Price comparison website Uswitch, which is owned by parent company RVU, is transforming the way it works with data, as it aims for greater personalization and building a more meaningful relationship with its customers. It is doing this by utilizing a range of Google Cloud Platform services, including BigQuery and Vertex AI, whilst building an in-house data sciences team.
Uswitch has been working with Google Cloud since 2015, and expanding the relationship ever since, after it decided that its prior approach to working with data was slowing down decision making and that too much resource was needed to maintain infrastructure.
Speaking with Siddharth Dawara, Head of Data Engineering at Uswitch’s umbrella company RVU, he explained that Uswitch had been using an in-house Hadoop cluster to understand the company’s behavioural data, but that this sat in the corner of the office with a post-it not on saying ‘do not turn this off’. Hardly a modern, scalable approach to data analytics.
The company also experimented with AWS Redshift at the time, but found that it too was too resource intensive. Dawara explained:
The trouble with each of those implementations was they always came with the need for a specialist. In the Hadoop world you need to be able to hire people and train them up in your business. We wanted to prevent that situation where your business then becomes like a revolving door for people you hire, train up and then leave for more lucrative opportunities.
You definitely don't want to become a revolving door with Redshift as well. We always needed a hardware specialist on hand, or an AWS specialist on hand, because the one thing that Redshift didn’t give us, at least in its early iterations, was adapting to our circumstances.
So if our data grew 2X or 3X through high traffic events, then our infrastructure would also have to grow to 2X or 3X - but it never came back down. It only grew in that one direction. And we always found ourselves in these situations where the cluster kept getting jammed up, where someone would run a complicated workload on top of our cluster and that would make it effectively unusable for everybody else for a duration of time.
During this time, Uswitch was also in the market for a CRM system. One of the company’s Google account reps got in touch with the team and came in to give a presentation about all the services the vendor had on offer and how Uswitch could take advantage of them. Dawara added:
Interestingly, GCP BigQuery was not actually part of the sale, it was more ‘here's how we use analytics’. It's when we started using those tools that we started getting experience with the tech underneath, and we realized it was so much better, everything just worked.
Nothing was slow. We didn't need hardware specialists or anything like that. And we've been in that space pretty much since that day. Ever since we started using Google Analytics to its fullest potential, and the nature of BigQuery being underneath, we've made this bet that all of our data sits in BigQuery. All of our data services sit in GCP.
The Google Garden
Dawara explained that at Uswitch, one of the main metrics for success the company measures its data work by is how long it takes to get to the data itself. He said that compared to the company’s competitors, it “runs pretty skinny on the ground”, but that GCP has enabled a significant increase in speed for the team. Dawara said:
We don’t have as many analysts as our competitors. The nature of us being able to actually get through work quicker and more efficiently, is quite a lot more meaningful.
It's actually one of the metrics that we optimize towards as a data organization - lead time to analytics, which is where we measure how long it takes teams to actually get through work.
Not just as a duration of time, but throughput as well. It's actually very important to us that we're efficient, and the thing that efficiency gives us is adaptability.
It is incredibly important for us to adapt to market events as soon as they happen. It is very important for us to get to our customers. So tomorrow if there's a wholesale change in gas prices, it is incredibly important for us to engage with our customers as soon as we can, in the most meaningful possible way.
Since implementation BigQuery has empowered Uswitch teams with the data they need to do their jobs, whether that be focused on improving customer experience, boosting conversion rates, or cutting down on bounces. Uswitch says that it has shifted away from older ETL (extract, transform, load) processes, towards ELT (Extract, Load, Transform).
Over the past 18 months or so, Uswitch has been investing in its data science capabilities to better understand its customers. It has been using Lifetime Value Modelling to identify different customer segments and respond better to their needs. Dawara said:
From a data science standpoint, we started hiring and spinning up a data science function, because we realized that if we're making all these big bets with data, we're at a place now where things look reasonable, things are running smoothly. We should be able to get something smart out of it other than ‘what was yesterday's performance’ or ‘help me run an experiment’.
A great example is the mobile phone market. Last week, we had 65,000 deals on our website and there’s no way you can expect customers to sift through all of that information. BigQuery enables us to display a simplified version of those contract bundles, based on what we know about a potential customer.
The company had been using Databricks to move towards more personalized services, as the platform is what RVU had been using more broadly. However, Uswitch found that it wasn’t getting the speed it was used to on GCP. Dawara said:
What we were seeing was we were able to get through work but not at not at a rate that satisfied us. And this is again the same learning that we've had in the general data space, that things should be efficient, things should be smooth. If you're doing work, it should be in the space where you can add value. If I'm trying to get my data scientists to solve general infrastructure problems, I'm failing.
As such, the team began to look at Google’s Vertex AI offering to dig deeper into customer behaviour. Vertex AI is fully integrated with BigQuery, which is saving the company on time and cost when it comes to data egress. Dawara said that it has enabled the team to streamline its data operations across the whole organization. He added:
We built a few prototypes, and we saw some real potential there. I don't think it is exactly where it needs to be for people to get a tremendous amount of scale out of it - I think it's quite early in its lifecycle. But it's about seeing that potential right. It's about making the right bet. It's about saying that, okay, if we have all of our data in GCP, and there’s something with a lot of potential that isn't 100% there yet, but we're within the Google garden [that makes sense].
We’ve not got data flowing out and then coming back in, it was a pretty easy bet for us to make and to say let's see what value it gives us.
A more meaningful approach
Driving all of this investment is a recognition that Uswitch can’t operate in the way that it has in the past - Dawara said that the company needs to move towards a more meaningful relationship with its customers, not one that sees them interact with Uswitch during one off events. He added:
We can't keep operating in the same flavour that we’ve always operated, which is you get customers from paid search, you get customers from organic search, and you get a good portion of people from email, and you started just maintaining this mass market, getting a database off the back of it.
I think more and more people are realizing that our products have to actually give value to people in their everyday lives, not just once a year when the energy contract is coming to cancellation, or once your broadband provider ups their price. I think it has to be a little bit more than that.
So for us to deliver that level of personalization, where you're not just coming in and looking at the same 10 deals on the same websites every single time it should be more meaningful. Like if you're telling us through our kind of accounts and things like that, that you care about renewable sources, that you're willing to sacrifice saving money up to a particular threshold, then that gives us more room to be able to say smarter things to you. And that's where data science comes in.