Navigating Five Generative AI Challenges for Financial Services

Craig Mackereth
EVP, Global Service Delivery
5 min read
Navigating Five Generative AI Challenges for Financial Services

Financial services have used AI in cybersecurity & anti-fraud operations for years, but Generative AI presents a new set of challenges; here’s how successful IT leaders can navigate them.

While AI isn’t new to financial services—most organizations use some form of AI to support their cybersecurity and anti-fraud operations—the U.S. Department of the Treasury reports that some financial institutions’ existing risk management frameworks may not be adequate to cover emerging AI technologies, such as Generative AI (GenAI). Yet deployment continues apace.

A recent Nvidia survey reports that 43% of financial institution respondents are using GenAI in their organization, and 46% already use large language models (LLMs), the AI deep learning technology behind GenAI and other innovations. And Grant Thornton’s latest CFO survey finds the portion of respondents using generative AI at an all-time high at 47% (Q1 2024).

GenAI poses significant challenges for financial service leaders who want to leverage its capabilities and stay ahead of the competition. As a recent guest on the AI in Business Podcast, I spoke on the topic of IT for Financial Services in the Age of GenAI; I encourage you to give it a listen. In the meantime, here’s a summary of my interview, including five GenAI challenges to be aware of, and how to successfully navigate them as you plot your AI strategy.

1. Stay flexible in your IT strategy

There are many players in the GenAI space, each with different approaches, models, and platforms. It is hard to predict which will emerge as dominant and which will best suit your needs. With such rapid development of this technology, your enterprise must have the flexibility to choose the best-fit AI vendor to deliver the best-fit AI solution at the right time in order to drive the best business outcomes.

A recent Bain & Company technology report notes that given few incumbent or leading providers, many early-stage companies will offer LLM-based data management, storage, and related apps which many eventually consolidate as larger platforms eventually provide such services in-house.

This is one of many reasons why you must be agile and flexible in your IT strategy and steer clear from being locked into a single vendor or solution. Instead, keep your options—and your resources—open for investment where it makes the most sense given your business objectives.

Tip: Experiment with different tools and pilot use cases to learn what works best for you. And while large ERP vendors could emerge as major AI players, there’s a lot of green field out there. Join successful organizations and leaders who follow their business-driven IT roadmap to stay ahead of the curve and confidently navigate winding roads.

2. Prioritize data security and quality

When you use GenAI, you need to feed it with data and content that may include your proprietary information, intellectual property, and customers’ personal financial information. In fact, Gartner predicts 80% of large enterprise finance teams will use internal AI platforms by 2026 for this process, according to a September 2023 press release.

In addition to data security, there’s the larger issue of data quality—which Forbes contributor Gene Marks dubs “the Achilles heel of AI that no one is talking about.” The adage “garbage in, garbage out” rings especially true for AI, as the quality of the output is a direct function of the quality of the input.

As the lifeblood of AI, quality data—measured in terms of accuracy, completeness, consistency, and timeliness—is paramount to AI effectiveness. Just as an analyst can’t build accurate forecasting models using inaccurate or outdated information, AI can’t deliver effective results using low-quality data.

Tip: Take caution in how data is securely shared and stored, and focus on the quality of data to ensure GenAI models deliver quality output. You also need clear agreements and policies with your providers and partners regarding the ownership and usage of your data and content.

3. Develop a robust QA and validation process

While GenAI has made progress since OpenAI unleashed ChatGPT in November 2022, the technology is far from perfect and can produce inaccurate, biased, or unethical results—often making up or hallucinating answers when it lacks the information needed for an accurate response. A leaderboard on GitHub that tracks how often LLMs introduce hallucinations when summarizing documents ranged recently from a low of almost 3% to a high of more than 16%, and these hallucinations can carry significant societal and business impacts.

For organizations using AI for forecasting, analysis, insights, and decision-making, AI hallucinations can present considerable financial and operational risk; strategies based on such output can lead to missed opportunities, misallocated resources, and missed objectives.

Successful organizations can address these challenges by establishing a quality assurance process, starting with a focus on data quality. Organizations can better monitor AI output for inaccuracies and inconsistencies through the timely involvement of human experts who help to oversee the models.

Tip: Develop a robust quality assurance and validation process to ensure that GenAI output meets your standards and expectations. You should also have a human-in-the-loop mechanism to monitor and correct the output when necessary. Be transparent and accountable for how GenAI affects your customers and stakeholders, and identify and correct inaccurate or inconsistent responses to better train the model while reducing the impact of hallucinations.

4. Focus on real-world business problems & ROI

This wave of GenAI is a building block, and it is incumbent on all business leaders to determine how to find real-world business problems to solve and that deliver a return on investment. You must start somewhere, and this is a technology that your business and IT teams want to experiment with.

Some of the more common real-world business needs that AI can help address involve predictive data analysis for actionable insights, automation to optimize business processes, and enhancements to customer and employee experience.

If GenAI is just viewed as something to play with and doesn’t really deliver anything beneficial, you will not likely get more shots at securing resources for pilots and projects. If a team implementing a project isn’t willing to commit that it will pay for itself in the first year and stand behind it, you should look much harder at that project.

Tip: Start by focusing on real-world business problems facing your organization which AI can help address; these typically involve data analysis and process automation. When assessing ROI, consider all potential benefits—increased efficiency and accuracy, lower costs, improved decision-making—not simply direct monetary gains. Apply extra scrutiny to project proposals that lack a commitment to break even in the first year; prioritize accordingly.

5. Learn from pilot experiments

If your teams make an ROI commitment but don’t succeed, that’s not necessarily a failure; lessons from such “failures” can prove beneficial when those learnings are applied to related projects. Besides, the nature of AI is that you will not be able to predict accurately every result you might get. In fact, according to a recent Harvard Business Review article, estimates show that 80% of AI projects fail.

Enterprise GenAI can be expensive to deploy, run, and maintain in operational production, so the proof is in the pilot. An initial project that might be seen as a failure could lead to great successes down the line; you might later regret tossing AI into the trash heap because it didn’t work just as you expected straight out of the gate.

Tip: Develop and implement AI pilot projects as an approach to learn about new technologies, pressure-test business value hypotheses, and experiment with implementation approaches. Ensure pilot use cases are ones in which AI can best address business needs. For pilots that prove out, determine if scale is feasible and apply learnings to related projects.

An expert partner can help fund & feed your AI strategy

The volume and velocity of innovative AI technologies continues to accelerate. And while the opportunities for AI are potentially vast, the challenges remain significant. You must be aware of the risks and how to manage them, but getting into the game is essential.

Financial service leaders looking to leverage GenAI capabilities and outpace the competition must maintain a flexible IT strategy, prioritize data security and quality, develop a robust QA and validation process, focus on real-world business problems and ROI, and learn from pilot experiments. And we can help.

Learn more: Discover how Rimini Street can help you plot a successful AI strategy by reallocating resources to further innovation, gain competitive advantage, and accelerate growth.