Upgrading Banking: Why AI/Machine learning might not be the optimum solution

There is increased marketing by banks & fintechs that Artificial Intelligence solutions could solve many of the existing issues and is the best solution but existing solutions might be "good enough" and can be tweaked a lot faster.

The initial areas where there has been success is Anti-Money Laundering screening, Fraud detection etc. So banks/fintechs are increasingly reliant to try and see if it could solve more problems

An AI solution might not be the best-fit solution.

AI-based solutions need to be evaluated like other solutions in terms of incremental benefits vs the cost provided.

One of the examples I have observed is trials to use AI for short term cash forecasting. The existing solutions which were pure logic based algorithms could already provide a level of accuracy 70-80%. An initial assessment of AI solution would improve this accuracy to 95%+ however the benefits offered would have to be evaluated against the costs/resources needed.

Banks/Fintechs should evaluate the costs /resources

Some of the tips to consider are

Tip #1: Training Data The AI Models are only as good as the training data. If you do not have any existing, historical data which is in a structured format, the model would only be Junk-In and Junk-Out.

Tip #2: Computation There is additional computing power which might need the data to be hosted on cloud services like AWS, Azure or Google Cloud. Internally hosted solutions might end up as very expensive.

Tip #3: Timelines Models take months and years to achieve the desired accuracy levels as they spend time from moving from training data to real life data.

The key is to go in with your eyes open when evaluating solutions.