AI in R&D advisory: seven control points

Much of the debate about AI in professional services asks whether it can replace expert judgement. In innovation tax and R&D advisory, that is the wrong question. The more consequential question – the one that determines where AI creates genuine value – is whether it improves the quality of the evidence and control environment on which that judgement depends.

The evidentiary standard in R&D tax advisory has risen considerably. HMRC’s compliance activity has intensified, and the expectation is that evidence of qualifying activity must be contemporaneous, traceable and capable of withstanding technical scrutiny.

AI’s most significant contribution is therefore not that it generates better first-draft technical narratives. It is that it can strengthen the control architecture underpinning a defensible claim. Seven ‘control points’ define where that architecture either holds or fails (see the table below).

AI use should be assessed against these control points, not against the volume or speed of output produced. Used properly, AI can improve the quality of captured evidence, apply methodology more consistently, make the review more traceable and support the governance standards increasingly expected by HMRC and clients.

The control points also define the limits of AI’s role. Decisions about eligibility, technological uncertainty, the boundary between qualifying and non-qualifying activities, and the assessment of competent professional evidence remain matters for professional judgement. AI may help to organise and analyse information, but the responsibility for those decisions rests with the adviser. There is also a risk that AI tools may favour material that appears supportive while overlooking relevant contrary evidence; human oversight therefore remains essential.

AI does not reduce the value of professional judgement in R&D advisory. It raises the standard of the material to which that judgement must be applied. 

Javed Bashir, Grant Thornton