Improving Profitability: AI Use Cases for Fintechs & Financial Institutions

Photo by Carlos Muza on Unsplash

This article is about use cases of artificial intelligence for fintechs, financial and banking institutions. It was borne out of concepts I learnt whilst studying data science and AI at Stanford University. It focuses on the application of association rule mining in the finance industry to improve profitability. Some sample use cases are:

• Enhance marketing (create product bundles and identify upselling opportunities)
• Optimise website layout (so that products purchased most often are listed next to each other)
• Drive recommendation engines (customers who brought this also brought)

Irrespective of the above chosen use case, the below process can be executed to determine associations within your current dataset. Once associations are determined, these can then be applied to enhance marketing, optimise website layout or drive recommendation engines.

Steps to Execute Association Rule Mining

Step 1: Determine your tool for analysis. Both R and Python are good choices. Tip: In case you use R for mining association rules, install arules and arulesViz packages. For Python install apyori.

Step 2: List the products for which association rules need to be determined. These can be products offered by your organisation. E.g. wealth management software, private banking software, portfolio management, reporting, tax loss harvesting, robo investing, self directed trading, data & analytics, payments processing etc.

Step 3: Get list of transactions associated with the above products which can be used for mining associations. E.g. The dataset you are looking for can comprise of a list of transactions with existing financial advisors or investor clients. Some data elements within it can be name of product, transaction amount, country etc.

Step 4: Choose your algorithm. APRIORI is a great choice for transactional datasets likely to be found in finance. Some other association rule mining algorithms that can be used are: Eclat, PPV, FP-Growth, Apriori DP, FIN, Pre Post and OPUS.

Step 5: Analyse the output. Once executed the algorithm will output a set of rules that can be exploited to make business decisions. For e.g. if APRIORI is used, the output will comprise of 4 components:

1. The Rule E.g. LHS (Left Hand Side): wealth management software + robo investing + tax loss harvesting = RHS (Right Hand Side): client retention period
2. Support: This denotes the fraction of transactions in the dataset that contain that product/ product set E.g. 0.175829
3. Confidence: The probability that the proposed rule is true based on the given data set. E.g. 0.83766
4. Lift: If lift = 1: It indicates that there is no association. If lift > 1: It suggests a positive association. A high lift value indicates a strong association. If lift < 1: It implies a negative or inverse association. A low lift value indicates a weaker association. Analysts or data miners typically look for rules with lift values significantly greater than 1 because these rules indicate a non-random association between items, making them potentially actionable and informative for decision-making.

Statistically, the most interesting rules will be the ones with the highest support, confidence and lift combinations.

To summarise this article, we:

• Identified use cases for AI for fintechs and financial institutions
• Listed steps to implement association rule mining
• Talked about how to identify meaningful patterns

Once done, the next step will be to take these patterns and use them to tweak your marketing strategy, optimise website layout or drive recommendation engines.

I have deliberately kept this article at a high level. If you feel that it will be valuable to go into the next level of detail, please reach out. All the best for your data mining adventures!