Does Integrating AI Into a product Count as R&D?
AI is increasingly being incorporated into software platforms, diagnostic tools, decision-support systems and other products. For businesses investing in this development, an important tax question is whether any of the work may qualify as research and development.
This matters where a business is considering a claim under New Zealand’s Research and Development Tax Incentive (RDTI) or the R&D Tax Loss Credit (RDTLC). Both regimes can provide valuable cashflow support. Identifying eligible R&D activities is an important part of that assessment, although each regime also has its own unique requirements.
In the AI context, the R&D question is generally not whether using AI is innovative. It is whether developing or integrating AI into a wider product gives rise to a genuine scientific or technological uncertainty that the development team must systematically investigate.
That question can be particularly relevant where a business is building around a third-party AI model. Unlike conventional software, an AI model may operate as something of a “black box”. The business can provide an input and observe the output, but may have limited visibility over how the model reached its conclusion. Outputs may be probabilistic rather than repeatable, and the model itself may change over time.
Those features can create real engineering challenges. However, they do not mean every AI integration project will qualify for the RDTI or RDTLC. Eligibility remains dependent on the particular facts, including the nature of the uncertainty, what a competent professional could already determine, the systematic work undertaken to resolve it, and the separate requirements of the relevant regime.
Developing your own AI
Where a business is developing its own AI model or underlying AI technology, there is a greater potential for eligible R&D. That work may involve developing new algorithms, training methods, model architectures or other technical capabilities where the outcome cannot be determined in advance. However, developing an AI tool is not automatically eligible. The business must still identify a genuine scientific or technological uncertainty and use a systematic approach to resolve it.
Using AI as a tool is unlikely to be R&D
Using AI to help develop something is different from developing the AI itself or integrating it into a product’s functionality.
For example, a business might use an AI coding assistant to write code, troubleshoot errors, design a workflow or automate a process. The resulting product may be new and commercially innovative, but the underlying development does not become R&D merely because AI made it faster or easier.
Integrating AI into a new product may involve R&D
R&D can also arise where a business incorporates a pre-existing AI model into a new product or platform. In this case, the potential R&D lies in the experimental work required to make the AI operate effectively within the wider system.
The business may need to develop and test new methods to determine whether the integrated system can:
- produce sufficiently accurate and consistent results;
- detect or prevent unreliable and fabricated outputs;
- validate outputs against authoritative information;
- meet the required speed, capacity, security and privacy standards;
- operate reliably when inputs are incomplete, unusual or outside expected parameters; and
- detect and manage changes in the external AI model without compromising the product.
It is not necessary for the business to have developed the underlying AI model itself. The potentially eligible activity may instead be the work required to resolve the technological uncertainties created by integrating that model into the business’s own product.
Using an established AI model does not necessarily prevent an activity from qualifying. Equally, adding an AI feature to a new product does not mean that the integration will be R&D. The focus is on the specific technological uncertainty involved in integrating AI into the product, rather than simply whether the product uses AI.
A practical comparison
Consider a business that uses an AI assistant to create a new internal workflow. The AI helps map the process, generate code and automate individual steps. This is unlikely to be R&D merely because AI was used to produce the workflow.
Now consider a business developing an autonomous resource-allocation platform for a large service business. The platform must rapidly allocate workers and equipment while accounting for qualifications, availability, travel times, fatigue limits, contractual requirements and unexpected operational changes.
The business uses a mathematical constraint-optimisation engine and develops its own AI agent to interpret operational information, interact with the engine and dynamically manage changes. There is technical uncertainty about whether these technologies can be integrated to achieve the required speed, accuracy, reliability and compliance - and the answer cannot readily be determined from existing knowledge.
Even then, the project does not qualify simply because it is complicated, autonomous or uses AI. The potentially eligible activity is the systematic experimental work undertaken to resolve the identified technological uncertainty. Routine platform development, user-interface work, reporting and subsequent operation of the system may not form part of the eligible core R&D activity.
Focus the claim on the actual uncertainty
For RDTI or RDTLC purposes, the key question is:
What was technologically uncertain about making the AI operate as part of the product, and what systematic experiments were undertaken to resolve that uncertainty?
The answer should identify the specific technical problem, explain why existing knowledge was insufficient and show how the development team systematically tested possible solutions.
Disclaimer:
This article provides general information only and does not constitute tax advice. The RDTI and RDTLC rules are detailed, and their application depends on the particular claimant, activities, expenditure, contractual arrangements and supporting evidence. The examples are illustrative and do not confirm that any specific project or cost is eligible.
Businesses should obtain advice based on their own circumstances before making a claim.
Need Specialist R&D Advice? Contact Us.








