Embracing the Gen-AI revolution – summary recommendations for financial institutions

This series of articles focuses on how firms in the highly regulated world of financial services can best embrace the Generative AI (Gen-AI) revolution. In this final piece, we seek to pull together the key insights and recommendations from across the entire series, offering a comprehensive overview of key takeaways for those seeking to successfully adopt and scale Gen-AI within their organisations.

Given how situational these recommendations can be, a conscious effort has been made to steer away from dealing with specific use cases – that is where the hard work needs to be done by those embarking on this journey. Instead, we have focused on the crucial foundational concepts that apply to all FS firms, including adoption strategies, management of risk, scaling of infrastructure and the strategies financial institutions should employ to manage their large language models (LLMs).

At every step, we have also highlighted key considerations for decision-makers within these organisations to evaluate, informed by insights from the deployment of Deloitte’s own proprietary Gen-AI solution, PairD¹.

Focus on adoption and value delivery
 

Gen-AI has the potential to revolutionise financial services, but its success depends on strategic leadership and the careful reimagining of business processes. In our first article we focused on the important role of the CEO in championing AI adoption from the top down, supporting initiatives through active leadership, strategic partnerships and a focus on responsible AI development.

We believe that CEOs need to become tech-savvy and actively engage in AI strategy. The lessons from Deloitte’s own PairD deployment reinforce our view that senior leadership’s involvement is critical for driving organisation-wide adoption. This is especially true considering the vision and courage needed to rethink core processes in the ways required to unleash the full potential of Gen-AI. After all, these transformative technologies should not just automate existing processes, but rather reimagine them for end-to-end optimisation.

AI solutions must also be intuitive and transparent, since both are essential in helping users understand and trust the technology. Human-AI collaboration, along with tailored training can help ease the way for AI integration into workflows. In addition, effective training, and particularly role-specific education, can help accelerate Gen-AI adoption further, generating the early wins that will be needed to building momentum.

We recommend:

• Fostering active leadership involvement in AI deployment and ensuring full transparency in AI decision-making.
• Adopting a human-centred design approach for all Gen-AI tools, backed by the vision and support from leadership necessary for organisations to fully reimagine long-established business processes.
• Providing targeted, role-specific training and clear communication of Gen-AI benefits to all key stakeholders, from the boardroom to users, customers and shareholders.

Safety first
 

For financial institutions, responsible AI adoption must prioritise robust risk management, data privacy and security to mitigate the associated perils and potential downsides. In our second article we outlined strategies for building a proactive risk management framework, establishing data governance policies, and fostering transparency to build trust.

At its heart, Gen-AI risk management needs to be proactive and integrated from the ground up. Deloitte’s experience with PairD highlights the value of starting with lower-risk AI applications while gradually scaling and refining risk controls. In financial institutions, where sensitive data is handled frequently, stringent security and standardised frameworks are essential for all Gen-AI deployments. Likewise, maintaining full compliance with all relevant data regulations, such as GDPR, and building robust data quality, control and storage strategies are also key lessons.

Trust in AI tools, like PairD, can also be built through transparent decision-making processes and continuous user feedback. Explainability initiatives can help to illuminate the decision-making processes of Gen-AI, and institutions should embrace an iterative approach to risk management, learning from each application and adjusting governance frameworks accordingly to achieve peak compliance and scale with confidence.

We recommend:

• Embedding risk management into AI design from the outset, starting with lower risk use cases.
• Prioritising data privacy and ensuring AI systems fully comply with all relevant regulations, including GDPR.
• Building trust through transparency, continuous monitoring and user engagement.

Scaling the infrastructure
 

Scaling Gen-AI infrastructure is vital for organisations looking to fully capitalise on the technology’s potential. In the third article in our series, we provided a roadmap showing how financial institutions can scale AI infrastructure while managing their data, optimising costs and ensuring secure deployment.

Cloud infrastructure is both a powerful catalyst and, we believe, an essential ingredient for scaling Gen-AI solutions effectively. In particular, it offers institutions the flexibility needed to accommodate the growing computational demands of Gen-AI applications. High-quality and fully compliant data is essential for effective Gen-AI, requiring a centralised governance framework to ensure deployments are managed consistently. Additionally, organisations should consider hybrid options with on-prem GPU stacks for experimentation and for use-cases involving sensitive data, which cannot be hosted on cloud. Elsewhere, modular architectures can also help organisations optimise cost without compromising on performance.

We recommend:

• Leveraging cloud infrastructure for AI scalability and flexibility.
• Using an on-prem compute stack for experimentation and use cases involving sensitive data.
• Developing a robust data management strategy that ensures compliance and high data quality.
• Implementing centralised governance for AI deployments and ensuring cost-effective scaling through modular AI architectures.

Better to train, or to use out of the box?
 

Organisations must also decide between training custom LLMs or leveraging pre-trained models, balancing cost, performance and compliance. In our fourth article we explored some of the trade-offs accompanying this choice, offering strategies for businesses to consider as they seek to maintain performance and safeguard transparency.

Evaluating LLMs requires a multi-faceted approach, though, one that goes beyond accuracy to encompass fairness, reliability and transparency. Automated testing and detailed user feedback panels can be key enablers for firms looking to optimise the performance of their LLMs. This is especially true when it comes to addressing common LLM challenges, such as prompt brittleness and bias, both of which need to be actively managed. Through savvy prompt engineering efforts and regular audits of LLM outputs, firms can easily mitigate these risks.

We recommend:

• Carefully assessing the decision whether to build custom LLMs or leverage pre-trained models, ensuring their choice aligns with the long-term goals of the organisation.
• Implementing a clear upgrade strategy for fine-tuned models, backed up with comprehensive evaluation techniques.
• Addressing ‘classic’ LLM challenges proactively through prompt engineering, frequent audits and user feedback loops.

Ready to embrace Gen-AI?
 

Across this series of articles, our own experiences with PairD’s deployment have informed our thinking, consistently emphasising the importance of leadership, risk management, infrastructure scaling and strategic decision-making in successful Gen-AI adoption. As financial institutions look to embrace Gen-AI – either as a start, or a continuation of their implementation journeys – these core considerations must be weighed by all Gen-AI decision-makers.

By ensuring that CEOs and other senior leaders take an active role in driving AI adoption, aligning programmes with wider business strategy, overseeing responsible deployment and injecting the vision to not just automate tasks but to reimagine entire workflows, firms will achieve greater efficiency and value through Gen-AI. And, by designing AI systems with users in mind, providing clear communication on AI capabilities and limitations backed up with proactive, adaptable risk management frameworks to maintain data privacy and security, these organisations can also build the trust and transparency needed for these programmes to flourish.

Leveraging the latest cloud technologies and developing strong data management strategies, supported by centralised governance, will allow firms to achieve these benefits in ways that are both cost-effective and secure. And, by balancing the costs and benefits of custom LLMs versus their pre-trained equivalents, institutions can find solutions that work for them ‘where they are’, implementing systems that will address the complexities of their individual use cases over the long term, supported by planful fine-tuning, evaluation, upgrades and performance management.

By following these guidelines, financial institutions can responsibly and effectively embrace the Gen-AI revolution, positioning themselves to unlock new levels of innovation, efficiency and competitive advantage.

We wish you good luck and good fortune with your Gen-AI journey.

_____________________________________________________________________

References:

¹ Developed by Deloitte’s AI Institute, PairD is an internal Generative AI platform designed to help the firm’s people with day-to-day tasks, including drafting content, writing code and carrying out research safely and securely. The tool is also able to create project plans, give project management best practice advice and suggest task prioritisation.