How should companies capitalise AI investments under IAS 38?

Full Explanation

IAS 38 provides the framework for capitalising internally developed intangible assets, including AI systems. The standard draws a critical distinction between the research phase (expensed) and the development phase (capitalised if criteria are met). For AI projects, this distinction maps onto the ML lifecycle but requires careful judgement. The research phase includes activities such as exploring potential use cases, evaluating algorithms, conducting proof-of-concept experiments, and performing initial data analysis. All costs in this phase — including data scientist salaries, compute resources, and data acquisition — must be expensed as incurred. Many AI projects never progress beyond this phase, which is precisely why IAS 38 requires expensing. The development phase begins when six criteria are simultaneously met: technical feasibility is established (the model achieves target performance metrics), the company intends to complete and use or sell the asset, the company has the ability to do so, probable future economic benefits can be demonstrated (the AI system will generate revenue or reduce costs), adequate technical and financial resources are available, and development expenditure can be reliably measured. For AI projects, the practical challenge is determining when technical feasibility is established. A model achieving 85% accuracy in testing may or may not be technically feasible depending on the use case — 85% accuracy for product recommendations may be commercially viable, while 85% accuracy for medical diagnosis may not. Companies should document the technical feasibility threshold for each AI project before development begins. Capitalisable costs in the development phase include: data engineering and preparation (post-feasibility), model training compute costs, integration and deployment engineering, and testing. Ongoing operational costs (inference compute, model monitoring, retraining) are generally expensed as incurred.

Related Glossary Terms

Related Questions