products, developing our recommended sort engines (how we order and display content on site across all our different channels), detecting fraudulent transactions, predicting the risk of insurance claims costs and forecasting future costs and metrics. All of these solutions are built, deployed and tested using our in-house developed data science platform. They act together in one ecosystem allowing us to leverage and share the outputs of each across all solutions to further improve performance.
CarTrawler is very proud of the technologies and tools it has developed in-house. Developing cutting-edge machine learning algorithms and technologies is key to CarTrawler’s success in the travel tech industry. Here’s a quick overview of some of these innovations.
ACDC is a service that allows CarTrawler to deploy and serve its machine-learning models directly into production. “When a customer comes to our site and makes a search, what happens in real-time is our production system calls out to one of our ACDC APIs and gets scores in real-time. So, we’re offering an optimised insurance price, optimised car price, and optimised sorting real-time in a couple of milliseconds, which is generated through machine learning algorithms taking in real-time data. It’s all hosted in the cloud and we manage and maintain it along with our colleagues in DX and IT. We package up all our models into individual containers, deploy them in AWS, and then they are called on a real-time basis.
“The great thing about this is it scales really well – so we can add more models as we build them without impacting latency or performance.” Coupled with some other tools the team have developed, ACDC also allows us to efficiently retrain and refresh models on a regular basis, so we can keep things up to date, and ensure that the models aren’t decaying.
“We’re on version 2.0 – we initially built it on bare metal servers with the support of our P+T department, and have now transitioned onto the cloud, which gives us a lot more flexibility and scalability.”
Another exciting tool that the Data Science team has developed in the last three years is ROPE – real-time optimized pricing engine. It’s an optimisation algorithm, that produces a selling price for both our insurance and car products.
The ROPE algorithm uses a propensity to purchase machine learning model as part of an optimisation process to determine the optimal selling price depending on the situation it’s presented with. The algorithm is deployed to our ACDC service and as the name suggests utilises real-time data, ensuring we are providing the right price at the right time. The algorithm is applied to both our car and insurance products. There are obvious nuances and differences between the two products but the framework and mindset behind the approach is relevant to both products. “These two tools are our main revenue drivers in the business, so are very important and need constant maintenance and upgrading.”
Classon HouseDundrum Business ParkDundrumDublin 14Ireland
14th Floor575 5th AvenueNew YorkNY 10017United States
3 Waterhouse Square138 – 142 HolbornLondonEC2N 2SWUnited Kingdom
Manage my BookingFAQs