Hey guys, I’m a data scientist in our two-person data team. We’re responsible for all the data related stuff for the company, anything from setting up and maintaining ETL pipelines so we can get clean accurate data, creating dashboards to track our KPIs, doing analysis/modelling to inform decisions, to creating machine learning data products.
One of the things I’ve been working on this week was modelling growth of our users. When a user signs up for a Monzo account, we track where they came from. Unfortunately, lots of signups can’t be attributed to a source. The model I’ve built helps us answer: of all the untracked users, how many came from word of mouth, and how many came from Monzo’s organic brand growth?
For those interested, I modelled this with a hierarchical/multilevel Bayesian model with heavily informative priors that impart our domain knowledge into the model. I also borrowed a few techniques from Facebook’s Prophet. Unfortunately, I don’t think I’m allowed to share my findings with you 
. All I can say is that the results look pretty good for Monzo so far! 
Another thing I’ve been working on is an attention based model for customer support. As @enginoid mentioned, we’re trying to make our customer support scale. This model basically takes a customer support enquiry and determines which sentences/words are important in order to resolve this issue. We can do many things with this model: build a classifier on top of it for auto-tagging conversations, use it as a search ranking algorithm for our in-app FAQs, or if we feed it transactional data, it can tell you what transaction a user is most likely asking about, making this a super versatile base model to use. An additional bonus is that it makes whatever model you’re creating down the line a lot more interpretable since you know why a model make a certain decision.
Here’s an example of what the model outputs (glad I can at least show this). It tells us that the first sentence is the most important, and the most important words in that sentence are ‘annual’, ‘withdrawal’, ‘limit’, ‘renew’. Pretty cool stuff! 
Happy to chat about this stuff if you’re interested! I’d like to nominate @yehudi for the next update