This week we’re lucky enough to have @Neal joining us to answer your questions. Little copy and paste from his site gives us the following:
I’m currently Director of Machine Learning at Monzo Bank in London, where we’re focusing on building machine learning systems to help make money work for everyone.
Neal also wrote a fantastic blog post about what the team work on which can be found here.
Neal will be joining us on Fri 2nd July around 2pm - 3pm (schedule permitting) to answer the lovely questions that you all have given him
Machine learning! Right up my street. Excited for this one.
Hi Neal! Just a few questions for now.
Does Monzo have any plans to use machine learning to augment the customer support experience? Atom Bank have had some fascinating discussions in regards to this, so would be interested to hear if Monzo are exploring something similar, and how they plan to approach it. The transition towards more self help focused support could really use a little machine learning powered oomph so things don’t wind up at a dead end.
Are there any plans to improve upon the machine learning that powers summary? This aspect of summary is a very annoying miss for me, and completely misrepresents my financial outlook, to the point I’d rather the feature be void any predictive aspects entirely. Is this something you’re looking at improving?
What other things are you using machine learning for that we may not be aware of as customers?
What is the coolest thing machine learning has allowed Monzo to achieve?
If you could improve one aspect of Monzo, what would it be, and why?
Since the above covers the present and future, can I focus on the past please
Your résumé is seriously impressive and you’ve done some spectacular sounding things. Since you graduated I imagine you’ve spent countless hours researching and working on side projects and stuff to further your skills in your free time.
Working in a tech role too this is what I really struggle with. How do you:
a) Find time
b) Motivate yourself
c) Come up with an idea/problem/project to try solve
Any tips on self-learning and development and how you managed to get where you are today will be greatly appreciated
If you were in a battle of the Dan’s against @Minnie, @danhughes and @dan5, who do you think would win. (Four way battle, to have the title of Best Dan)
You’re allowed to bring one weapon of your choice. So far Minnie is in the lead with a Skillet.
For the budget, if I set it to £300 then it thinks I have £10 a day and above/below this, it says I will have money left over/run out.
But no all days are equal. Do you envisage a time when it’s smarter? Let’s say on average I spend £2 a day M-F but £30 a day at the weekend. If I have 3 days to go (MTW) and have £7 left then I’ll be okay, but if I have £50 left for 2 days (SS) then I’ll run out.
On a slightly more serious note though, as someone not in tech, I am fascinated by the scale of what machine learning can do.
My current level of understanding, is things like the GPT-2 ‘tool’ for machine generated paragraphs, people using Machine Learning to solve games, or things like deep fakes (which I’m not sure if that’s actually a machine learning thing).
I guess there are really broad strokes of ambition between things like fraud detection, “help” systems, and spending analysis. please
What are your/Monzo’s goals when it comes to rolling out machine learning + AI in those areas, and what will that mean for the consumer, in terms of ‘adding more value’.
What approach do you take to finding the right balance of machine learning algorithms and the human element? Do you see machine learning as a means for replacing some human elements, or as something that can augment them?
What are your thoughts on how other banks approach Machine learning? What have others in the industry done that you think monzo has done, or could do, better thanks to machine learning?
Is an AI chatbot something Monzo have explored, or planning to do, to serve as as a concierge between FAQs and human based support? How easy do you think these systems should be for users to bypass the AI and speak to a human? A shortcut (/human) like N26, or click buttons and hope for the best like RBS?
We use machine learning in this area already! When you get in touch with customer support, our agents (COps) have a ton of saved responses that they can use to reply to you - so that they don’t have to type out how to get a replacement card hundreds of times over. We have a system that gives them recommendations on how to answer - but they can pick the right one. We also use machine learning as part of a system that routes customer queries to the right type of specialist - when we built this last year, we showed that it helped our customers get to an answer quicker!
I would love to use ML for summary- we have only done a couple of internal hack days on this so far. Let me know your ideas and I’ll pass them on to the teams
There’s quite a broad range - from things you can interact with yourself (the search bar on the help tab of the app), to a lot of ML that we use to detect and prevent different kinds of financial crime.
I think the coolest things we’ve shipped have been built when we joined forces with Engineers and Designers - that’s where I think the magic happens! I often say internally that if customers look at our app and think about AI, then we’re just distracting them rather than giving them truly wonderful experiences.
One thing I loved about joining Monzo in 2018 was the culture of one-to-one meetings: we didn’t need an agenda or a reason to go for a 30 minute walk and share what we’re working on with each other. That has become much more difficult to do during the pandemic, and I’d love to see it return as the world becomes a safer place.
Ooh, that’s a great question @danhughes! It’s also an area with a lot of fascinating research. A couple things that jump straight to mind.
We don’t have a separate team who is responsible for “ethical AI” - it’s all of our jobs to ensure that our models behave the way we expect them to. That means that there are some areas where we could use ML but we just don’t.
For areas we do work in, we spend a lot of time thinking critically about what we’re putting into the model - ensuring avoid any features that could be explicitly or implicitly biased. But, more broadly, we also don’t rely on just aggregated stats about how our systems work: we look at the data in depth, and design controls for our live systems to think about all the things that could go wrong.
