Responsible AI in Finance: A path to opportunity

Today’s leading CFOs are turning disruption into opportunities for competitive advantage and growth while improving the delivery of products and services to stakeholders. Organizations expect finance to serve as a true business partner in this drive to sustained competitiveness and growth. In the wake of these escalating demands, the finance function and CFO are turning to data-driven decisions powered by emerging and smart technologies.

1. Predicting and simulating future business scenarios using AI-based scenario planning models
2. Using robotic process automation to drive back-office automation and cost efficiency
3. Using machine learning models to identify M&A targets
4. “Optimizing” finance processes like order to cash (O2C), purchase to pay (P2P), record to report (R2R) using artificial intelligence (AI) predictive models
5. Improving data quality, traceability, and providing a single view of finance using AI tools
6. Driving finance transformation and migration to cloud

However, the adoption of digital technologies like AI and the growing interconnectedness among businesses create their own set of challenges and exposes finance to many of the following risks:

• Explainability: This is a limitation affecting many AI-based tools, models, and solutions, especially those using more advanced algorithms. A Scenario Modeling Engine makes recommendations which can improve an organization’s bottom line; however, the CFO and Planning head would want to understand the driving factors behind the recommendation before acting upon it.

• Bias and fairness: Bias could arise in AI-based workforce planning models that favor a certain race or gender. Biased outcomes are possible during the O2C process, specifically around AI-driven collection efforts or payment term setting favoring customers or suppliers based on their locations.

• Robustness: If AI-based models are not stable and robust, slight changes in underlying data could result in erroneous projections resulting in inconsistent variances which can negatively impact business or shareholder value.

• Brand/Reputation Risk: AI model biased outcomes can have a direct impact on staff, customers, vendors, and suppliers. As an example, the incorrect identification of expense fraud claims, or money laundering alerts can result in regulatory and reputational issues.

• Compliance Risk: The use of AI and natural language processing (NLP) algorithms to review legal contracts and regulatory requirements is increasing. However, absence of a robust governance framework and review process can affect the ability to track regulatory changes and exposes the organization to compliance risk.

How to Implement Responsible AI

We define Responsible AI as the practice of designing, building, and deploying AI in a manner that empowers employees and businesses and fairly impacts customers and society. Working with companies across the globe, we have developed a four-pillar framework to help them implement and adopt Responsible AI in the finance function.

Organizational

A key element is the introduction of a Responsible AI culture, a critical task for the CFO. Guiding principles for the adoption of Responsible AI should be elaborated and communicated throughout the finance function.

Leaders should develop a culture which motivates individuals to build Responsible and Ethical AI into their working DNA and empowers them to raise concerns and failures and act ethically. Organizations should work towards creating a diverse team working on AI-based solutions and tools since diversity brings different viewpoints and experiences which help improve the fairness of these tools and prevents bias.

Finally, the CFO and finance leaders should be aware of the challenges and risks AI presents, recognize the shortcomings associated with Responsible AI and be ready to address these.

Operational

Transparent, cross-domain, end-to-end enterprise governance is the starting point for a Responsible AI journey. AI governance can help put the Responsible AI principles into practice and monitor their effective implementation, key to mitigating any AI-generated compliance or reputational risk exposure.

AI-led decisions can impact the broader finance function which includes multiple units like treasury, tax, and controllership. A Finance AI Center of Excellence (CoE) with accountability to all finance units and with authority to implement the Responsible AI principles should be established.

Technical

Organizations should be able to deploy technical methods to make the AI models more transparent, fair, less biased, and more robust. Areas of focus should include:

Bias Mitigation: The organization should integrate algorithmic bias mitigation tools into its AI-modeling pipeline to eliminate data, model, and operational biases. Accenture has introduced an Algorithmic Assessment toolkit that automatically detects bias in AI algorithms and can help data scientists effectively mitigate the issue.

Explainability: Adding a layer of explainability to an AI model can help in interpreting the decisions it makes. A widely used explainability layer is the Shapley values, and can be used to measure the contribution of each feature in the decision made by the model. Other techniques include Permutation Importance or “LIME”.¹ Hybrid explainable modelling approaches can also address this problem by combining an explainable modelling approach with a complex black-box method.²

Robustness and Security: AI models should be tested for robustness by creating adversarial changes in the data and by doing extensive validation, maintenance and frequent monitoring of models and decisions taken. Security of AI processes should cover the entire AI solution and tool spectrum from data, AI models, system/platforms to the final reporting and communication.

Reputational

Every finance organization needs to adhere to certain regulatory guidelines in addition to its accountability to shareholders and other functions. Business decisions driven by unmonitored usage of AI could result in reputational damage for the entire firm and could lead to sanctions in extreme cases.

A proactive approach to establishing an organizational, operational, and technical framework would help companies reap the benefits of AI without any reputational risk exposure. This should also help organizations define their Responsible and Ethical AI mission and facilitate the embedding of guiding principles into the heart of the organization.

To find out more on the topic, please contact me or my colleague Ashish Aggarwal.

Source: CFO Now: Breakthrough speed for breakout value, 2021