Chapter 2
Building an optimised data strategy that fuels AI solutions
Summary
The financial sector is actively embracing AI, with key challenges centred around integration with existing systems (56%), completeness and variety of data inputs (54%), and accuracy of data used as inputs (50%).
Firms are implementing both automated (50%) and hybrid (47%) processes to evaluate data readiness for AI models.
Top AI use cases include data quality checks (52%), regulatory compliance (38%), and reducing manual operations (37%).
AI solutions are predominantly benefiting data management & operations (54%), risk management (31%), and trading & portfolio management (26%).
What are your biggest challenges when using data-led AI solutions? (Respondents were asked to select all that apply)
Integration with existing systems
Completeness and variety of data as inputs
Accuracy of data used as inputs
A lack of internal expertise
Data governance
Compatibility
Data silos
Input vs output transparency
A lack of use cases
Scalability
"It comes down to GIGO or perhaps now that we don’t pull any punches the less pleasant SISO. Until we address the data silos and accelerate data integrate then we will continue to fall short of an accurate 360 barely even achieving a believable 180! We’ve had 2 years of GenAI POC’s where less than 20% have made it to production.
On reflection the last decade has been overwhelmed with average descriptive dashboard creation and rigid data warehouses limited largely to structured data only (Databricks and Snowflake are changing that!), coupled with humans thinking they’re capable of analysing a broad number of inputs and then “making decisions” (giving humans better information does not necessarily lead to better insights, humans cannot solve problems beyond 7 pieces of information – paraphrased Daniel Hulme Chief AI Officer WPP).
So, we first must address the persistent problem of data quality and accessibility along with organising and cataloguing structured, semi and unstructured data assets. Then we can address the use of AI for decision intelligence and opinion generation, but only then will GenAI have a chance of being trustworthy."
Lawrence Freeman, Director of Next Generation Tech, Kubrick Group
How do you evaluate data readiness for inputs into AI models?
Automated exception-management process
A combination of manual and automated evaluation
Manual approval process of data
What are the top three AI use cases for your organisation? (Respondents were asked to select three options)
Data quality checks of inbound information
Regulatory compliance
Reducing manual operations processes
Fraud detection
Low-code/no-code interaction for users in technical platforms
Risk assessments
Document digitisation
Algorithmic trading
Investment research
Chatbots and customer service assistants
Content creation
"The top and bottom responses stand out to be of interest to me here and one describes the other. The bottom responses both include reference to use of AI for opinionated response generation - one via chatbots and the other as content creation.
This concern tallies with a recent report sponsored by Databricks and The Economist in “Unlocking Enterprise AI: opportunities and strategies” where 37% of data execs blamed data quality, and 33% blamed concern for compliance for AI governance as reasons why few enterprises have fully productionised AI.
Surely this explains why the number 1 use case is to tackle this problem and use AI for data quality checks of inbound information. It's less “Turkeys voting for Christmas”, and more “Turkeys voting for better Turkey-food”. Hell, those Turkeys are even willing to go and help source their own high-quality food now!"
Lawrence Freeman, Director of Next Generation Tech, Kubrick Group
In your organisation, which functions are benefiting the most from implementing AI solutions? (Respondents were asked to select two options)
Data management & operations
Risk management
Trading & portfolio management
Trade operations
Cash management
Margin & collateral management
Reconciliations
Regulatory reporting
“Based on the variation in the survey responses, it’s clear that implementing AI at scale can surface so many different challenges. For many financial services firms, one of the first concerns is how to adopt new technology without compromising information security, compliance, and client data privacy. Then there's the issue of data quality; if your data isn't clean and well-managed, AI models will be flawed. Ensuring transparency and explainability is another element that will be crucial for trust and accountability. It's a balancing act between innovation and responsible implementation.”
Dmitry (Mitya) Miller, Senior Vice President, General Manager, Arcesium
