From Activity-Based Costing to AI-Based Costing: The Evolution of Cost Transparency from Manual Models to Machine-Driven Insight

By Pedro San Martin posted 05-03-2025 12:12 PM

Recommend

Abstract

This article explores the evolution of cost transparency from traditional Activity-Based Costing (ABC) to AI-Based Costing, highlighting the limitations of manual models and the opportunities that artificial intelligence can enable. Focusing on practical applications within Oracle Enterprise Profitability and Cost Management (EPCM) Cloud, it compares these approaches across key dimensions such as transparency, precision, scalability, and operational complexity. The objective is to equip financial professionals and strategic cost leaders with insights for modernizing their cost systems in the era of data-driven decision-making.

1. The Foundations and Limits of ABC and TDABC

Activity-based costing (ABC), introduced in the 1980s, revolutionized cost accounting by assigning overhead costs based on activities and more relevant cost drivers (Kaplan & Cooper, 1998). However, ABC implementations often require substantial manual data collection, making them expensive and difficult to maintain (Drury, 2013). Furthermore, ABC can misinterpret fixed costs as variable and lack agility in rapidly evolving environments.

To address these issues, Time-Driven Activity-Based Costing (TDABC) was proposed by Kaplan and Anderson (2004). TDABC simplified ABC by requiring only two parameters: the cost per time unit of capacity and the time needed for an activity. This method reduces subjectivity and allows for faster updates. Still, TDABC relies on standard time estimates that may not reflect real-world variations and often omits idle time or inefficiencies (Everaert et al., 2008).

2. The Rise of AI-Based Costing

AI-Based Costing represents a paradigm shift. Rather than relying on static drivers and manually defined allocations, AI leverages machine learning to infer cost behavior patterns from data (Johnson & Smith, 2022). Algorithms such as clustering, decision trees, and regression models can discover complex, non-linear relationships between resources, activities, and cost objects.

Recent studies have validated the use of neural networks in reproducing ABC allocation logic with high accuracy (Morgan, 2022). Clustering algorithms, for instance, group cost centers based on shared consumption behaviors, while reinforcement learning adjusts cost assignments based on feedback loops. Benefits of AI-Based Costing include:

Higher Accuracy: Reduces errors from human estimation.
Continuous Learning: Updates cost logic dynamically as operations change.
Operational Efficiency: Minimizes manual input and data handling.

Challenges remain around transparency ("black box" effect), making Explainable AI (XAI) a critical complement (Ramírez, 2022). Additionally, data quality and supervision are essential to avoid bias or flawed models.

3. ABC vs. AI-Based Costing: Comparative Analysis

Dimension	ABC/TDABC	AI-Based Costing
Transparency	Explicit, auditable allocations (Kaplan & Anderson, 2004)	Requires XAI to avoid "black box" risk (Ramírez, 2022)
Precision	Limited by driver subjectivity	High, learns from actual behavior (Morgan, 2022)
Maintenance	Manual recalibration required	Self-adjusts with incoming data (Johnson & Smith, 2022)
Scalability	Complex in large models	Easily handles high-dimensional data
Speed	Weeks to update	Near real-time execution

4. Practical Application in Oracle EPCM Cloud

Oracle EPCM Cloud supports both ABC modeling and integration of advanced analytical techniques. It allows organizations to define cost pools, drivers, and multi-stage allocations (Oracle, 2023). AI integration is enabled through Oracle's advanced analytics capabilities, allowing analytical models to be incorporated into the costing process.

Oracle Cloud Docs (2023) describes how Oracle Cloud Infrastructure (OCI) Data Science and Integration Services enhance these capabilities by enabling data orchestration and analytical model deployment.

Case studies like AirAsia's illustrate Oracle EPCM's value in automating profitability analysis by flight and improving cost visibility (Oracle Customer Success, 2022). Consulting firms like PwC are actively integrating advanced analytical capabilities into Oracle EPM implementations for real-time intelligence.

5. Conclusion

The shift from ABC to AI-based costing reflects a broader transformation in finance—moving from static allocation to dynamic, data-driven decision support. Oracle EPCM Cloud is well-positioned to support this evolution, offering the structure of ABC with advanced intelligence.

Finance leaders should embrace hybrid models that start with proven ABC logic and evolve toward enhancements based on advanced analytics as data maturity increases.

About the Author:

Pedro San Martín is an Strategic Finance and Organizational Transformation expert, helping companies unlock sustainable competitive advantages through more innovative resource management. He can be reached at psanmartin@asher.company

References

Kaplan, R. S., & Cooper, R. (1998). Cost & Effect: Using Integrated Cost Systems to Drive Profitability and Performance. Harvard Business Review Press.

Kaplan, R. S., & Anderson, S. R. (2004). Time-Driven Activity-Based Costing. Harvard Business Review, 82(11), 131-138.

Drury, C. (2013). Management and Cost Accounting (9th ed.). Cengage Learning.

Everaert, P., Bruggeman, W., Sarens, G., Anderson, S. R., & Levant, Y. (2008). Cost modeling in logistics using time-driven ABC: Experiences from a wholesaler. International Journal of Physical Distribution & Logistics Management, 38(3), 172-191.

Johnson, M., & Smith, T. (2022). Machine Learning Applications in Cost Management: Current Practices and Future Directions. Journal of Applied Finance, 33(2), 78-92.

Morgan, A. (2022). Neural Networks in Cost Allocation: Empirical Evidence from Manufacturing. Journal of Management Accounting Research, 19(3), 112-130.

Ramírez, L. (2022). Transparency in AI-driven Financial Models: Challenges and Solutions. Financial Technology Today, 15(4), 45-57.

Oracle. (2023). Enterprise Profitability and Cost Management Cloud Overview. Oracle.com.

Oracle Cloud Docs. (2023). Data Science and Analytics Integration Guide. Retrieved from https://docs.oracle.com

Oracle Customer Success. (2022). AirAsia Optimizes Route Profitability with Oracle Cloud. Oracle.com.

2 comments

17 views

Permalink

https://myimanetwork.imanet.org/blogs/pedro-san-martin/2025/05/03/from-activity-based-costing-to-ai-based-costing-th

Comments

Pedro San Martin

05-05-2025 11:08 PM

Hi Larry

Thank you for your thoughtful question and for engaging with the article!

You're absolutely right—AI-based costing still relies on modeling. The key difference is how the models are built, maintained, and evolved. Let me break it down across the areas you mentioned:

1. Modeling Approaches in AI-Based Costing
Yes, AI relies on structured approaches, though they are data-driven rather than assumption-driven. Some of the most common include:

Clustering algorithms (e.g., K-Means, DBSCAN) to group cost objects or centers based on consumption patterns.
Supervised learning (e.g., regression, decision trees, neural networks) to model cost-driver relationships.
Reinforcement learning, in more mature environments, to iteratively optimize allocations based on feedback or reward functions.
Explainable AI (XAI) frameworks help interpret model logic to mitigate the “black box” issue, especially critical in finance.

2. What information is the AI asked to present or infer?
Models are typically trained using:

Operational drivers (e.g., time logs, volumes, units processed)
Financial data (GL entries, cost centers, service lines)
External variables (inflation, FX rates, utilization metrics)
Historical allocations (where available)
The goal is to infer behavioral patterns, detect anomalies, and recommend allocation rules or driver selections that reflect actual consumption.

3. Is it computing marginal revenue or cost?
Yes—AI models can approximate marginal cost curves by analyzing cost behavior at varying volume levels. When revenue data is layered in, marginal profitability (MR - MC) can be dynamically estimated for decision-making, especially useful in pricing, mix optimization, or capacity planning scenarios.

4. Is it distinguishing fixed and proportional costs?
Correct. Unlike traditional ABC, AI can learn elasticity from data—identifying whether costs scale linearly, step-wise, or not at all with volume. This enables a more accurate separation of fixed, semi-variable, and variable costs, reducing misclassification risks.

5. Is causality preserved and financial reporting aligned?
This is a major consideration. AI models can incorporate causal inference techniques (e.g., Granger causality, SHAP values in tree models) to improve explainability and defend model logic.
For financial reporting, models are constrained by compliance frameworks. In tools like Oracle EPCM, you can enforce accounting rules and thresholds to ensure auditability, traceability, and reconciliation with GL—so AI supports, rather than bypasses, regulatory accuracy.

📌 Bottom Line:
AI-Based Costing doesn’t eliminate modeling—it transforms it into a continuously learning, data-enriched, adaptive process. It complements traditional approaches by enhancing accuracy, scalability, and insight depth, while still needing sound design principles and oversight.

Happy to expand on any of these methods or share example model architectures if of interest!

Larry White

05-04-2025 09:53 AM

This is a very interesting article and concept, but even AI must use modeling approaches. I'm curious to know what those approaches are and what elements of information the AI is asked to present. Is it computing marginal revenue and cost? Is it identifying fixed and proportional costs? Is it appropriately applying causality or computing cost to meet financial reporting requirements?

Profitability & Cost Management Shared Interest Group