Life sciences companies are grappling with how to account for the costs associated with increasingly prevalent artificial intelligence (AI) technologies, according to accounting firm Crowe. The challenge centers on adhering to U.S. Generally Accepted Accounting Principles (GAAP) as AI becomes integrated into clinical trials, data management, and analytics.
Finance leaders at both sponsor and contract research organizations (CROs) are focused on not only deploying AI effectively but also on mitigating risks, including identifying inaccuracies – often termed “hallucinations” – and verifying the reliability of AI-generated data, and processes. The financial reporting treatment of AI costs can vary significantly due to the complexities of existing accounting rules, according to Jennifer Dzierzak and Karen McDaniel, partners in Crowe’s audit and assurance group.
AI is being utilized across the clinical trial lifecycle, from initial planning and patient recruitment to data analysis and regulatory submissions. Determining whether costs related to AI development or implementation should be capitalized as an asset or expensed immediately depends on several factors, including the intended use of the software.
If the software is intended for sale, lease, or commercial marketing, accounting follows guidance under Accounting Standards Codification (ASC) 985, which requires the software to be technologically feasible before related costs can be capitalized. However, if the software is for internal use or service delivery, certain costs – such as coding and direct implementation expenses – can be capitalized during the application development stage, continuing until the software is ready for its intended use.
A further complication arises when internal-use applications are later offered externally. This transition requires careful consideration of reclassification and capitalization rules when moving between ASC 350-40 and ASC 985-20, a scenario common in AI development where projects often evolve without a clearly defined endpoint.
AI assets also present unique challenges compared to traditional software. They may deteriorate more rapidly due to factors like model drift, outdated training data, and evolving regulatory requirements, increasing the risk of impairment. Ongoing evaluations, guardrails, and monitoring for issues like hallucinations also demand additional resources, blurring the line between development and maintenance activities and complicating accounting treatment.
Determining whether costs relate to software, another intangible asset (like a process or database), or research and development requires careful judgment. Research and development costs are generally expensed as incurred under ASC 730, and most internally developed intangibles are expensed unless specific capitalization rules apply.
Accounting Standards Update (ASU) 2025-06, issued in September 2025, updates the guidance for internal-use software. Effective for fiscal years ending after December 15, 2027, with early adoption permitted, the update removes the project stage capitalization framework. It replaces it with two criteria for capitalization: management approval to complete the project and a probability that the software will be completed and used as intended. The level of capitalization achieved under the revised guidance can vary significantly depending on the level of development uncertainty. Crowe notes that traditional software capitalization timelines don’t always align with AI projects, and the “probable completion and use” criteria may be demanding to apply to work that is continually in process.
The proper treatment of these costs directly impacts a company’s financial position and performance reporting under GAAP. Expensing costs records them immediately as operating expenses, reducing current period net income. Capitalizing costs records them as assets, avoiding an immediate impact to earnings, with the cost recognized gradually through depreciation or amortization. This distinction also affects earnings before interest, taxes, depreciation, and amortization (EBITDA). capitalizing AI-related costs typically has a net-zero impact on EBITDA, while immediate expensing reduces it.
Cost classification also influences compliance with debt covenants, particularly those tied to operating income or EBITDA. Expensing reduces earnings and EBITDA, potentially risking noncompliance, while capitalization is EBITDA neutral. Accurate cost classification is therefore crucial for financial performance metrics, investor perception, and covenant compliance.
To ensure accurate accounting for AI, finance leaders must understand the technology’s purpose, intended use, and associated costs. Crowe recommends several steps, including understanding which accounting model applies, determining when to apply the guidance of ASU 2025-06, maintaining contemporaneous documentation, accurately tracking internal and third-party costs, supporting useful life and amortization estimates with evidence, monitoring for impairment indicators, and staying aligned with evolving accounting guidance. The firm suggests engaging technical accounting advisors for assistance, given the rapid advancement of AI technologies and evolving interpretations of existing standards.
