The financial services sector is demonstrating great interest in AI, particularly in autonomous AI agents, to deliver agentic banking.
Recent research by OpenText reveals that 96% of banks are experimenting with agentic AI, however, only 19% are actually deploying it.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
Why are so few deployments taking place? The answer, according to the State of AI in Banking 2025 Report from MIT Sloan, is that 95% of enterprise AI pilots fail. Largely due to AI accessing data that is of poor quality, vague goals and organisational inertia.
Also, in the banking world there are concerns around using AI when needing to ensure compliance with strict regulations.
As a result, deployments of agentic AI at financial institutions are mainly occurring in the back office to aid labour intensive KYC and AML checks, including complex credit analysis for real-time fraud mitigation.
The opportunity for AI in financial services
Forward thinking banks are looking at creating AI-powered agentic advisors to autonomously manage money on behalf of customers, such as moving cash into higher yielding accounts. In the future, there’s also the possibility of AI agents representing the customer in negotiating with their banks’ AI agents to obtain the best financial products.
Data quality is key
To take this next step in creating autonomous AI agents that can effectively operate beyond the back office and add value to customers, while help ensure adherence to strict industry-wide regulations, financial institutions need to have to a solid data foundation. It also requires that AI agents are able to pull from a foundation of high quality, API – or machine-readable – data.
Access to clean data is often the main obstacle to banks as they attempt to train, deploy, scale and determine the return on investment (ROI) from their AI initiatives. Inaccurate data causes unreliable automation, ineffective personalisation, inaccurate recommendations, and an erosion of customer trust.
Frequent data problems
Common data issues faced by those in financial services include having out-of-date and duplicate data. This often results in the vast majority of data teams spending more than half of their time on data preparation – cleaning and structuring data so it’s ready for analysis – rather than insight generation.
Then there’s inconsistent or incomplete data, which can cause data accuracy and bias issues with AI recommendations.
Legacy and fragmented data is another big issue which leaves financial institutions struggling to define consistent business metrics across departments.
Finally, poorly structured and non-machine readable data is a significant problem, with many banks spending a significant amount of time per month manually reconciling data – comparing separate datasets to identify and correct discrepancies.
To have AI ready data it’s best practice for financial institutions to take five steps:
- Cleanse and update customer records on an ongoing basis by verifying name, address, email and phone number in batch and in real-time as data is captured. As part of this parse and structure data into a usable format. This will help to prevent biased results and actions.
- Match and merge duplicate records using an advanced fuzzy matching tool to create a single, accurate customer profile that you can trust. Data duplication is an issue with duplication rates of 10% to 30% not uncommon on customer databases. It happens when errors in contact data collection take place at different touchpoints, two departments merge their data, or when amalgamating datasets after a business acquisition.
- Enrich data with demographic, firmographic, geographic, social media, property attributes, and add missing email and phone information to support analytics, personalisation and omnichannel marketing efforts.
- Monitor and maintain your data across the entire data lifecycle to prevent bad data from entering your database, and keep it clean over time.
- Use well labelled data to ensure accuracy across mission critical AI applications, because high-quality data labelling helps maintain accuracy, consistency, and trust as systems grow. Particularly when errors can have financial, operational, legal, or safety consequences.
Have AI operating within a governed system
With concerns over meeting compliance requirements when using AI, rather than have it acting completely independently have AI operating inside a governed system where it has access to qualified data and is able to abide by agreed business rules. This is vital approach to take for those seeking to benefit from AI while operating in heavily regulated industries like banking.
In summary
Agentic AI is set to radically transform financial services and improve the customer experience. However, it relies on clean, trusted data to operate effectively. Inaccurate data leads AI to deliver hallucinations and bad outcomes. Therefore, have robust processes in place to deliver accurate data. This is non-negotiable for successful agentic banking operations that benefits all parties and provides a strong customer experience that supports growth.
Barley Laing, UK Managing Director at Melissa
