There is an intrinsic data driven culture at mobile-only bank Monzo. Neal Lathia, machine learning lead, believes there are a few reasons for this. The first is that it is a very young tech company that was founded just four years ago.
The second is that there is a lot of support and encouragement from the senior leadership to look to data to answer questions. Lathia even mentioned that the CEO is an avid user of data exploration platform Looker.
And finally, almost everything that has been built can be quantified in some way. With such a vast amount of data on which to run experiments, tests are carried out all the time.
Lathia gave the example of ‘golden tickets’. These are invitations that an account holder can extend to a friend or acquaintance so they too can join Monzo. The Monzo employees experimented with this by seeing if changing the placement of the golden ticket button would encourage more users to share it.
“All of these questions have led to a lot of experiments being run and personally that is very satisfying, as opposed to people debating which is best,” he said.
Monzo has a staff count of approximately 700 people, of which 20 are in the data team. The data team is split into three roles: product analysts, domain analysts and the machine learning team.
Lathia heads up the latter. He explained that the product analysts will join teams of engineers, designers and product managers, and support all of their data needs. They might do this by helping to test new features on customers and seeing what the uptake is.
The domain analysts deal with the analytics that supports the customer support function. There, data is used to help the customer support staff look at how well they are working, but also get feedback from the customers. “We might say, ‘oh look. A lot of customers are asking us about x. We could put that into the app and make it super easy for them to do that themselves.’
The machine learning team has the biggest mix of data and engineering. “We build parts of the app or parts of our tooling that use machine learning to help achieve some outcomes,” said Lathia.
He said that the Monzo way of dealing with data is to have a technological foundation where all of its data goes into BigQuery and the members of the data team serve that data to the teams they are partnered with.
“They’ll create a number of different data models and all of these will usually be fairly generic,” he said. One model is called ‘user stats’ and the principle behind it is that 90% of the questions asked of someone in data are ‘how has x changed over time?’ The employees making the requests might want to know the number of customers, new conversations or new overdrafts, for example.
To contend with those requests, the data goes into “a huge BigQuery table” so the team would not have to start from scratch every time.
Lathia said: “If someone asks ’how many new overdrafts did we give last week’, then they can build off the work that everyone else has done so they able to answer that, and see how that compares to the number of new conversations. One of the key things we are trying to proactively avoid is doing one-off analyses.”
He said that by having this big set of data in a table that holds the answer to 90% of Monzo data-related questions, he and his team can train people to answer their questions themselves. He added that Looker is set up to help answers that vast majority of data requests with everyone having access to Looker, either as a viewer or an explorer.
The remaining 10% are very specific questions that need more time or more data wrangling and are just relevant to certain teams.
Having easily accessible data meant that when Lathia happened to see a tweet about a breach at British Airways with customers being requested to contact their banks, he and the financial crime team were able to immediately start a proactive investigation. “I said ‘BA has given us these dates. Let’s go find the customers who have transacted with BA.’ We had a bit of filtering to do but eventually came up with 3,000 cards,” he said.
Accounts were frozen, cards were blocked and the affected customers received a notification that they were being issued with a new card and had the full support of Monzo. Lathia was pleased with the quick reaction and customer service he and his team displayed, especially when he saw the positive reaction of Monzo customers in comparison to other bank customers.
He was also involved in a project to improve the written content on the help screen and make sure that all the articles had the most relevant information in them. Instead of going through all 800 of them one by one, “a brutally manual project,” he and his team set up a Looker dashboard, which looked at the number of customers who chose to get in touch despite having read an article. “This conversion metric was used to prioritise which articles should get improved first,” said Lathia.
With easily accessible data, and the tools to access and explore it, the staff are putting Monzo ahead of the curve.