AI is moving fast. As organisations move from experimentation to enterprise-wide adoption, attention is increasingly turning to this question: how should the costs of AI be accounted for under IFRS?
AI spend can take many forms, from developing capabilities internally to accessing models, platforms and services from third parties. Understanding the substance of these arrangements is therefore critical to determining the appropriate accounting treatment.
Jayesh Gajare
Director, Accounting Advisory Services
- call
1. The AI invoice is changing
AI is changing not only how organisations operate, but also how they buy technology. Traditional software licence models are now being supplemented by consumption-based pricing, while organisations combine third-party models, proprietary data and internally developed applications.
AI and Large Language Model (‘LLM’) providers commonly charge customers based on consumption. A token is a unit used to measure consumption and, ultimately, the cost. Token consumption is typically generated through the following activities:
Access
Consuming AI capabilities provided by a third party.
- Employee productivity tools
- Prompting
- Experimentation
Build
Creating, configuring or developing AI-enabled solutions.
- Software development
- Data
- Testing
Operating and maintaining AI-enabled solutions after deployment into production.
- AI agents and workflows
- End-user interactions
- Production inference
The accounting outcome depends on the underlying activity, the contractual arrangements, the rights obtained and whether the organisation controls an identifiable resource or is receiving a service.
2. Where accounting judgements are likely to arise
Under IFRS, the accounting assessment may involve several areas:
What exactly is being assessed: the model, data, codebase, application, agent workflow or the broader solution?
An AI-enabled solution may combine several elements obtained or developed in different ways. The accounting outcome may depend on whether those elements are separate resources or form part of a single asset or service arrangement.
Defining the appropriate unit of account is therefore an important starting point that requires judgement. Without first identifying what is being assessed, it may be difficult to determine whether an asset exists, which accounting guidance applies and how the related expenditure should be treated.
Does the organisation control an identifiable resource, or is it receiving access to a supplier’s AI capability?
Understanding contractual rights is one of the critical factors in this assessment. The IFRIC agenda decision on cloud computing arrangement explains that access to software controlled by the supplier is generally a service rather than a software asset. However, activities performed alongside that service, such as developing additional code, may create a separate resource controlled by the organisation. Finance teams should therefore identify what has been created, who controls it and whether it is distinct from the supplier’s underlying technology.
Has the organisation moved from exploring what is possible to building an approved and technically feasible solution?
In practice, one of the biggest accounting challenges is determining when experimentation, prompt engineering, model evaluation and agent design move from research activities into a committed development programme capable of meeting the IAS 38 asset recognition criteria.
Determining when an AI project meets all the asset recognition criteria may require significant judgement. Demonstrating probable future economic benefits may be particularly challenging while a use case is still evolving.
Where has an asset been recognised, which expenditure is directly attributable to its development?
The challenge is determining which costs are directly attributable to creating the asset and which relate to experimentation, training, operation or ongoing service delivery.
In particular, agentic AI solutions can consume significant volumes of tokens through testing, model evaluation, autonomous workflows and live operation. The key question is whether tokens, computation, orchestration and related costs can be directly linked to an approved development activity capable of meeting the relevant recognition criteria.
What is the useful life, and how should subsequent expenditure be treated?
AI solutions generally continue to evolve after deployment. Finance teams need to determine whether expenditure relates to maintaining, monitoring and operating an existing solution or whether it creates additional capability.
The useful life should reflect factors such as expected use, technological change, changing data and dependence on third-party technology. Organisations also need to consider whether such changes suggest any impairment.
3. Questions for finance leaders
Answering these questions requires finance and technology teams to work together. Understanding the architecture, data, contractual arrangements, operating model and intended business outcomes is often important to reaching an appropriate accounting conclusion.
The questions below bring together the commercial, technical and accounting perspectives.
Understand the arrangement and control
Establish recognition, useful life and cost boundaries
Reassess after launch
4. Conclusion
As AI adoption accelerates, finance leaders will increasingly need to combine accounting judgement with a clear understanding of technology, data, governance and business objectives. Getting this right will enable organisations to move beyond experimentation and embed AI into finance in a controlled, scalable and value-driven way.
How can we help?
Apply the relevant IFRS requirements to token usage, compute, data, development and post-launch expenditure, while documenting significant judgements and accounting conclusions.
Cost visibility and monitoring
Map the AI cost stack, improve token and cloud-cost visibility and align usage information with projects and business outcomes.
Tax and sustainability perspectives
Bring in tax specialists to consider relevant tax and R&D relief implications, and sustainability specialists to consider energy use, emissions, water and workforce implications and compliance with reporting requirements.
Systems, processes and governance
Translate the accounting requirements into data, systems, controls, policies and governance that work in practice.
