Why Finance has lost the AI race
Although Artificial Intelligence is widely used in our daily lives, asking Alexa to research available flights, checking your app for daily commute estimation, etc. banking has only joined the race in recent years. Customer facing functions, like Front Office, have now robots and chatbots in their workforce, however Finance functions are still reluctant to join the race.
Why is that?
DATA
Incumbent Banks have typically grown through mergers and acquisitions resulting in an intrinsic system landscape characterised by partially integrated systems. This system landscape results in data breaks throughout the bank impacting Finance processes.
As such, although Finance has high volume of data available, the data quality is often low impacting potential predictions from Machine Learning.
PROCESSES
Finance processes are highly regulated and typically include restricted and/or highly restricted data increasing the complexity of getting compliance and IT approvals to perform pilots. Approval processes are sometimes so complex and time consuming that the teams disengage and do not embrace testing and experimenting with experiment the latest available tools.
PEOPLE
AI tools require a programming/data science background for set up and maintenance. Classic finance teams include mostly accountants who do not have code experience but have a deep process knowledge and understanding of accountancy rules.
COST
New AI tools are costly to acquire and do not always provide full benefits to Finance. Tools are created for finance processes but do not always align to organisational specificities (e.g. systems complexity and integration will vary depending on the organisation, processes will have particular characteristics which an out of the box tool may not 100% solve for).
In order to join the race, Finance teams should aim to perform pilots in ring fenced processes.
While choosing the process to pilot, teams should consider processes with relatively clean data and potentially using historic or sanitised data to speed up compliance discussions.
For tool selection, teams should take into account support and training provided by vendors as well as time required to be proficient in the tool. Aiming for tools that are low code and limited training will support selection of tools that can be rolled out across the team.
Performing quick and short pilots will address both cost challenges and benefit validation. Experiments will eliminate tools that do not fully support organisation-specific systems and process characteristics and bring to life tools with a perfect match.
Find this article useful? Please do like and share to help more people to discover it!
This is a personal blog. Any views or opinions represented in this blog are personal and belong solely to the blog owner and do not represent those of people, institutions or organisations that the owner may or may not be associated with in professional or personal capacity, unless explicitly stated.