R&D Tax Credits: what to ask before using AI to prepare a claim
Patrick Murray’s interview got me thinking about a question some CFOs may already have asked themselves. Is AI now good enough to write the technical narrative for an R&D Tax Credit claim? And if so, could we not prepare the whole thing ourselves and save the advisor’s fee?
It’s a reasonable question but before deciding, I’d want to know what risks we’d be taking on and where responsibility would lie.
Say we get to the end of the year, give Claude or ChatGPT a few project notes and ask it to write the technical narrative. It might well come back with a convincing account of the work. Who would check whether it had stuck to what the engineers actually did?
Someone would still need to sit down with them and ask what they were trying to achieve, what they couldn’t readily work out and what they did to resolve it. Then someone would have to decide which work qualified and which costs we could include. If we were doing the claim ourselves, who in the business would make those decisions?
But if HMRC, which I understand uses Copilot itself when reviewing claims, asks us to explain why we included a particular project, what could we show them? There’s a reason HMRC’s GfC3 guidance recommends keeping records from the start of the project. If all we have is a convincing account written by AI months later, what are we going to defend it with?
If we’re considering an advisor who relies heavily on AI to prepare claims, I’d ask them to talk me through how they’d prepare ours. Who speaks to our engineers and checks the narrative against what they tell them? Who makes the final call on what qualifies? If HMRC opens an enquiry, who will deal with it and is that included in the fee?
Before signing off, I’d ask that advisor to pick a project and explain why it’s in the claim and what evidence they relied on. However we prepare the claim, I’d want to know how we’d defend it before we file it.
Commentary from RCK Partners Innovation Advisory Council