As a team which primarily supports claimants within the technology sector, the number one question we receive from our clients is “we’re working with AI, do we qualify for R&D credits?”. As you might expect, the answer is… it depends. As we see it there are 3 types of company conducting R&D in the AI space (for the purposes of this article, we’ll focus on the AI darling of the moment, Large Language Models, other flavours of AI are available!):
1. At the top end of the spectrum, we have the big names in big tech, Google, Meta, OpenAI or Anthropic etc, the companies conducting the fundamental research into the development of LLMs. It is reasonable to assume that these businesses are conducting R&D in its meaning for tax purposes.
2. At the opposite end, we have businesses who’ve added AI features to their products, perhaps they’ve augmented their customer service function with a chat bot or maybe conducting some minor prompt engineering. In these cases we’d assume that this type of activity fits more within the remit of routine product development and would be unlikely to qualify.
3. In the middle of these two extremes, we have a big grey area of businesses working in the domain of applied AI. They are (usually) not conducting the fundamental research into the design of AI models that big tech invests in, but equally their work goes beyond the implementation of tech developed by others. So, what happens if your business falls into this area?
Scenario 1: LLMs applied to specialist areas
Current LLMs are generalist models, they can give limited advice on a broad range of subject matter, however their capabilities to give detailed answers to questions in specialist domains is limited. Lets take the example of a LegalTech company developing an AI tool capable of analysing commercial contracts. While foundation models already exist, the company finds that outputs are generic and fail to interpret the specific wording of individual contracts and specialist legal terminology accurately. Their work to improve the performance of AI models operating in a specific domain would more likely to qualify.
Scenario 2: Reducing hallucinations in a generative AI platform
The most well-known issue with current LLMs is their ability to “hallucinate”, or make up seemingly plausible facts or context in response to a query. A SaaS provider develops an AI-powered knowledge assistant for customers. Early testing reveals that the LLM sometimes generates inaccurate responses, creating commercial and regulatory risks. As the company is seeking to tackle a core issue preventing widespread adoption of AI technologies, their work would be more likely to include qualifying R&D activity.
Scenario 3: Making AI more efficient without reducing performance
As AI platforms scale, the cost and computing power required to process requests can increase significantly. For example, a company providing an AI-powered customer support assistant found that growing usage was placing increasing demands on its infrastructure. The company sought to determine whether it could maintain the same quality and speed of responses while significantly reducing the computational resources required behind the scenes. As there was no readily available solution, the team carried out iterative experimentation with different AI architectures and optimisation techniques. This work would be more likely to qualify as R&D due to the technical risks involved in achieving equivalent performance more efficiently.
Key takeaways for founders and finance directors
- AI projects are not automatically eligible for R&D tax relief.
- Work that seeks to overcome issues faced by all businesses seeking to adopt AI are more likely to attract relief
- Having experts in AI at your business (or contracting with you) is essential for claiming
- Maintaining good records throughout the year on what the team have been working on, and the associated costs, will make your lives much easier at year end
How can HaysMac help?
HaysMac has a long-standing track record of supporting businesses across the technology sector, from early-stage start-ups and scale-ups to established enterprise vendors. We understand the unique challenges faced by technology businesses from bootstrapping to IPO/exit. Our multidisciplinary team provides a full range of advisory, tax and accounting services tailored to the sector, helping management teams and investors make informed decisions at every stage of the business lifecycle.
This includes specialist R&D tax relief support, where we work closely with founders, finance teams and technical leaders to identify qualifying activities, associated costs and prepare claims that stand up to scrutiny, whether from HMRC or in a deal room.
To discuss in more depth about how your business, or claim is affected by any of the above, reach out to Jack Williams at Jwilliams@haysmac.com
