We recently sat down with Shilin Chen, Co-founder and CEO of our AI partner, SmartClaim. We explored how AI is already changing the way R&D tax claims are being prepared and delivered, where accountability lies, and what firms should be considering to protect both themselves, and their clients.
Here are some of the key themes we explored, but by no means the entire conversation:
We began by asking our audience where they stand on AI today, specifically in the context of their R&D service. The responses from this first poll were as follows:
These results certainly reflect the feedback and conversations we’ve been having with our clients and wider audience across the sector for some time now – AI is firmly on the agenda, but curiosity is tempered with caution.
In a second poll, we asked about data privacy and security when using AI in client-facing work. The results were as follows:
These results were also no surprise. Client data from sectors such as government, defence, biotech, or patent-heavy industries (to name just some examples) can be extremely sensitive. The reputational risk of unintentionally releasing this information into the public domain is too high to ignore. Appropriate safeguards cannot be viewed as optional, they are essential.
The discussion turned to the practical limitations of AI. Even with its strengths, AI has well-documented weaknesses. Ask it to perform mathematical calculations and the results can potentially be unreliable. Push further into specialist territory, such as R&D tax legislation, and the risks multiply.
Generic AI may not be trained on the most recent rules, it may misinterpret ambiguous HMRC guidance, or worse, present confident but erroneous interpretations. That’s not a fault of the technology itself, but it’s a reminder that it cannot be trusted as the sole interpreter of complex legislation.
When it comes to drafting narratives, humans must always be in the loop. Even with safeguards that reduce the risk of hallucinations, small but critical details can still be missed, or misrepresented. In the context of tax, those small errors can have devastating consequences.
And crucially, AI is a mathematical machine, not a regulated adviser. It cannot be held legally responsible for the advice it generates. The accountability always rests with the adviser, which is why professional oversight is non-negotiable.
We naturally moved into a conversation about client relationships – another area where AI simply cannot compete. Yes, AI can transcribe meeting notes, or suggest follow-up questions, but only a human can pick up cues that a client may be uncertain, reframe a question, or identify when the right person (i.e. the competent professional) is / is not in the room.
AI only knows what it has been trained on. Advisers bring the judgement, intuition and empathy needed to ask the right questions, read between the lines, and guide their clients to stay within the boundaries of HMRC’s rules.
You don’t become an experienced R&D tax adviser simply by reading the theory. You get there by working on real claims, exploring the grey areas, and learning through practice. Over time, that experience is what allows you to balance risk, interpret ambiguity, and ultimately give sound advice.
The conversation shifted to the importance of guardrails. Shilin Chen, co-founder and CEO of SmartClaim, used a vivid analogy:
“Imagine the AI is like a super-powerful dragon… you want to make sure you keep it in a nice cage so it doesn’t run around doing things that are not intended!”
By design generic AI tools optimise for plausible, fluent answers. Without safeguards, they may fill gaps with misleading information, or present outputs that appear convincing but are dangerously inaccurate. In R&D tax, this positivity bias can be more dangerous than silence.
Purpose-built models, such as SmartClaim, trained on structured data, designed with compliance in mind, and subject to human oversight, are the best way to mitigate this risk.
Despite being a small sample, our polls seem to reflect the consensus within the sector. The question is no longer whether artificial intelligence (AI) will be adopted, but how. The technology is advancing quickly and many firms are already experimenting with it.
Looking five to ten years ahead, the conversation will have shifted. By then AI adoption will be assumed. What will separate firms is not whether they use AI, but how responsibly they apply it – with ethical standards, strong safeguards, and an approach that leverages benefits for both firms and their clients.
At WhisperClaims, we believe navigating complexity is not about shortcuts. It requires structure, responsibility, and a compliance-first framework.
You can watch the full event recording on demand by clicking below:
If you’d like to have a chat and see a demo of how WhisperClaims software and Advice Line can support your firm’s R&D tax preparation processes, why not book a demo?
Subscribe to email updates
© 2025 Wobbegong Technology Ltd (Registered number 10754811), trading as WhisperClaims.
WhisperClaims is a registered trademark of Wobbegong Technology Ltd (Trade Mark No.: UK00003360482).
