AI IN FINANCIAL SERVICES (FINEXTRA REPORT OCTOBER 2019) KEY TAKEAWAYS
THE STATE OF AI IN FINANCIAL SERVICES
(A REPORT ON A SURVEY FROM FINEXTRA AND QUANTEXA – OCTOBER 2019)
EXECUTIVE SUMMARY & KEY TAKEAWAYS
This survey sought to illuminate the evolution leading to large scale investment, adoption of AI approaches and return on investment
Senior business and technology managers from 31 countries, representing 63 financial institutions, about their current state of AI adoption and their future plans and challenges
I. CURRENT STATE
By far the most common type of AI project in production or development is using machine learning and predictive models to support decision-making.
Deep learning or machine learning-based models were the next most common type of AI being implemented by survey respondents.
Followed closely by organisations that have become confident enough in the accuracy of their models that they are using it in an automated fashion to make system and business decisions without human input
II. CURRENT AND FUTURE USE CASES
AI will play an important role in the short and medium term in areas as diverse as operational efficiency, risk management, fraud prevention and customer insights.
Compliance and financial crime have proven to be functions with a particularly compelling business case for the adoption of AI techniques. This is largely due to the large staff requirements and high false positive alert rates for non-AI driven approaches, as well as increases in regulatory oversight and penalties.
III. DATA READINESS AND OTHER CHALLENGES
From a business perspective, internal aggregation of disparate data sources into a single view of the customer has been achieved by 18% of respondents.
On a more technical level, 17% have implemented a data lake– a centralized repository for storing structured and unstructured data at any scale.
Skills availability is another issue.
IV. DEPLOYMENT APPROACHES, COST AND ROI
33% said they had built and deployed themselves with solely internal teams. But this group made up more than half of those with projects that failed to deliver any ROI.
The next most common deployment approach has been to create a platform for multiple use cases (20% of respondents) which shows desire to avoid the fragmentation of AI solutions and approaches that can occur when project initiation and approvals arrive from multiple sources within a large financial organisation.
Much smaller numbers are going live purely with end-to-end packaged solutions (10%) or cloud providers (8%).
29% of respondents had departmental allocation of budget for these projects in excess of USD. 1 million.
26% of respondents who had live deployments of any AI-driven project had seen strong ROI or outstanding results with payback within months.
AI is just a tool – a way of making better decisions based on data.
But the reality is that those decisions don’t become compelling until the organisation is implementing these intelligent decisions at scale.
The most common implementation of AI-assisted decision making (31%) is still human-in-the-loop.
And those organisations seeing the most success here are removing a lot of the data presentation usually associated with business intelligence approaches and only presenting to employees the areas of data and context they need to focus in on, appraise or enhance with more data before a final decision.
Our experience is that the more mature customers that focus on outcomes rather than technology achieve the best ROI – typically within a year.