A manufacturer may still own factories and machinery, but an increasing part of its competitive advantage could come from data, software, automated processes, proprietary models and specialist knowledge. A technology company may invest heavily in training an AI system without producing anything that looks like a traditional physical asset.
The spending is real. The expected benefits may also be real. The accounting outcome, however, is not always obvious.
This creates an important question for financial reporting. Do the financial statements show enough information about the resources driving modern businesses, or do valuable internally developed capabilities remain largely invisible?
For ACCA SBR candidates, this is exactly the type of current issue that can support a strong discussion about relevance, faithful representation, judgement and disclosure. It also provides a useful test of whether you can apply IAS 38 without simply repeating definitions.
Candidates looking to strengthen this type of analysis may benefit from working with an ACCA SBR tutor who focuses on application, professional judgement and clear written conclusions.
Why intangible assets matter more in the AI economy
A traditional business model is relatively easy to see in the financial statements.
A company buys a building, installs equipment and purchases vehicles. These assets are recognised, measured and depreciated. A reader can see where money has been invested and how that investment is being consumed.
The AI economy is different.
A business may spend millions developing datasets, refining algorithms, training employees, testing automated systems and integrating cloud-based tools. These activities may improve efficiency, reduce costs or support new revenue streams.
However, the resulting value may be spread across several connected resources rather than located in one clearly identifiable asset.
The data may be useful only when combined with specialist software. The software may depend on access to an external cloud platform. The AI model may need constant retraining. The business knowledge needed to operate the system may sit with employees who are free to leave.
This makes the accounting more difficult.
The problem is not simply that the assets are intangible. Accounting has dealt with patents, licences and software for years. The difficulty is deciding whether a separate resource exists, whether the company controls it and whether its cost can be measured reliably.
The starting point under IAS 38
An intangible asset is an identifiable non-monetary asset without physical substance.
That definition sounds broad enough to cover many modern resources. However, meeting the definition is only the beginning.
The entity must be able to identify the asset. The resource must either be separable from the business or arise from contractual or legal rights.
The entity must also control the resource. This means it has the power to obtain the expected economic benefits and restrict the access of others to those benefits.
Finally, the recognition criteria must be met. Future economic benefits must be probable and the cost of the asset must be measured reliably.
These principles remain useful. The difficulty is applying them to resources such as data, AI models, cloud systems and rapidly changing software.
Is an AI system one asset or several resources?
Imagine that a company develops an AI system to improve its pricing decisions.
The system contains purchased software, internally written code, customer data, external market information and an algorithm that has been trained over time. Employees also contribute specialist knowledge when selecting inputs, testing results and correcting errors.
What exactly is the asset?
It may be tempting to describe the entire system as one AI asset. However, accounting requires a more careful analysis.
The purchased software licence may be a separate intangible asset.
Internally developed code may qualify for recognition if the relevant development criteria are met.
Access to external software may be a service rather than an asset if the supplier controls the underlying platform.
Customer data may create economic benefits, but the company must consider whether the data is identifiable and controlled.
Staff knowledge is valuable, but the company will not normally control employees strongly enough to recognise their skills as an asset.
The answer therefore depends on the rights, contractual terms and substance of the arrangement. Calling something an AI project does not determine the accounting.
Why control becomes difficult
Control is one of the hardest issues in the AI economy.
A business may benefit from a resource without controlling the underlying asset.
For example, a company might subscribe to an AI platform that analyses its sales data. The company receives access to the service, but the supplier controls the software, updates the model and decides how the underlying platform operates.
The customer may control its own data, but it does not necessarily control the AI software.
This distinction matters because paying for access to a resource does not automatically create an intangible asset.
The same problem arises with cloud computing arrangements. A business may spend heavily configuring a supplier’s software for its own operations. However, configuration expenditure does not automatically create a resource controlled by the customer.
The accounting must focus on what the company actually controls.
Does it own separate code?
Can it obtain the benefits from that code and restrict others from accessing it?
Can the code operate independently of the supplier’s platform?
Does the contract provide only a right to receive future services?
These questions produce better analysis than simply stating that the expenditure relates to software.
Data may be valuable without being recognised
Data is often described as one of the most valuable resources in a modern business.
It can improve forecasting, personalise customer experiences, identify fraud and train AI models. In some organisations, the quality of the data may be more important than the underlying software.
However, describing data as valuable does not prove that it should be recognised as an intangible asset.
A company must determine whether there is an identifiable resource. It must also demonstrate control and measure the cost reliably.
Internally collected data creates particular difficulties.
The company may collect information through normal customer activity. The cost of gathering and maintaining that data may be mixed with the cost of running the wider business. It can therefore be difficult to separate the cost of creating a data resource from ordinary operating expenditure.
The data may also change continuously. Old information loses relevance, new information is added and the database is cleaned or reorganised. This makes it difficult to identify when the asset was created and what costs belong to it.
There may also be legal restrictions on how the data can be used. Privacy rules, customer consent and contractual obligations can affect the company’s ability to obtain benefits from the information.
A strong SBR answer should therefore avoid the simplistic conclusion that valuable data must be recognised. Value and accounting recognition are not the same thing.
The research and development divide
Internally generated intangible assets are divided into a research phase and a development phase.
Research expenditure is recognised as an expense when incurred. During this phase, the entity cannot demonstrate that an intangible asset exists that will generate probable future economic benefits.
Development expenditure may be capitalised, but only when the entity can demonstrate that all the recognition conditions have been met.
This distinction can become blurred in AI projects.
AI development is rarely a straight line. Teams experiment with different models, test datasets, reject ideas, adjust objectives and return to earlier stages. The project may move repeatedly between investigation and practical development.
This makes it harder to identify the exact point when the development criteria are satisfied.
A company cannot capitalise expenditure from the beginning simply because management hopes the project will succeed. It needs evidence that the project is technically feasible, that there is an intention and ability to complete it, and that probable future economic benefits will arise.
The entity must also have adequate resources to finish the project and be able to measure the expenditure attributable to the asset reliably.
The decision should be supported by evidence rather than optimism.
Agile development creates another challenge
Traditional development projects often have defined stages.
There may be an initial research phase, a formal approval decision and a clear development period. Costs incurred after approval can then be tracked against the recognised asset.
Agile software and AI development may not operate in this way.
New features are released continuously. Teams test small changes, gather user feedback and update the product. Some expenditure creates new functionality, while other expenditure simply maintains or improves the existing system.
This makes it difficult to distinguish between development of a new asset and expenditure that maintains existing benefits.
A company should not capitalise every salary and supplier invoice connected with an AI team. Management needs a reliable process for identifying qualifying development activity and separating it from research, maintenance, training and general operating expenditure.
The absence of a clear traditional project structure does not remove the IAS 38 requirements.
Training an AI model is not the same as training employees
The word training creates potential confusion.
Training employees is normally expensed because the business does not sufficiently control the future economic benefits arising from staff knowledge. Employees can leave and take their skills with them.
Training an AI model is different in substance, but the accounting still requires analysis.
The business may use data and computing power to refine a model. The resulting model might be controlled by the company and capable of generating future benefits.
However, management must still determine whether the expenditure creates or improves an identifiable asset. The company must also distinguish qualifying development expenditure from research, testing and routine operation.
The word training therefore does not determine the treatment.
The answer depends on what resource has been created, whether the company controls it and when the recognition criteria were met.
Purchased AI and internally developed AI can look very different
One of the long-standing tensions in intangible asset accounting is the difference between acquired and internally generated assets.
When a business acquires another company, identifiable intangible assets may be recognised separately from goodwill. These could include technology, customer relationships, brands and in-process development projects.
The acquired business may not previously have recognised all those resources in its own financial statements.
This can create a strange result.
A company that builds valuable technology internally may expense much of the expenditure. Another company that acquires that business may recognise the technology as an asset through the acquisition accounting.
The underlying economic resource may be similar, but the accounting presentation is different.
In the AI economy, this difference may become more visible because so much value is generated through internal investment in data, software and knowledge.
A strong discussion should recognise both sides.
Recognising more internally generated intangibles could improve information about investment in future capabilities. However, it could also introduce unreliable measurements and increase the risk that ordinary operating expenditure is capitalised to improve reported profit.
Why simply recognising more assets is not an easy answer
It is easy to criticise the existing rules for failing to recognise important resources.
The harder question is what should replace them.
Recognising an asset requires a sufficiently reliable measure. For many internally developed resources, cost may be difficult to separate from general business expenditure.
Fair value may be even harder to determine because there may be no active market for a unique AI model, dataset or internally developed process.
Forecasts may depend on uncertain assumptions about adoption, competition, regulation and technological change.
There is also a risk of management bias.
Capitalising more expenditure increases reported assets and reduces expenses in the current period. If the rules become too flexible, companies may be encouraged to classify unsuccessful or routine expenditure as investment.
The old rules may feel strained, but caution still serves an important purpose.
The useful role of disclosure
Recognition is not the only way to improve financial reporting.
Where important internally generated resources are not recognised, better disclosures may help users understand the company’s investment and business model.
Useful information could include the nature of significant expenditure, the objectives of major development programmes and the way management monitors progress.
Companies may also explain the risks surrounding AI investments, including technical uncertainty, dependence on third-party platforms, data quality and regulatory restrictions.
However, more disclosure is not automatically better disclosure.
A generic statement saying that AI is strategically important tells users very little. The information should explain what the business is investing in, how the investment is expected to create value and what could prevent those benefits from arising.
The disclosure should also be consistent with the financial statements. A company should not describe a major AI transformation as a valuable strategic investment while providing no explanation of where the associated expenditure appears in the accounts.
Questions that improve an SBR answer
When you see a scenario involving AI, software, data or another modern intangible resource, work through these questions:
- What specific resource may have been created?
- Is the resource identifiable through separability or legal rights?
- Does the company control the resource and restrict access by others?
- Is the arrangement an asset purchase, a licence or a service?
- Is the expenditure part of research, development, maintenance or training?
- When were the development recognition criteria satisfied?
- Can qualifying costs be measured separately and reliably?
- What useful disclosure is needed if significant expenditure is not recognised?
These questions move the answer away from generic commentary and towards professional analysis.
How to write the issue in an exam
A weak answer might say:
“AI is becoming more important and IAS 38 may need to change.”
That statement is broad and does not analyse the scenario.
A stronger answer would say:
“The company has incurred significant expenditure developing an AI pricing tool. Management must determine whether it controls an identifiable resource and whether the expenditure moved from research into a development phase that satisfies the IAS 38 recognition criteria.”
This version identifies the accounting decision.
The next paragraph should apply the facts.
If the project is still experimental and management cannot demonstrate technical feasibility or probable future benefits, the expenditure should be recognised as an expense.
If the system is technically feasible, approved for completion and expected to support measurable future benefits, qualifying development expenditure may be capitalised from the date the criteria are met.
The conclusion should then state the treatment clearly and identify any disclosure required.
Do not turn a current issue answer into a prediction
The examiner is unlikely to reward unsupported predictions about how accounting standards will change.
Candidates should not assume that every AI project will eventually be recognised as an asset. They should also avoid saying that IAS 38 no longer works.
A better approach is to explain the tension.
Modern businesses generate value through resources that may be difficult to identify, control and measure. Existing recognition requirements provide discipline and reduce the risk of unreliable assets. However, they can also result in significant expenditure on internally generated capabilities being recognised immediately as an expense.
That is a balanced discussion.
It shows that you understand both the relevance problem and the need for faithful representation.
Link the issue to financial performance
The accounting treatment affects more than the statement of financial position.
If expenditure is recognised as an expense immediately, current profit is lower. If qualifying development expenditure is capitalised, the immediate expense is reduced, but amortisation and possible impairment charges arise in later periods.
This can affect performance trends and management measures.
It may also affect comparability between companies.
One business may purchase an AI system and recognise an asset. Another may develop a similar capability internally but expense a larger proportion of the cost. Their financial statements could look different even when the underlying economic investment is similar.
A good SBR answer should explain this impact rather than discussing intangible assets in isolation.
Impairment becomes important quickly
Technology can lose value rapidly.
A model may become outdated, a competitor may release a better product or regulatory restrictions may reduce the available uses of data. The company may also discover that the expected cost savings or revenue growth will not be achieved.
Recognised intangible assets must therefore be reviewed for impairment when indicators arise.
Assets not yet available for use require particular attention because commercial success may still be uncertain.
Forecast cash flows should reflect current evidence rather than management enthusiasm. Assumptions about customer adoption, cost reductions and useful life need proper support.
AI may be a fashionable investment, but fashionable language should not protect an asset from impairment.
Why this is a strong SBR topic
This topic combines technical knowledge with professional judgement.
Candidates can discuss the definition and recognition of intangible assets, research and development expenditure, cloud arrangements, control, reliable measurement and impairment.
It also allows discussion of wider reporting qualities.
Recognising more resources might improve relevance. However, uncertain measurements could weaken faithful representation. Different treatments for acquired and internally generated assets can reduce comparability. Better disclosures may provide useful information where recognition is not appropriate.
This is exactly the type of connected analysis that distinguishes a strong SBR response from a list of memorised rules.
Candidates who want structured question practice and feedback can explore an ACCA SBR course designed to turn technical knowledge into applied, board-ready answers.
What to do next
Take a scenario involving an AI project and avoid beginning with the definition of an intangible asset.
Start by identifying the resource.
Then assess whether the company controls it, whether it is identifiable and whether the expenditure meets the recognition criteria. Separate research from development and distinguish owned software from access to a supplier’s platform.
Finish by explaining the effect on profit, assets and disclosure.
The old rules may feel strained because modern value is difficult to see and measure. That does not mean the principles should be ignored.
It means candidates and preparers must apply them with sharper judgement.