IAS 38 in the AI and SaaS Era: What Finance Should Capitalise, Expense and Document
Source: IASB — Staff paper AP17: Changes to definition and guidance, SaaS test case·Published July 2026
Spending on software subscriptions, implementation partners and AI development now dominates many technology budgets, yet the standard that decides whether this spending becomes an asset dates from a world of licences and servers. IAS 38 still works, but only if finance applies its tests precisely and documents the judgements. This article walks through the recognition criteria, the two IFRS Interpretations Committee agenda decisions on cloud computing, the accounting for internally developed AI, and what the IASB's intangible assets project, still in its research stage in October 2026, may change.
At a glance
- A SaaS subscription is usually a service contract: the customer does not control the supplier's software, so no intangible asset arises at contract inception (agenda decision, March 2019).
- Configuration and customisation costs are generally expensed, either when the work is performed or over the access period, unless they create code the customer controls (agenda decision, April 2021).
- Internally developed AI can be capitalised only in the development phase and only once all six criteria in IAS 38.57 are demonstrably met.
- The IASB's intangible assets project is exploring a rights-based model using SaaS as a test case; staff expect decisions in early 2027, and IAS 38 is unchanged for now.
The problem: digital spending that does not fit the old categories
A typical finance transformation programme in 2026 combines a cloud ERP subscription, a systems integrator, internal developers, data engineers and, increasingly, a team that builds or fine-tunes machine-learning models. The invoices look similar, but the accounting answers differ sharply. Some costs create an intangible asset that is amortised over several years; others hit profit or loss immediately; a third group becomes a prepayment released over the subscription term. The difference can move EBITDA by millions, which is why auditors, analysts and enforcement bodies pay close attention to it.
IAS 38 Intangible Assets does not mention cloud computing or artificial intelligence. The answers therefore come from its general principles, two agenda decisions of the IFRS Interpretations Committee and, eventually, whatever the IASB concludes in its current review of the standard.
The IAS 38 tests that decide every case
Before any cost can be capitalised, the item must meet the definition of an intangible asset: an identifiable, non-monetary asset without physical substance. Identifiability means it is separable or arises from contractual or legal rights. The entity must also control the resource, meaning it has the power to obtain the future economic benefits and to restrict others' access to them, and those benefits must be expected to flow to the entity.
The recognition criteria then require that future economic benefits are probable and that cost can be measured reliably. For internally generated intangibles, IAS 38 adds a strict separation:
- Research phase: all expenditure is expensed as incurred.
- Development phase: expenditure is capitalised only when the entity can demonstrate all six criteria in paragraph 57: technical feasibility of completion, intention to complete and use or sell, ability to use or sell, how the asset will generate probable future economic benefits, availability of adequate technical, financial and other resources, and the ability to measure the expenditure reliably.
If the entity cannot distinguish the research phase from the development phase, it treats all expenditure as research. Expenditure once expensed cannot later be reinstated as part of an asset's cost. Internally generated brands, customer lists and similar items are never recognised. Only directly attributable costs, such as employee costs and fees for services used to create the asset, are capitalised; training staff and general overheads are not.
SaaS and cloud: what the two agenda decisions say
March 2019: access to software hosted on the cloud: The IFRS Interpretations Committee considered a contract that gives a customer the right to receive access to a supplier's application software over a contract term. It concluded that such a contract, at its commencement, gives the customer neither a software lease under IFRS 16 nor an intangible asset under IAS 38, because the customer does not control the software. The arrangement is a service contract; fees are expensed as the service is received, with any payment in advance recognised as a prepayment.
April 2021: configuration or customisation costs: The follow-up question was how to treat the cost of configuring or customising the supplier's application software. The Committee, in an agenda decision finalised in April 2021, said:
- The customer often does not recognise an intangible asset, because it does not control the software being configured. An exception arises where the work creates additional code that the customer controls, which must then be assessed against IAS 38's recognition criteria.
- Otherwise the costs are expensed when the services are received. Applying IFRS 15 by analogy, if the configuration service is distinct from the access, the expense is recognised when the supplier performs it; if it is not distinct, it is recognised over the period of access, with an upfront payment carried as a prepayment.
- The same logic applies where a third party, not the SaaS supplier, performs the work.
- Entities should disclose their accounting policy where it is relevant to understanding the financial statements.
In practice, the most contested items are integrations, interfaces and data-migration tools. A bespoke interface that runs on the customer's own infrastructure and that the customer owns can be a separate intangible asset. A configuration setting inside the vendor's platform almost never is.
Sources: IFRS Interpretations Committee agenda decisions of March 2019 and April 2021.
Internally developed AI: research or development?
The same IAS 38 framework governs AI. The distinctions that matter are what the entity actually controls and in which phase the spending occurs.
- Consuming a hosted model through an API is economically similar to SaaS: the entity typically does not control the model, and usage fees are an operating expense.
- A licensed model deployed on the entity's own infrastructure, with rights to use and modify it for a defined term, can be an acquired intangible asset if the licence terms give the entity control.
- Building or fine-tuning a model starts in the research phase. Experiments with architectures, data samples and prompt strategies to find out whether a use case works at all are research and are expensed. Only once technical feasibility has been demonstrated, for example a model meeting predefined accuracy thresholds on validation data with an approved plan to put it into production, can subsequent directly attributable costs be capitalised.
- Data is a frequent grey area. Purchased data under a licence the entity controls may be an intangible asset; internally accumulated customer data resembles customer lists and is generally not recognised. Cleaning and labelling data can be a directly attributable development cost, but only for a specific asset in the development phase.
Measurement after recognition is equally important. AI models can become obsolete quickly, so useful lives should be short and supported by evidence. Assets not yet available for use must be tested for impairment at least annually under IAS 36, and a failed deployment or a superior vendor model are typical impairment indicators.
Where the IASB's intangible assets project stands
The IASB's project page, checked on 21 October 2026, shows the intangible assets project as active, with the stated aim of a comprehensive review of IAS 38. Work has begun through two initial streams: understanding what information investors need about recognised and unrecognised intangibles, and considering whether to update the definition, guidance and some recognition criteria, using newer types of intangibles as test cases.
In July 2026 the staff presented a rights-based model for a customer's accounting for intellectual property licensing contracts, with a SaaS arrangement as the test case. Under that model, a right delivered at a point in time, such as a traditional licence, would be an intangible asset that is amortised, while a right to access delivered over time, such as SaaS, could be treated differently. The Board continued the discussion in September 2026 without making decisions. The staff paper indicated further discussion and outreach in the second half of 2026 and expected IASB decisions on IAS 38 improvements in early 2027. The next listed milestone is a decision on project direction.
Nothing in the project has changed IAS 38 yet. Any amendment would require an exposure draft, comment period and an effective date, so current policies remain governed by the existing standard and agenda decisions.
Practical example: classifying a finance AI programme
Illustrative case, all amounts in EUR thousand
A group implements a cloud ERP and builds its own AI tool for automated invoice matching. Costs for the year:
- ERP subscription, 3 years paid upfront (900): prepayment, released at 300 per year. No intangible asset.
- Vendor configuration of ERP workflows, not distinct from access (240): prepayment released over the three-year access period.
- Integrator building a bespoke interface the group owns and runs on its own servers (180): assess as an intangible asset; capitalise if the IAS 38 criteria are met, amortise over its expected useful life.
- Data migration and staff training (120): expense as incurred.
- AI proof of concept, three model experiments (150): research phase, expense.
- After the steering committee approves the production build on documented accuracy results: engineering salaries and training compute (260): development phase; capitalise if all six criteria are met and documented.
Result: of total spending of 1,850, only 440 is potentially capitalised as intangible assets, 1,140 is initially carried as prepayments and 270 is expensed immediately. Without evidence of the point at which technical feasibility was achieved, the 260 would also have to be expensed.
What finance should document
- A contract analysis for each cloud or AI arrangement: who controls the software or model, and what rights the entity holds.
- A distinctness assessment for each configuration or customisation service, linked to the 2021 agenda decision.
- A dated record of the transition from research to development, with evidence for each IAS 38.57 criterion.
- Time-recording and cost allocation that separates directly attributable costs from training, overheads and maintenance.
- A useful-life rationale and an impairment review schedule for each capitalised asset.
- An accounting policy disclosure that explains the approach to SaaS and configuration costs.
Limits and scope
This article addresses IFRS financial statements. German commercial law differs: section 248(2) HGB allows, but does not require, capitalisation of self-created intangible assets, with distribution restrictions, and tax law prohibits it. The analysis does not cover the supplier's side (IFRS 15 revenue for software vendors), leases of hardware, or crypto-assets. The example is simplified and does not replace a contract-by-contract assessment. The description of the IASB project reflects staff papers and the project page; staff proposals are not Board decisions and may change.
Conclusion
IAS 38 gives clear answers for most digital spending, provided finance asks the right questions in the right order: does the entity control the resource, and in which phase was the cost incurred? Cloud subscriptions and most configuration costs end up in operating expenses or prepayments; internally built software and AI can be capitalised, but only with evidence of feasibility and directly attributable costs. The IASB's review may eventually reshape the treatment of rights to access, but until it reaches a decision, strong documentation is the best protection for both the numbers and the audit.
Key terms in this article
| Deutsch | English | فارسی |
|---|---|---|
| immaterieller Vermögenswert | Intangible asset | دارایی نامشهود |
| Aktivierung | Capitalisation | سرمایهای کردن (احتساب بهعنوان دارایی) |
| Forschungsphase | Research phase | مرحلهٔ پژوهش |
| Entwicklungsphase | Development phase | مرحلهٔ توسعه |
| technische Realisierbarkeit | Technical feasibility | امکانپذیری فنی |
| Verfügungsmacht | Control | کنترل |
| Cloud-Computing-Vereinbarung | Cloud computing arrangement | قرارداد رایانش ابری |
| Konfiguration und Anpassung | Configuration and customisation | پیکربندی و سفارشیسازی |
| aktiver Rechnungsabgrenzungsposten / Vorauszahlung | Prepayment | پیشپرداخت |
| Nutzungsdauer | Useful life | عمر مفید |
| Wertminderung | Impairment | کاهش ارزش |
| Agenda-Entscheidung | Agenda decision | تصمیم دستور کار (کمیتهٔ تفسیر) |
The terms were selected for direct relevance to the topic and for use in reports, meetings and professional documentation.
Sources
This article is an independent professional analysis based on the primary sources listed and does not reproduce their text.
- IFRS Foundation — IAS 38 supporting materials
- IFRS Interpretations Committee — Customer's right to receive access to the supplier's software hosted on the cloud (IAS 38), agenda decision · March 2019
- IFRS Foundation — IFRIC Update March 2021 (addendum: Configuration or customisation costs in a cloud computing arrangement) · April 2021
- IFRS Foundation — Work plan: Intangible Assets project
- IASB — Staff paper AP17: Changes to definition and guidance, SaaS test case · July 2026
Sources checked on: 21 October 2026