← All insights
The Novel Shift

How AI is reshaping R&D tax relief

In March, the spotlight has been on the intersection of artificial intelligence (AI) and R&D tax relief claims, here's what we know so far.

Sanela Tomlinson
·
March 31, 2025
·
Last updated
·
7 min read
How AI is reshaping R&D tax relief

In March, the spotlight has been on the intersection of artificial intelligence (AI) and R&D tax relief claims. I got a chance to sit down with Shilin Chen and ask some of my burning questions about the use of AI in the industry. Below I’ve summarised my posts over the last month, some of your insightful comments and shared a bit more information that I found interesting.

The Big Story: AI Meets Novel

March kicked off with the announcement of the partnership between SmartClaim and Novel.

SmartClaim leverages cutting-edge AI to help write technical narratives quicker, slashing the time consultants spend on writing technical narratives by 2-5 hours per project. Novel offers an all-in-one solution, connecting financial data, claim management and submission into a seamless claim preparation workflow.

So far the feedback we received from our users has been positive and i’m looking forward to finding more ways we can improve the experience of using Novel with AI.

Technical vs Financial: Where Does AI Shine?

One of my posts sparked some valuable insights when I asked whether the financial narrative could benefit from the same efficiency gains as the technical narrative with the help of AI. I’ll admit, this question stemmed from a mix of skepticism and perhaps an outdated belief that AI struggles with numbers (i’ll add links on this topic below). This is just the information I’ve come across, therefore I would be more than happy to hear opposing opinions!

From Shilin’s perspective, AI is less useful for financial calculations because spreadsheets and software already handle it well, such as traditional accounting software like Xero or specialised software like Novel. Although AI could potentially be used to help with invoice matching and spotting inconsistencies.

Paul Rosser shared a compelling example of how AI could significantly contribute to financial narratives. His idea was to leverage AI to provide deeper context about companies involved in transactions. For example, the AI can summarise the company's details, the subcontractors' locations, and whether it is a connected company or an offshore-based company. Even with my current skepticism, I can get behind this approach since it utilises AI for deep research rather than numerical calculations.

Chris Williams proposed another use case: using AI with narrative prompts to look up specific conditions within payroll, such as identifying an employee’s time spent and associated costs on a particular project. Although I could see this approach involving some arithmetics, if the prompts are kept simple and very linear, it could work well too.

The takeaway I gathered from this discussion is that while everyone agrees AI isn’t quite there yet, there’s little doubt it will get there eventually. As Samantha Watkins put it, “like anything, I would grow to trust it with time.” Same here.

Some interesting links about AI and numbers:

Fraud & Compliance: The Impact of AI

A prominent topic on LinkedIn this month focused on the use of AI in R&D claims, with one key layer of discussion being whether AI could inadvertently facilitate more fraud. In an industry already grappling with / recovering from significant fraud cases, there’s heightened sensitivity to anything that might enable it. When I asked Shilin how one can prevent fraud and ensure compliance when using AI for tax relief claims, she pointed out a distinction.

“These are two separate topics in my mind. One is AI making mistakes or hallucinations. The other is humans using AI to do wrong things, which is fraud.”

Although hallucinations are in the very nature of AI models, continued improvements and advancements in the technology, such as retrieval-augmented generation (RAG), have helped tackle this issue.

Somehow, it appears that the issues with AI hallucinations are easier to tackle than preventing humans from using AI for fraudulent purposes. For me, this complexity was underscored by Rufus Meakin in one of his recent polls, where he asked whether people should disclose the use of AI when preparing R&D claims. The answer isn’t as black and white as you would think. It is definitely worth having a read through the comment section on that post.

The Future: AI vs AI

What does the future of our industry look like with AI, and where does HMRC fit into this picture? Shilin believes it’s only a matter of time before HMRC adopts AI to help manage the scale and complexity of R&D claims. From phone support to compliance and operation, she says HMRC “cannot serve the country without using AI” in at least one of these steps.

“Ideally”, said Shilin, “we get HMRC’s AI battle with fraudulent AI agents and agreeing with compliant AI agents so we can move from random enquiries to really targeted enquiries”. Richard Edwards took this idea further, suggesting we might reach a point where it “becomes so difficult to know what is real, what is being prepared and the underlying reality behind all these reports” that it forces us to go full circle. In the end, “HMRC might resort to a different system” altogether "where inspectors are going to spend half an hour talking to competent professionals” to gain a better understanding of their projects.

However, this raises a follow-up question: how does HMRC find enough inspectors with the necessary expertise to facilitate such a system? Ironically, is the answer AI again aka the exact reason why we ended up in this place? We will probably find out the answer sooner rather than later.

All I hope for is that we don’t get to a situation where dealing with HMRC’s AI assistance is like navigating a website’s chatbot, endlessly having to type ‘speak to a person’ in an effort to break through the circle of automated responses to an actual representative.

What about Novel?

We’ve got some exciting plans to bring AI into more areas of Novel, with development already underway. I have spoken about this before but when it comes to integrating AI, we’ve always taken a more methodical approach, never adding features just for the sake of it or that sound ‘flashy’ but aren’t truly useful.

There is one feature in particular that I had a glimpse at last week, and while we’re keeping it under wraps for now, it has the potential to truly transform how you prepare your claims (I really mean it). I don’t know when it will be live but if you want to be the first to know, I suggest signing up to our newsletter list.

Last week we also got a lovely review from James Alexander at Blinkhorns Accountants , I'll leave a snippet here:

"Novel ensures our claims remain consistent with the evolving R&D legislation, providing technical expertise and analysis regarding the R&D process. The specialised software is expertly designed for easy use, providing us with the ability to present our clients’ innovative projects in a concise and professional manner, also offering them the opportunity to personalise their claims. Thanks to the Novel team and their speedy response time, our practice has received support every step of the R&D process."

‍

Newsletter

Read every issue of The Novel Shift

Subscribe on LinkedIn
Book a walkthrough

See how Novel fits the way your team already works.

Book a 30-minute walkthrough