AI in Finance: Balancing Innovation with Responsibility

Behind the headlines about chatbots and image generators, a quieter AI revolution is taking place in finance departments across sectors. Here, the stakes are high. Financial systems demand accuracy, compliance, and security. In addition, these departments face a growing pressure to deliver more insights, as resources become more limited. 

Much of what is written on AI focuses on theoretical possibilities. In contrast, this article explores how real companies are putting AI to work in their finance departments, examining both the practical benefits and the challenges they're encountering. From creating effective guardrails to ensuring fairness and maintaining proper data governance, these examples reveal important patterns for anyone considering AI adoption in finance. 

Constructing Appropriate Guardrails 

The integration of AI into finance functions raises important questions about balancing automation with human oversight. 

Flight Centre implemented an AI expense auditing tool that reduced manual review requirements by directing human attention only to flagged exceptions. While this cut audit labour by 95% in the first quarter, the real value comes from focusing human expertise where it matters most: on complex cases requiring judgment rather than routine compliance checks. 

Caterpillar's finance group took a more cautious approach with their machine learning forecasting implementation. Their "shadow testing" methodology ran AI predictions in parallel with traditional processes, representing a prudent strategy for validating model accuracy before relying on the system's outputs for financial planning. 

These contrasting approaches reveal a key insight: the most effective AI guardrails in finance aren't one-size-fits-all. They must be calibrated to the specific risk profile of each function. High-stakes forecasting demands more cautious implementation than expense processing, where errors have lower organisational impact. This risk-based approach to AI guardrails will likely become standard practice as finance departments gain experience with these technologies. 

Addressing Fairness and Transparency 

As finance departments deploy AI, transparency in how systems operate becomes increasingly important. 

Unilever developed an AI tool for financial supplier analysis that goes beyond traditional cost metrics. Alongside these, it incorporates ESG performance and diversity criteria. While their system helps identify suppliers that align with corporate social values, it also introduces complex weighting decisions: how much should a diversity score influence supplier selection compared to pricing or payment terms? These trade-offs, once made subjectively by procurement teams, now require explicit quantification. 

Mott MacDonald tackled transparency by developing specific governance rules for their financial AI systems. Facing upcoming EU AI regulations, their framework sets explicit guidelines for how financial algorithms must document decisions, maintain audit trails, and allow for human review. This represents a shift from treating AI as a technical tool to managing it as a regulated financial process.  

The deeper insight here concerns the shift from implicit to explicit values in financial decision-making. Traditional finance processes often incorporated unwritten preferences and priorities. AI systems, however, demand that these values be codified, weighted, and made transparent. This forced articulation of organisational values may prove to be one of AI's most significant impacts on finance departments. It compels organisations to explicitly define what "fair" or "optimal" actually means in their particular context. Companies that navigate this transition effectively will likely gain both operational advantages and stronger alignment between financial practices and stated corporate values. 

These implementations demonstrate that successful AI in finance isn't just about algorithms. It's about creating systems where humans understand and trust the AI's recommendations and where accountability remains clear. 

Data Governance in Financial AI 

Finance departments handle particularly sensitive information, making data governance a central concern in AI implementation. 

Telefónica created a finance-specific AI governance model that requires financial algorithms to meet stricter data protection standards than other AI applications. Their approach gives finance teams veto power over AI systems that access financial data. However, innovation is encouraged within these guardrails. This balances the finance department's compliance responsibilities with the company's broader AI ambitions. 

DS Smith tackled the foundation of financial AI – data quality – when implementing their spend analytics system. Rather than accepting fragmented financial data, they first unified information from 12 separate ERP systems, standardizing supplier codes and accounting classifications. This investment in financial data infrastructure delivered a clear lesson: fixing underlying data issues pays dividends beyond the initial AI use case. 

These examples point to an emerging pattern in financial AI governance. The most successful implementations recognise that financial data requires specialised governance approaches beyond general AI policies. Companies seeing real benefits are creating finance-specific AI frameworks that address the unique regulatory, compliance and accuracy requirements of financial information, while still enabling the flexibility needed for innovation. This domain-specific governance approach may ultimately prove more effective than trying to create one-size-fits-all AI policies. 

The Measured Path Forward 

Unlike many business areas, finance can't afford to "move fast and break things." With regulatory compliance, fiscal accuracy, and risk management at stake, finance teams need AI approaches that enhance control rather than undermine it. 

What patterns emerge from these real-world examples? Successful AI in finance shares three key characteristics: 

• It automates predictable tasks while keeping humans in charge of exceptions and strategic decisions 
• It establishes governance that enables innovation without compromising compliance 
• It recognises that getting the data right is half the battle, and often the more difficult half 

This suggests a more pragmatic path forward than much of the AI hype implies. Rather than wholesale transformation, finance departments are finding value in targeted AI applications that solve specific problems while maintaining appropriate human oversight. 

At Trilateral Research, we help organisations develop Responsible AI approaches tailored to finance's unique requirements. Our focus isn't technology for its own sake, but practical solutions that enhance rather than replace human financial judgment.