AI Hackathon

Personalisation can be applied in a range of different areas in the bank and something we work with daily. In some cases it is a time-consuming task which would benefit greatly from the help of AI. Personalisation is about increasing relevancy in the communication with our customers in e.g. Email, mobile bank, SMS, websites, 1:1. By increasing the relevancy we can make tailored service plans and better customer experience.

Ways to use AI here include text generation, image generation, text translation etc. Here one may use customer segments, geographical features and possibly transaction data to better understand the customer and to create personalised content. Once content is created, it may need to be translated into other languages while keeping the tone of voice.

Topic 1: Text generation 

Generate textual content to be used in different communication channels such as email, SMS, ads, or websites. One can also include the possibility to translate between languages while keeping the tone of voice

Data: N/A

Topic 2:  Image generation 

Generate personalised images to be used in different communication channels e.g. change background depending on target group.

Data: N/A

Working with the latest information and being up to speed with changes that might have financial impacts is crucial at large financial institutions. By analysing financial reports, news, morning reports, earning calls etc, the bank can get a good picture of current and potential customers as well as the market in general. This information can be used to e.g. assess risk profiles for companies or to identify what customers to target.

Utilising this information can be hard and time-consuming. It often involves reading through and summarising a vast amount of text or listening to hours of calls. With the help of AI one can imagine the amount of time and resources that could be saved.

Topic: Transcribe calls and derive insights

a) Transcribing audio files such as earning calls to be used for analysis, or use already transcribed calls

b) Get a summary on what have been discussed. Any updates on their products? Have the questions a general theme? What are the top X questions from investors per sector etc.. Are there any themes related to any given industry?

Data: Transcriptions: https://www.kaggle.com/datasets/tpotterer/motley-fool-scraped-earnings-call-transcripts

Topic: Financial reports

Potential ideas of exploration could be to: 

a) Use LLMs to help summarise or understand large amount of financial reports such as annual reports

b) Identify trends within the company as well as within the sector or country e.g. semi-conductor shortage, covid etc

c) Find companies with growth plans in specific countries

One could also include the possibility to translate between languages to broaden the knowledge base Financial reports can be found on: https://mfn.se/all/s

Data: Financial reports can be found on: https://mfn.se/all/s

Topic: Morning reports

Understanding trends on the markets by following companies morning reports or different newsletters. Use LLMs to find trends and to summarise the latest within some sector or market

Data: Transcribes calls + Financial reports + Financial news. https://www.kaggle.com/datasets/jeet2016/us-financial-news-articles/code 
 

Topic: Price prediction

Predict price movements using 1) share prices 2) financial reports etc.     

Data: Yahoo finance + financial reports

Our biggest communication channel is our websites and therefore it is crucial to ensure that the information there is easily accessible to our customers. In this use case one can focus on the content available on our XX pages (DK, SE, NO, FI) and how to make it easier to access and understand it. Here it is worth considering that it is important that only relevant content is shown to the user, e.g. customers in Sweden shouldn't be shown Denmark specific content. Can a modern LLM chatbot be used here or maybe a more sophisticated search engine?

Potential ideas for further exploration for this use case can include 

1) multi-lingual conversations - many users might not understand the local language

2) whether an LLM agent should be able to perform any actions such as blocking cards

Topic: Chatbot

Have an LLM powered chatbot to assist customers on tricky questions or situations through conversation

Data: Nordea.se .dk .fi. no

Topic: Search engine

Have a sophisticated search engine to make finding information on our sites easier

Data: Nordea.se .dk .fi. .no

Working with regulation is essential for a large regulated bank. Part of working with it is to follow up on comming regulation and start when one gets the first draft version. Once the first draft version has been published then the task is to break it down into what does it mean for us. However, when a second version arrives then the team has to compare it with the first draft version that have been approved to make sure we are still aligned with the new version. This can become a manual time consuming process that we are keen to automate. Is it possible to use LLMs to compare versions of the same regulations?

Topic: Regulation analysis

In this topic, one could imagine using with a few legislations and build a prototype on how to compare versions of the same legislations. One could also imagine summarising it based on personas, e.g. as a data scientist, what do I need to know about AI act?

Data: For instance, AI Act: https://www.europarl.europa.eu/legislative-train/theme-a-europe-fit-for-the-digital-age/file-regulation-on-artificial-intelligence