More and more companies are buying and developing software that uses AI tools. We look at the kinds of associated costs and when these can be capitalised in accordance with IAS 38 Intangible Assets.
IAS 38 Intangible Assets provides a comprehensive framework for the recognition, measurement and disclosure of intangible assets. Artificial intelligence (AI) applications and software are an emerging consideration for many companies, and have a multi-faceted impact on the accounting profession.
AI refers to the development of computer systems that can perform tasks that usually require human intelligence, such as learning, problem-solving, and decision-making. It is an overarching term for various technologies that enable machines to simulate human-like cognitive abilities. Core technologies behind AI include:
- Machine learning (ML): algorithms that allow systems to learn from data sets
- Deep learning: a subset of ML using neural networks
- Natural language processing (NLP): enabling machines to understand and subsequently generate human languages
- Computer vision: allowing machines to interpret and process visual information
- Robotics: combining AI with mechanical systems to perform physical tasks.
Given their ability to enhance internal productivity and help with revenue generating activities, and their widespread application in numerous sectors, AI tools and applications are quickly becoming critical strategic assets for organisations.
As the level of investment in AI technologies grows, its impact on accounting is rapidly becoming an area of focus for industry experts and the professional services sector alike.
Costs – internally or externally generated?
So which costs incurred for AI technologies can be capitalised? IAS 38 provides guidance that covers both internally and externally generated intangible assets.
The development of AI technologies and applications may involve traditional software development costs (ie internal labour or external developer or contractor costs). But some applications may be more advanced and could incur additional types of cost not previously considered in scope. As well as these costs subject to IAS 38, companies need to carefully consider and identify costs that are tangible in nature. These would be within the remit of IAS 16 Property, plant and equipment.
Where an organisation has or is intending to develop AI applications and technologies, what kind of costs might be incurred? They could include:
- Software developer costs
- A mixture of internal labour costs such as employees and other staff, or external subcontractor labour costs related to development of the asset.
- Data acquisition costs
- AI technologies are often developed and used to access and process vast amounts of data. Developers will often incur costs to acquire large data reservoirs needed to develop models that perform various NLP tasks.
- Computational resources
- AI applications and technologies require computational resources, which can affect costs. These resources include high-performance CPUs (central processing units), GPUs (graphics processing units), or TPUs (tensor processing units) for training and running machine learning models. There are also large-scale storage systems for managing vast datasets. Cloud-based platforms like AWS, Google Cloud, and Azure offer scalable computing options, but costs can accumulate based on usage, especially for tasks like deep learning model training. On-premises infrastructure may involve upfront capital expenditure for servers, cooling systems, and maintenance.
- Storage costs
- Given the volume of data needed to train an AI application, entities may incur storage costs. Options range from a lease arrangement for hardware to the purchase of cloud storage services from third party providers.
- Installation fees
- The installation costs for AI applications and technologies can vary widely depending on the scale, complexity, and specific use. For SMEs initial costs may include purchasing or subscribing to AI software, acquiring hardware (such as GPUs or edge devices), and integrating the AI system with existing IT infrastructure. For larger enterprises or custom AI solutions, costs can escalate due to the need for specialised development.
- Configuration costs
- A critical element of the implementation budget, these costs include the customisation of AI models to suit specific business needs, involving the integration with existing systems, and the setup of data pipelines to ensure high quality input.
- Testing costs
- This involves validating the AI system’s performance through rigorous trials, which may include simulated environments, A/B testing, and real-world pilot programmes. Expenses can also arise from hiring or contracting data scientists, ML engineers, and quality assurance specialists. Iterative tuning and debugging may also be needed to optimise the model, further increasing costs. These phases are essential to ensure the AI solution functions correctly before full-scale deployment. There could also be identified inefficiencies and preliminary operating losses before the strategic asset achieves planned performance.
- Staff training
- There will be costs associated with staff training in order to use the strategic asset effectively.
Research and development phases – what to expect
AI applications and technologies are by their nature equivalent in function to software. This means the general considerations for the appropriateness and capitalisation of software-related costs under IAS 38 are applicable to AI applications and technologies too. So the expenditure must be a directly attributable cost of preparing the software/asset for its intended use by management.
One common pitfall when assessing whether an intangible asset qualifies for initial recognition is when an entity cannot identify when there is an identifiable asset that will generate future economic benefits. Another is the inability to reliably determine costs associated with the asset.
The key here is being able to distinguish between internal costs associated with the development of the AI application and technology, and costs of maintaining the asset or day-to-day operational running costs.
Research phase
Under IAS 38, to assess whether an internally generated asset meets the criteria for recognition, an entity must classify the generation of the asset into a research phase and a development phase.
Where an entity cannot easily separate the research phase and development phase of a project to generate an intangible asset, all expenditure is classified under the research phase only.
All expenditure incurred in the research phase of an internal AI application or technology project is expensed and charged through the statement of comprehensive income. This treatment under IAS 38 is driven by the idea that future economic benefits from a project in the research stage cannot yet be readily demonstrated because of uncertainties.
But AI projects often take longer than other software projects to demonstrate future economic benefits, given the greater risks surrounding a successful outcome. So while each project is different in scope and duration, AI-related projects are usually expected to have extended research phases compared to other intangible assets.
At the end of the planning phase and start of development, companies must carefully track and record all internal and external costs associated with building the AI system, including new procedures like employee training and time-tracking.
Development phase
Costs arising during the development phase of a project may be capitalised as an intangible asset where an entity can demonstrate all of the following:
- the technical feasibility of completing the intangible asset so that it will be available for use or sale.
- its intention to complete the intangible asset and use or sell it.
- its ability to use or sell the intangible asset.
- how the intangible asset will generate probable future economic benefits. Among other things, showing the existence of a market for the output of the intangible asset or the intangible asset itself or, if to be used internally, the usefulness of the intangible asset.
- the availability of adequate technical, financial and other resources to complete the development and to use or sell the intangible asset, ie a technical business plan or a lender’s indication of its willingness to fund a plan.
- its ability to measure the expenditure attributable to the intangible asset during its development.
To work out whether an asset will generate probable future economic benefits, an organisation must apply the principles set out in IAS 36 Impairment of Assets.
Relevant development activities may include:
- prototype design and testing – developing and refining AI models, algorithms and pilot systems
- AI integration – coding and embedding AI modules into existing platforms
- infrastructure development – building back-end systems to support AI tools
- UI/UX and visualisation – designing interfaces (eg dashboards or chatbots) and visual tools for AI outputs
- data preparation – creating training datasets or synthetic data for model development
- model validation – conducting simulations, stress tests and fine tuning for production readiness and deployment.
For AI-related projects and other intangible assets, the following costs of an internally generated asset cannot be capitalised, so should instead be incurred as expenditure:
- Selling, administrative and other general overhead expenditure
- Identified inefficiencies and initial operating losses
- Expenditure on training staff to operate the asset.
Useful economic life and amortisation
AI applications and related technologies are inherently susceptible to technological obsolescence over a short period of time and therefore should be treated as a finite intangible asset.
The accounting of AI-related intangible assets is based on their ‘useful life’, as judged by an entity. Any AI-related intangible asset is amortised over its useful life, and that period is reviewed annually by the entity per IAS 38. Amortisation periods are subject to change where the management amends the corresponding accounting estimate under IAS 8. In these cases the amortisation method is adjusted and accounted for prospectively to reflect the newly chosen amortisation period.
It’s important to consider a range of factors when assigning an appropriate useful economic life of an AI-related intangible asset. For example:
- How the entity plans to use the asset and whether another team could manage it better
- How long similar assets usually last and what public sources say about their typical lifespan
- Whether the asset might become outdated due to new technology, market changes, or for other reasons
- How stable the industry is and whether demand for the asset’s output is changing
- What competitors or new market players might do that could affect the asset’s value and other general market trends
- How much maintenance is needed to keep the asset working well and whether the company plans to do it
- How long the organisation can legally use the asset, including lease end dates or other restrictions
- Whether the asset’s useful life depends on the life of other assets the company owns.
In the same way as AI applications and technology, software-related intangible assets are susceptible to technological obsolescence and finite useful lives too. Given they are similar in nature to AI applications, they can give entities a steer as to appropriate useful lives for AI-related intangible assets.
Typical economic lives for software range from 3 to 10 years. But it’s important to remember that this period may vary depending on the nature of the AI tool, its integration into the business process, and the speed at which underlying models and programmes become outdated.
If you would like further information or advice on any of the issues raised in this article, please contact Imogen Massey or Calum McChrystal.
