Avoiding AI Pitfalls in Finance: The Key Role of High-Quality Data 

Back in 2018, Amazon had to discard a recruitment tool after realizing that it had developed a preference for male candidates in technical roles. This happened because the AI tool was trained on résumés the company had received over a period of 10 years, and as tech is a male-dominated industry, most of those résumés came from men. As a result, the system was unintentionally trained to choose male candidates over female ones.  

Fast forward six years, AI has developed significantly, and data continues to play a vital role in our daily lives. The importance of understanding the original source of data and the context and truth behind the data has never been greater – both in our daily lives and in business.

The risks of using bad data 

Low-quality, inappropriate or incomplete data can have a highly negative impact on the performance of AI agents. There are so many examples. Imagine a system designed to assist doctors in diagnosing diseases. If the model is trained on patient data that is incomplete—such as missing family medical history, medication details, or lifestyle information—it could misinterpret symptoms or fail to recognize important risk factors when going through the diagnostic process, leading to a misdiagnosis or incorrect treatments.   

As with human health, low-quality data can have serious repercussions on financial processes, leading to inaccuracies that harm businesses. For example, financial forecasting relies heavily on accurate historical data, including records of past invoices and payments. Low-quality data, such as missing transaction details or incorrect payment dates, can skew forecasts, resulting in overestimating cash flow or underestimating liabilities. Such poor decisions could affect budgeting, investment decisions, and strategic planning, ultimately hindering the company’s growth and stability.  

Consider the process workflows used, for example, in accounts payable or order management. Imagine the chaos if an invoice goes through an automated system with a decimal place incorrectly placed or a value missing. The outcome is likely to result in incorrect payments and unhappy suppliers, as well as too much money being spent for a given service or goods and ultimately a lack or loss of trust in the brand. Loss of trust in a brand ultimately results in loss of business in its simplest terms.

The financial impact of poor data 

Data plays a foundational role in all AI driven projects. As experts know, AI algorithms are powerful tools capable of processing large volumes of information, learning from patterns, and making autonomous decisions. However, their performance is strictly related to the information they receive. GIGO (Garbage In, Garbage Out)

At the same time, quality is far more important than quantity. It is useless to have 100 pages of incomplete data, while one page of clean, precise, and representative figures can make a pivotal difference in a decision.  

More than one-quarter of global data and analytics employees who claim poor data quality is an obstacle to data literacy at their organization estimate they lose more than $5 million annually due to poor data quality, with 7% reporting they lose $25 million or more, according to Forrester’s Data Culture And Literacy Survey, 2023.

For companies using AI systems, the cost of poor data quality is even higher, as low-quality data can lead to expensive model retraining, biased outputs, and legal liabilities. With AI becoming more integrated into sectors like finance, healthcare, retail, and manufacturing, addressing data quality could unlock substantial savings and drive better overall outcomes globally. 

In a recent interview "The Role of AI Agents in Intercompany Reconciliation: An interview with EY Advisors",Vish Dhingra, a Partner at EY’s Financial Accounting Advisory Services with over 20 years of experience in transformational finance, stressed the importance of data accuracy and master data management. As an example, he cited the issues multi-national companies have when it comes to intercompany reconciliation. Some companies have millions of unreconciled accounting items because of data mismatches, costing them untold millions of dollars.   

Key steps to achieving high quality data 

Quality data ensures higher accuracy and consistency in AI models. Just as a house needs a strong foundation, AI systems require accurate and comprehensive data to learn patterns and make informed predictions. Inaccurate or incomplete data can misguide AI algorithms, leading to flawed conclusions and unreliable outputs. At the speed AI operates, poor quality data could place you in a precarious position many times faster than a manual process. So, it is essential to pay attention to data quality and consistency.

Consistent data ensures that transactions are recorded uniformly, which reduces discrepancies and improves the accuracy of financial records. All of this is crucial for maintaining trust and reliable financial reporting. When AI algorithms use high quality data, they can function optimally and are able to identify anomalies and patterns with far greater precision. For example, an AI agent would be able to flag a duplicate invoice entry or detect overcharging from a vendor based on historical data patterns. This in turn enables more efficient processing, faster identification of issues, more reliable outputs, and ultimately streamlines the accounts payable process.  

Besides improving efficiency and accuracy, high-quality consistent data also enables enhanced decision-making capabilities. Organizations are able to make informed decisions based on reliable data and accurate, up-to-date information. For example, with the information provided by an AI agent, a company might be able to negotiate better payment terms with vendors or optimize management of their cash flow. Having higher quality data also minimizes the risk of errors that can result in financial losses or compliance issues. Better analytics, deeper insights into spending patterns, vendor performance and cash flow management are all benefits from using high-quality data. In essence, data integrity directly influences the effectiveness of sophisticated AI agents, ensuring they deliver maximum value and performance for accounts payable and order processing.

In conclusion, here are four key steps to help establish high data quality and consistency to optimize the outcome of your automation journey:  

1. Establish clear data standards by defining a single source of truth. Standardized data formats, naming conventions, and validation rules will make sure that data across various systems is compatible and interpretable, reducing errors and discrepancies  

2. Connect data across all systems to minimize discrepancies, especially if multiple platforms or departments are involved. Use APIs, middleware, or master data management solutions to centralize data and improve transparency across processes 

3. Develop automated checks to detect anomalies and validate accuracy in real-time to maintain accuracy as data flows through systems and, more importantly, stop data from being processed unnecessarily

4. Regularly monitor and optimize AI models in alignment with evolving financial processes. Regular performance evaluations will help ensure that your AI systems are learning from high-quality, consistent data 

Reach out today to learn how the team at Hypatos can help you ensure that your data is accurate and appropriate as you transform your financial processes with automation and agentic AI.  

Get in contact with us!