Generative AI in Finance: 5 Use Cases & Risk Controls
Generative AI is moving quickly into financial services, but finance is not an industry where organisations can simply connect a large language model to sensitive systems and see what happens. The opportunity is real. Generative AI can summarise complex documents, assist analysts, improve customer interactions, generate scenarios and reduce repetitive work.
But financial institutions also operate with sensitive personal data, regulated decisions and extremely low tolerance for errors.
That means the important question is no longer simply:
“Where can we use Generative AI?”
It is:
“Where can Generative AI create value while keeping the right controls around it?”
In Singapore, this balance is becoming increasingly important. MAS's 2024 review of banks found that Generative AI adoption was still largely focused on augmenting employees and improving productivity, rather than allowing models to make high-impact decisions independently.
1. Customer Service and Personalisation
One of the clearest use cases is helping financial institutions respond to customers more efficiently. A Generative AI assistant could summarise previous interactions before a service agent joins a conversation, help draft responses to common enquiries or explain financial information in simpler language. It can also help personalise educational content based on what a customer is trying to understand.
The value comes from reducing the amount of time employees spend searching through information or repeatedly writing similar responses. However, financial advice is different from answering a basic service question. If an AI-generated response influences someone's investment, insurance or borrowing decision, the potential consequences become much more serious.
A sensible approach is therefore to let AI support the employee, while the employee remains responsible for important customer-facing decisions.
2. Fraud and Risk Investigation
Generative AI can also help risk teams deal with large amounts of information.
An investigator reviewing a suspicious transaction may need to examine transaction histories, customer information, previous alerts and written case notes. Generative AI can help summarise this material and highlight information that deserves closer attention.
It can also assist with scenario generation. Teams can create synthetic examples of unusual behaviour to test whether existing fraud and risk processes respond appropriately.
But Generative AI should not be confused with the predictive models already used for fraud detection or credit scoring. Its strength may be in helping humans interpret, investigate and communicate complex information, rather than independently deciding whether someone is fraudulent or creditworthy.
The closer the system gets to making a decision that materially affects a customer, the stronger the validation and human oversight need to become.
3. Investment Research and Scenario Analysis
Financial analysts spend significant time reading information before they can begin forming a view. Generative AI can accelerate this process by summarising earnings calls, annual reports, regulatory filings, research notes and market news. Instead of manually reviewing hundreds of pages, an analyst can use AI to surface relevant themes before conducting deeper analysis.
It can also generate scenarios. For example, an investment team could ask how a portfolio might be affected under several combinations of interest-rate movements, currency changes or deteriorating economic conditions. The model can help structure possible scenarios for analysts to investigate further.
What it should not become is an unquestioned prediction engine. Financial markets are affected by variables that models may not have seen before. An articulate AI-generated explanation can sound convincing even when its assumptions are weak. The output therefore needs to remain an input into professional judgement, not a replacement for it.
4. Compliance and Documentation
Finance produces enormous amounts of documentation. Compliance teams need to review regulations, internal policies, onboarding material, transaction records, contracts and regulatory reports. This creates a natural opportunity for Generative AI.
A model might compare a new regulatory document against an existing policy and highlight sections that may require review. It could summarise lengthy documentation, draft an initial compliance narrative or extract information required for a case file.
A real Singapore example comes from Bank of Singapore, which described a 2024 proof-of-concept using Generative AI to support Source of Wealth write-ups during client onboarding and maintenance reviews. The project was designed to improve the efficiency and quality of a labour-intensive compliance process while retaining appropriate control over the final output.
This is a good example of where Generative AI can work well: assisting with documentation without removing accountability from the professional responsible for the case.
5. Financial Analysis and Reporting
Generative AI can also reduce the amount of manual work involved in turning financial information into something people can understand. A finance team could use it to create an initial narrative explaining why actual performance differed from budget, summarise management reports or extract key themes from large volumes of business-unit commentary.
Instead of replacing the underlying financial model, Generative AI operates on top of the information to help people explore and communicate it. This distinction matters.
Numbers such as revenue, margins or forecasts should still come from controlled financial systems and validated models. The language model can assist with explanation, but it should not quietly invent the numbers it is supposed to analyse.
The Main Risk: AI Can Sound Right When It Is Wrong
Generative AI creates a particular problem for financial institutions because its output can appear highly convincing. A fabricated regulatory requirement, an incorrect financial explanation or a plausible but inaccurate customer response may be more dangerous than an obvious error because the user has less reason to question it.
MAS's AI model-risk work specifically identifies concerns including hallucinations, unexpected behaviour, explainability challenges and the complexity of testing Generative AI systems. The control question is therefore not simply whether the model works. It is how the organisation detects when it does not.
Protect the Data Before Thinking About the Model
Financial institutions handle some of the most sensitive categories of customer information.
Teams need to understand what data enters the AI system, whether that information is actually required, where it is processed and whether a third-party provider retains or uses it.
Singapore's PDPC guidance on AI makes clear that organisations using personal data to develop or deploy AI remain responsible for complying with the PDPA.
This means teams should not casually copy customer records into a public AI tool simply because the output would be convenient. The data architecture needs to be designed before the use case is scaled.
Human Oversight Should Match the Consequence
Not every AI use case needs the same level of governance. An internal tool that summarises a meeting carries a very different level of risk from a system supporting credit, investment or compliance decisions.
For low-risk applications, occasional review may be sufficient. As the impact on customers or financial outcomes increases, stronger validation, approval and escalation processes become necessary.
A useful principle is:
Higher consequence → stronger human control
This is consistent with the risk-based direction of Singapore's AI governance approach. MAS's AI model-risk work emphasises governance, model identification, validation, monitoring and proportional controls across the AI lifecycle.
Governance Must Cover the Whole AI Lifecycle
Responsible AI governance should begin before deployment. Someone needs to own the use case. The organisation needs to understand what model is being used, what information it can access, what the expected output is and what happens when something goes wrong.
Testing should include normal situations as well as difficult ones: misleading prompts, incomplete information, unusual cases and attempts to manipulate the model.
Monitoring must continue after launch because AI systems, surrounding data and business processes can change.
In 2026, MAS also announced the second phase of Project MindForge, producing an AI Risk Management Toolkit covering traditional AI, Generative AI and emerging agentic AI systems. The direction is increasingly towards turning high-level responsible-AI principles into practical operational controls.
Start With the Use Case, Not the Technology
The biggest mistake organisations can make is beginning with:
“We need to use Generative AI.”
A better starting point is identifying a real business problem. Perhaps analysts spend hours summarising documents. Compliance teams repeatedly write similar case narratives. Service agents waste time finding information across multiple systems. Those are problems worth exploring.
Then teams can ask whether Generative AI is actually the right intervention and what level of risk comes with using it.
The process becomes:
Business Problem → AI Use Case → Risk Assessment → Prototype → Test → Control → Scale
That order prevents technology enthusiasm from overtaking business judgement.
How Emerge Creatives Approaches AI Innovation
At Emerge Creatives, AI innovation starts by connecting the technology to a meaningful business need before exploring implementation. Teams need to understand where AI creates genuine value, what role humans continue to play and what could go wrong if the system behaves differently from expected. This means innovation and governance should not happen separately. The strongest AI concepts are designed with their controls from the beginning.
Conclusion
Generative AI has real potential across financial services, from customer support and risk investigation to investment research, compliance and financial analysis. But finance also demonstrates why AI adoption cannot be separated from governance.
The more consequential the use case becomes, the more important data controls, model validation, monitoring and human oversight become. The objective should not be maximum automation.
It should be the right level of AI assistance for the right problem, with controls proportionate to the consequences if the system gets it wrong. That is what turns Generative AI from an impressive experiment into something a financial institution can responsibly use.
Last Words
I will be adding more articles on Design Thinking, Strategy and Innovation throughout the year. Articles of these 5 Step Action Plan and Modern Soft Skills will be added periodically to give my readers a broader insights to how to crush complex problems, overcome future challenges and spot AI opportunities.
Check out more articles via my blog: https://www.emerge-creatives.com/blog-1
Follow me on social media:
Instagram- @designthinkersgroupdtv: https://www.instagram.com/designthinkersgroupdtv?igsh=ajg2NGplbXJ1cWU4&utm_source=qr
About the Author
Daniel Ling is a regional ex-Design Leader turned educator, and business owner of Emerge Creatives, an registered SSG training provider (RTP) to deliver modern soft skills to professionals through Design Thinking, Business Strategy, and AI Innovation.
With over 15 years of experience in the financial and e-commerce tech industries- including key leadership roles at Lazada, NTUC Income, OCBC, and DBS- Daniel has led cross-regional design teams, built design functions from the ground up, and spearheaded large-scale transformation initiatives. But beyond industry success,
Daniel has reinvented himself as a “designer in a business suit”- equally fluent in creative strategy and commercial impact.
How to sign up for his WSQ Certification Courses
Advanced AI Design Thinking:
Entrepreneurship Business Strategy:
AI Business Innovation:
Where to get Complete Design Thinking Guide for Successful Professionals
Download PDF Digital Copy here: https://payhip.com/b/hM4U
Purchase from Amazon here: https://www.amazon.com/dp/1514202735
Download eBook from Apple Store here: https://itunes.apple.com/us/book/complete-design-thinking-guide/id1022432207?ls=1&mt=11
Comments