Last week we hosted our 13th Demo Day, where our latest fellows presented their projects to the companies they have worked for, peers and our industry friends. It was amazing to see what these 20 artificial intelligence (AI) projects delivered in just six weeks. It shows how much potential there is for data science to bring meaningful business benefit.
For eight weeks our talented postdocs were trained and mentored by our top data science team to hone their raw skills. The fellows also worked with their assigned host companies as part of a six-week placement discovering how best to apply AI to a real opportunity.
Demo Day was a chance to see what they delivered. I have picked out some of my favourite projects.
1. easyJet: Seat price optimisation
Airlines must decide how to price seats over time and as other seats sell. This is often left to intuition and testing new strategies is risky – it might leave revenue on the table.
easyJet took part in the fellowship to create a simulator to help test seat pricing strategies. The fellow, Simon, used reinforcement learning to find the optimum pricing in order to maximise revenue for flights.
It is easy to see how the technique can be applied to many pricing strategy businesses – retailers, car rental and consumer packaged goods (CPG) to name a few.
2. HSBC: Finding high-risk customers
Our fellow, Kyle, worked with HSBC to better detect high-risk customers. The current system for detecting customers takes hundreds of days and costs thousands of pounds. HSBC wanted to create an ongoing detection process.
Kyle trained a machine learning (ML) model on a data set of 230,000,000 points from 60,000 customers. Using cutting-edge ML techniques he was able to sort a vast number of customers into low-, medium- and high-risk categories, with only the high-risk categories triggering a manual review.
The benefits are outstanding, with the potential to reduce manual reviews by 81% and increase the time HSBC can spend analysing high-risk customers by 550%.
3. Abaco Group: Combating farm subsidy fraud
Farm subsidy fraud is a problem where payments are made to people who aren’t growing the crops they are meant to. The current system of inspecting farms is inefficient and relies on an inspector visiting the site, which can be costly, slow and sporadic.
Our fellow, Luis, worked with low-resolution satellite images that didn’t appear to show anything about the farmer’s fields. We trained a ML algorithm to look for tiny signatures of colour change over time – the ‘fingerprint’ for crops. This has the potential to move the current system from one-off on-site inspections to continuous monitoring.
This is a great application of visual recognition trained in a specific problem. Our project for Abaco shows how AI can find important events in the world, even when the data seems to make no sense.
What will you build?
These inspiring short projects show what’s possible with AI in just a few weeks. Faculty’s fellowship has seen us work with more than 130 organisations that have either submitted projects or recruited fellows to work for them. If you want to be one of these companies, or to find out how we can help you with an AI project, get in touch.