Unlocking the Power of AI in Finance: Navigating Common Pitfalls

The Shift from Excel to AI

Traditionally, finance professionals have relied on Excel for modeling, but its limitations often lead to inaccuracies due to overgeneralization of data relationships. In contrast, AI can develop more accurate formulas based on historical data trends, combined with future-related assumptions, leading to improved modeling outcomes. Additionally, AI helps finance teams boost efficiency and effectiveness, especially in light of resource shortages caused by the pandemic.

Gaining a Competitive Edge with AI

Businesses can gain a competitive edge by being early adopters of AI. However, the success of AI applications is contingent on the quality of data and processes used. By understanding AI's potential and constraints, finance professionals can better manage AI applications and mitigate risks. Here are nine critical pitfalls to avoid when embarking on the AI journey in finance:

Data Pitfalls

1. Lack of Historical Data

AI models require ample data to learn correlations between different data points. Increased granularity necessitates more years of data. However, companies often purge data after seven years due to tax audit requirements or during system updates due to the cost of maintaining historical data. Insufficient historical data can negatively impact an AI model's accuracy and the number of future periods for forecasting. Finance teams must determine relevant data and bear the cost of storing it.

2. Poor Data Quality

Poor data quality leads to issues during AI implementation. Factors affecting data quality include missing data for chart of accounts (COA) members, late entries, top-down adjustments, and accruals. Additionally, COA hierarchy changes during divestitures and acquisitions impact the dataset's statistical properties.

3. Sparse Data

Finance often faces a data sparsity problem, with several COA members having zeros or missing values. Sparse data presents two challenges: AI models still perform calculations for these missing values, using up processing resources, and sparse datasets affect the efficacy and precision of outcomes.

4. Poor Curation of Training Data

Best practices recommend manually extracting and preparing data for AI models in the early phases of implementation. However, a lack of attention to detail during manual processing can lead to mistakes. Variation in data processing from month to month may cause AI models' training to deviate, resulting in erroneous outputs.

5. Data Silos

Data silos restrict access to data to a small number of people and develop from data distribution across many source systems. For example, finance professionals may have access to financial data in enterprise resource planning (ERP) systems but need to rely on different teams for inventory or marketing data. This creates friction in experimentation and continuous improvement of the AI pipeline. Data silos can also arise from organizational culture, system access, or historical processes, preventing finance teams from having a comprehensive understanding of the business.

Process Pitfalls

6. Accuracy as the Sole Metric for Measuring Efficacy

Accuracy is often chosen to gauge a model’s efficacy, but success shouldn't be judged solely on forecast accuracy. Instead, compare the model's forecasts to actuals, manually forecasted data, and the naïve forecast (applying the previous period’s forecast without adjustment). If AI forecasts are closer to actuals than manual and naïve forecasts, it's a win. Combining this with time savings creates significant business value for finance functions.

7. Premature Automation

Investing in automating data extracts, feeds, and building data warehouses before determining the right use case for AI is a recipe for failure. The goal should be to use AI to generate meaningful outcomes that support data-driven decision-making. Moreover, AI model building and deployment is cheaper than building data warehouses and automating integrations. Automation should follow once the AI path is determined and proven.

8. Lack of Awareness of AI’s Limitations

Not all financial use cases lend themselves well to AI modeling. Even with world-class AI models and software, the output depends on data signals and having the right drivers. Unreasonable efficacy standards prevent finance professionals from benefiting from incremental improvements in data-driven decision-making. Preliminary training in data science and machine learning (ML) helps finance professionals understand how AI and ML work and how to reap their benefits.

9. No Room for Experimentation

AI in finance is still in its early stages, and there is potential for trial and error until data points and procedures are codified. Unlike an ERP configuration, AI cannot currently be applied in a “big bang” manner. Consider data availability, manual modeling efforts, and data predictability when choosing use cases. An experimental approach to deploying AI can yield better outcomes than a waterfall strategy, as early AI stages will likely require significant improvement.

How AI Can Improve Forecasting

AI transformation requires finance teams to think unconventionally and continuously learn. Successfully implementing AI can reduce business risk through quicker, more effective decision-making.

By avoiding these common pitfalls and strategically integrating AI into their operations, finance professionals can enhance their forecasting capabilities, drive efficiencies, and gain a competitive edge in a rapidly evolving landscape.