Raffles Consulting Services

EIS AI internal build and payroll costs: draw the boundary

Test whether EIS AI internal build costs, payroll and vendor services fall within the YA 2027 and YA 2028 AI adoption measure.

EIS AI internal build costs should not be assumed to qualify merely because employees are building or using artificial intelligence. The announced YA 2027 and YA 2028 measure focuses on qualifying expenditure for subscriptions or licences obtained from another person and qualifying AI business services. Internal payroll, general software development and mixed vendor work need a careful statutory and invoice-level screen.

Identify the acquired AI item

The IRAS support-measures annex describes a 400 per cent deduction on the first S$50,000 of qualifying AI expenditure per Year of Assessment. It refers to subscriptions or licences of an AI system and subscriptions, acquisitions or licences of a qualifying AI business service from another person.

Write down the supplier, product or service, contract period, users, business function and price. Distinguish the AI component from ordinary hosting, data storage, support, implementation and hardware. Marketing language that calls a product AI-enabled does not decide the tax treatment.

Separate EIS AI internal build costs from third-party expenditure

Create a project ledger with external subscriptions, external services, employee payroll, contractors, data work, integration, training and hardware. The announced wording is built around specified expenditure acquired from another person. Do not place internal salaries in the qualifying column unless the enacted law and current IRAS guidance clearly support that treatment.

Internal work still matters commercially. Record it as total project cost and use it to assess payback, but keep the tax claim schedule separate. This prevents a valid business investment case from becoming an unsupported enhanced-deduction claim.

Test a qualifying AI business service

Review what the vendor actually provides. A qualifying service needs to meet the statutory definition and any exclusions, not simply use an AI tool behind the scenes. Ask for a scope that identifies the AI function, access right, output, service period and price.

Where a supplier combines consulting, implementation and access to a model, obtain a reasonable price split. Keep contracts, service descriptions, invoices and acceptance evidence consistent. A retrospective internal allocation made only for the tax computation is weaker than a contemporaneous commercial breakdown.

Deal with mixed invoices

Map each invoice line to the contract and system access. Separate non-AI modules, ordinary maintenance, user training, hardware and unrelated professional work. If a reliable split is impossible, obtain clarification from the vendor or take a conservative position rather than estimating a favourable percentage.

Allocate expenditure to the correct basis period and Year of Assessment. Record credits, rebates, refunds and government support. The enhanced deduction applies to qualifying net expenditure, so finance must not claim amounts that another party ultimately bears.

Keep tax choice and grant support apart

The IRAS Enterprise Innovation Scheme page controls the broader EIS conditions. The AI activity has been announced for YA 2027 and YA 2028, and IRAS states that the cash payout option is not available for this AI measure. Model the deduction using the company’s tax position rather than describing it as cash funding.

If the same project receives a grant, reconcile the funded amount and any reduced cost base. Keep the grant approval, invoices, payments and tax schedule together. A grant and a tax deduction may have different eligible-cost definitions and evidence periods.

Review before filing

The business owner should confirm that the AI system or service was acquired and used for the stated function. Finance then confirms expenditure and net support. Before the return is filed, the tax reviewer should confirm the enacted provision, current IRAS guidance and Year of Assessment.

Retain system-access records and a short business-use note. Do not collect confidential prompts or personal data merely to prove use. The evidence should establish the commercial function and expenditure without creating a new security or privacy problem.

Test EIS AI internal build costs in a mixed project

A company licenses a third-party language model, pays a vendor to configure a customer-service tool and assigns two employees to build integrations. The licence may fall within the announced AI category if the legal definition is met. The vendor’s service needs its own test. EIS AI internal build costs such as employee salaries should remain outside the enhanced-deduction schedule unless the enacted rule and current guidance expressly include them.

Finance should still record every cost in the commercial project budget. The tax schedule can then show the licensed system, qualifying service if supported, excluded payroll, ordinary cloud cost and any grant offset. This preserves a true investment total without inflating the EIS claim.

Keep evidence proportionate

Retain the contract, licence terms, user or consumption report, vendor scope, invoices, payment records and calculation. Add a short description of the business function and the period in which the system was available for use. Where a service is mixed, keep the vendor’s contemporaneous price split or another supportable method.

Avoid saving confidential prompts, customer conversations or personal data solely for tax evidence. A controlled access log, deployment record and business-owner confirmation may prove use with less risk. The tax file should support the expenditure and function while respecting the company’s security and privacy duties.

Questions before commitment

Before filing, ask the vendor to identify exactly what was licensed or supplied. The business owner should confirm how it was used, while finance checks whether credits, bundled items or government support reduced the cost. The tax reviewer should verify the law and latest IRAS material in force for the relevant Year of Assessment, because the announced measure is new and detailed guidance may develop.

Final review checklist

  • AI system or service identified
  • Third-party expenditure separated from payroll
  • Mixed invoice split supported
  • Basis period confirmed
  • Grant offsets reconciled
  • No AI cash-payout assumption used

The safe approach is to keep EIS AI internal build costs visible in the project budget but outside the enhanced-deduction schedule until the law and current IRAS guidance support them. A clear boundary protects both the tax claim and the integrity of the AI investment case.

Read this with the EIS AI adoption overview and mixed-invoice guide. The business support decision guide is the pillar cornerstone.