Unlocking the benefits of GenAI for next-generation lending

A number of macroeconomic developments combined with major technological breakthroughs are ushering in a new era powered by GenAI in the lending sector. Financial Services Institutions (FSIs) must respond quickly to remain competitive in a market, where they’ll face challenges but also opportunities.

Recent reductions in interest rates have exerted pressure on the balance sheets of FSIs. Additionally, they face growing regulatory compliance burdens: for example, Basel III requires FSIs to have more stringent capital and liquidity buffers depending on the RWA, rating model approach, and portfolio composition. Moreover, consumers are increasingly expecting financial solutions that cater to their specific needs, such as tailored loans or green loans. Larger universal banks and wealth managers must particularly adapt by offering personalised products and experiences while balancing booking centre-specific requirements versus group-wide harmonisation. Furthermore, the rising popularity of FinTech, B2B and B2C platforms (see further information1), as well as private credit, continues to disrupt the market through innovative and competitive offerings. Tokenisation and digital journeys enable the use of more exotic items as collateral, such as digital currencies and luxury goods (e.g. art, watches, cars). These developments necessitate that FSIs adapt their internal credit operating models, to make use of more data. Our recent webinar 2 which included live voting results from over 150 experts has shown that 36% of participants plan to introduce GenAI in their lending business by end of this year.

As FSIs navigate this rapidly changing landscape, GenAI offers powerful new use cases to enhance operations and overcome many of the typical challenges. Over the next five years, its integration into lending processes will be a decisive factor in defining competitive positioning. A GenAI powered transformation will enable businesses to target three high-value benefits:

• Efficiency: Improving operational performance through enhanced productivity by doing more with less (e.g. handling large loan volumes and future growth with fewer credit underwriters, thereby increasing the profit margin, or enabling cheaper pricings for the end customer)
• Experience: Providing customised experiences for both customers and employees (e.g. providing a real-time self-service platform with ideal credit structure propositions to assist relationship managers or clients directly)
• Capabilities: Developing and enhancing digital and business capabilities (e.g. enhancing credit risk monitoring to identify loan losses earlier)

In this blog, we explore how GenAI will shape the future of lending through enhanced yet simplified credit processes, and how your organisation can unlock the potential benefits ahead of the competition.

Unlocking the potential of GenAI

Current lending and evaluation processes are often lengthy, involving several time-consuming manual steps, as well as repetitive exchanges between the client advisors and the credit office due to incomplete or inconsistent information in the loan application. Root causes include:

• Insufficient guidelines or defined KPIs to measure success against the strategic vision
• Time-consuming manual documentation and validation of collateral and customer information
• A highly manual process in the credit structuring and decision proposal leading to inefficiency
• Complex internal policies with incomplete or contradictory information, leaving room for differences in interpretation
• A high proportion of non-standard business due to the overcomplication of exception to policies in the creditworthiness evaluation
• Unclear separation of responsibilities (e.g. between the 1st and 2nd line of defence, unclear decision authorities)
• Manual aggregation of data and analysis for reporting, and limitations in data quality to adequately measure risks and monitor complex products
• Consequently, no capability for using available data for enhancing risk-based pricing in line with the credit risk appetite, optimising capital structure and using of stress testing and scenario modelling
• Legacy IT architecture preventing integration with external partners and adjacent internal processes

GenAI can transform the lending process to become more efficient and scalable, significantly reducing processing time across the value chain, from client acquisition to loan approval and monitoring. This transformation will benefit all parties involved, from clients to the risk functions. At Deloitte, we have a large library available with over 100 proven use cases (see Figure 1).

Figure 1: Selected use cases across a traditional end-to-end lending process

GenAI capabilities are typically applied to a discrete process to resolve a specific challenge. Combining practical use cases that collectively enhance and accelerate an end-to-end business process provides compounded value, as the output from one use case can act as input for the next. At Deloitte we refer to this as the String of Pearls, chains of related processes which can be re-imagined end-to-end to deliver substantial additional value (see Figure 2). Therefore, applications of GenAI should be assessed based on their impact on the end-to-end process they support. Rather than tackling issues individually with GenAI (e.g. Relationship managers, Credit Operations, Credit Risk), the String of Pearls can create further value by providing an improved version of the as-is process, looking at the end-to-end process in combination with all practical use cases. This new journey may look entirely different with AI-enabled features (e.g. self-service, open banking, real-time pre-approval) previously not accessible (see Figure 3).

Figure 2: Example String of Pearls in lending

Adopting the String of Pearls leads to the implementation of multiple use cases and enhancements across the holistic operating model. Therefore, while the associated change costs are likely higher, the impact is also greater. First observations show short-term ~80% ROI for an individual use case in back-office functions, where efficiency improvements are most expected. Front- and mid-office functions observe a higher ROI of ~140%, as it is important to also include the top-line effects in the business case, e.g. from increasing the lead funnel, conversion rates and cross-selling. The overall ROI potential in applying the String of Pearls concept would be significantly higher than the sum the individual use cases.

Figure 3: Visionary end-to-end lending process greenfield based on GenAI use cases (Assistance from AI)

Deep dive into selected use cases

While AI can uplift the entire end-to-end lending process, we will now explore three selected use cases which offer the largest efficiency gain.

1. Client documentation analysis

The automated extraction and analysis of information from client documentation, along with profile screenings, enable GenAI to create pre-filled credit requests, thereby improving efficiency and accuracy. Traditionally, a credit analyst would spend a significant amount of time reviewing financial statements, tax returns, pay slips and business reports, identifying key data points and inputting them into a credit application. With AI-driven solutions and fine-tuned models, the process can be automated by extracting and classifying structured and unstructured data internally (e.g. transaction history) and externally (e.g. public credit scores) while cross-referencing it with internal documents and regulatory data bases to assess eligibility. Key metrics, such as income, liabilities and debt-to-income ratio are automatically mapped to the credit application, while ensuring alignment with the organisations’ risk appetite and internal credit policy (e.g. max LVs, collateral type). Although this was partially feasible in the past through trained Optical Character Recognition models (see further information here3), leveraging Large Language Models (LLM) unlocks the opportunity to handle not only multiple sources in parallel, but also interpret the figures (e.g. applying the credit notes or explanatory footnotes in annual reports to the data tables). A reduction in manual processing of up to ~50% per application will allow credit (risk) officers to focus on interpreting results, especially for complex cases that require deeper credit knowledge and analysis. This use case enables scalability in credit structuring and underwriting while increasing consistency with the internal risk appetite.

2. Collateral and counterparty rating

Collateral valuation is a key component of the end-to-end lending process, and GenAI unlocks opportunities to significantly improve the valuation methods. In the case of SME lending, unsecured credit lines cannot rely solely on traditional rating agency reports. However, GenAI can process non-traditional data sources (e.g. social media, news reports, public and investor sentiment), along with macro metrics (e.g. regional industry-specific market data). By combining this with the usual historical financial data and fundamental indicators, signs of financial stress or growth potential and stability of an SME can be identified. This will enable financial institutions to make informed decisions, manage their risk, and offer tailored, data-driven credit solutions to businesses. Additionally, GenAI offerings can be leveraged for property valuation. Currently, property valuations often rely on hedonic models provided by external agencies, which uses historical sales data, market trends and comparable property information to provide an estimate. GenAI can enhance this process by integrating more advanced data sources, such as satellite imagery (e.g. property condition, structural changes, potential greenhouse gas emissions) as well as identifying environmental factors such as flood risks, traffic, or proximity to key amenities. There is also increasing use of social media data (e.g. Snapchat) and cell phone data, including foot traffic and consumer characteristics, which are useful in assessing income-generating real estate, to better reflect current market conditions.

3. Monitoring

One of the first obvious use cases is the automation of internal and regulatory credit portfolio and credit risk reporting making up to 15% of back-office function efforts. Next, monitoring can be significantly enhanced with GenAI, particularly by using it for early warning indicators to predict defaults and improve loan loss provisions. By analysing vast amounts of structured and unstructured data (in combination with the two above use cases), GenAI can identify patterns and anomalies that signal financial distress or credit deterioration up to 12-18 months before manual grading would detect it or before the default occurs. This enables banks to take pre-emptive measures, such as restructuring loan terms or initiating targeted enhanced monitoring, which results in a ~20% loan loss reduction. This is especially important for SME lending, where banks often still rely on outdated annual reports. Furthermore, GenAI enhances loan loss provision models by generating more accurate forecasts of potential defaults, based on a combination of historical data, economic shifts, and borrower behaviour, ultimately reducing the volatility in loan loss provisions and resolutions. This enables up to ~10% RWA release depending on the internal modelling approach (which could help counterbalance the expected up to ~5% RWA increase from Basel III). Additionally, using the String of Pearls approach across the end-to-end lending process allows organisations to adapt their new request approval and risk-based pricing to the real-time risk in the portfolio, e.g. to hedge concentration risks.

The three biggest challenges and associated success factors

While GenAI offers numerous opportunities across the lending value chain, obstacles remain that need to be addressed to unlock its full potential. Three commonly encountered challenges in AI adoption, are data challenges, siloed governance and talent gaps and adaptability.

Data challenges

Data challenges stem from limitations in data privacy, quality and availability constraints. On one hand FSIs are expected to collect more and more data points (e.g. Basel III or Swiss Bankers Association with sustainability aspects). On the other hand, data privacy is critical, with stricter regulations coming into effect (e.g. GDPR). Additionally, the lack of evolving regulations does not allow for a clear understanding of future legal requirements around GenAI compared to AI. For example FINMA added a section focusing on Artificial Intelligence in its Risk Monitoring circular4. More is also to be expected from the EU AI Act5 and the Swiss Federal Council AI Regulation6. Furthermore, poor data quality, such as inconsistencies, missing records and incomplete or distorted historical data, limits data reliability and hinders the adoption of AI applications. 55% of surveyed FSI organisations have concerns about data quality at source, 73% about data privacy and security, and 82% about the quality of AI-generated data7. Finally, fragmented data landscapes across legacy systems, third-party sources or booking centres limits access to real-time insights and creates significant operational challenges. For example, a commonly observed challenge is data hosting cross-border due to banking privacy. Data challenges, especially regarding data quality, are beginning to improve, with the harmonisation of data processes through standardised governance and data models. An end-to-end data architecture, focusing on golden sources and combining data management, analysis and marketplaces significantly reduces operational complexity and enables quicker and smoother adoption of efficiency gains through use cases. Additionally, clear data governance ensures a higher level of traceability and explanations around decision making stemming from AI-powered insights. It is also important to note additionally that AI should not only be used to apply the data but can also be leveraged as a solution to identify these data challenges within the existing lending processes (e.g. inconsistencies or duplications between systems and reporting).

Siloed governance

While 72% of surveyed FSI organisations consider it critical to establish a dedicated governance around (Gen)AI, many still face challenges in aligning governance models across various functions7, resulting in lack of oversight, accountability, and decision-making for AI-initiatives. Siloed thinking and uncoordinated approaches due to insufficiently clear and enabled governance models mean that AI development often outpaces the underlying structures, leading to fragmented roles between technology, risk, and business teams. Success can be unlocked through the development of a flexible governance model that adapts to advances in AI, enabling innovation without hindering risk management. Best practices balance top-down governance and executive oversight with bottom-up management and flexibility. The most recent FINMA Guidance 08/20248 supports this view, claiming that especially operational risks (model, IT, cyber, legal, reputation) are to be considered with adequate controls and RCSA (see further information here9), bringing this closer in the risk taxonomy to the financial risks from the lending business itself.
GenAI use cases are often adopted in silos across organisations, meaning that similar features that could be applied at many points in the lending process are not utilised, or are built in parallel without applying the String of Pearls approach. The adoption of AI requires multiple functions to work together to identify and overcome new risks and challenges (e.g. the 1st line of defence integrating AI in their risk management, the 2nd line of defence monitoring and enforcing AI policies, while the 3rd line of defence building AI auditing capabilities). Similarly, this cross-functional collaboration and synergy needs to be observed across senior management levels within an organisation.

Talent gaps and adaptability

As with any major change, employees express some resistance to AI adoption due to changes in the nature of their tasks, combined with the fear of the unknown. A clear change management approach and communication framework is key to overcoming the resistance (see further information on importance of non-IT change management10). Additionally assessing the impact of the change on the work force (e.g. roles and responsibilities, skills) is crucial. While the integration of AI will eventually redefine target profiles, FSIs face challenges in identifying and training staff for AI-specific roles. 33% of surveyed FSI organisations, believe there is a clear lack of understanding and training around GenAI7. Training existing staff in AI tools requires a structured development programme to keep skills relevant. Therefore, FSIs must identify skills gaps and training needs in roles impacted by AI adoption across the organisation and invest in targeted training programmes to develop in-house talent and improve employee retention and satisfaction. Additionally, with the increasing focus on GenAI, the composition of credit and credit risk teams will change in the future – the role of credit underwriters inputting and calculating metrics from data will diminish as the need grows for fewer experienced credit officers to interpret the data and the results during the credit structuring process. As a starting point however, the mere appreciation of data transparency along the value chain and its benefits in daily use (e.g. live KPI dashboards) will prompt an important cultural shift towards accepting AI-based systems.

Next steps for the use of GenAI in lending

As we look to the future, the transformative potential of GenAI in FSI in general, and lending in particular, becomes increasingly evident. We foresee a future within the next five years where fully self-serviced real-time lending offerings for basic products are the norm, driven by seamless integration of internal and external information. This shift will not only streamline decision-making processes but also enhance the efficiency and accuracy of overall banking operations. With credit making up the major share of the business model of most banks, and in the case of mortgages often serve as an entry product for adjacent services, the changes are expected to be substantial, with over ~75% of a bank’s staff affected.
The ripple effects of this transformation will extend far beyond individual institutions. As GenAI continues to evolve, its capabilities could be leveraged to stabilise and optimise the financial market overall. By leveraging advanced analytics and predictive modelling, GenAI could provide unparalleled insights, fostering a more adaptive financial network resilient against credit crises as well as credit crunches. The adoption of GenAI in lending is not only a technological transformation but is also an industry-wide shift which completely redefines the financial industry. Eventually, not only will the bank push value streams for the consumer, but consumers themselves will create their own individualised AI-powered (credit) solutions, with the bank serving as the platform.

If you are interested in learning more about GenAI technology and its possible applications, please get in contact with us, e.g. to explore a “proof of concept” across the end-to-end lending value chain.

[1] The Swiss mortgage lending landscape in transformation - Platforms as one of three underlying drivers, 30/04/2021.
[2] Lombard Lending in Modern Banking: Key Insights from Deloitte’s Global Webinar, 20/06/2024.
[3] Smart process automation and analytics: How Optical Character Recognition can enhance productivity in core banking processes, 05/07/2021.
[4] FINMA Risk Monitoring (2023) - Artificial Intelligence in the Swiss financial market.
[5] Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence.
[6] AI regulation: Federal Council to ratify Council of Europe Convention.
[7] Deloitte FSI CDO Survey (with over 100 participants), 2024.
[8] FINMA guidance on governance and risk management when using artificial intelligence, 18/12/2024.
[9] Banking's evolving risk landscape: The case for smart internal controls, 14/01/2025.
[10] Navigating tech-enabled transformation of core banking processes| Part 3: The importance of non-IT change management, 14/06/2024.

Authors

• David Klidjian - Partner, Industry Solutions Banking
• Marc Beierschoder - Partner, AI & Data
• Eric Gutzwiller - Director, Industry Solutions Banking
• Matilde Pearce - Assistant Manager, Industry Solutions Banking