R&D tax credits UK and the AIF problem: why ‘good R&D’ still fails

Intro 

R&D tax credits UK can be lost for reasons that have little to do with whether the underlying work is innovative. The most common failure mode in UK R&D tax credits in 2026 is a claim that looks plausible in isolation, but breaks when HMRC compares the AIF narrative to the cost model. The fix is operational, not rhetorical.  

What the AIF is trying to standardise 

The Additional Information Form forces every claimant to describe projects in a consistent format and to identify who is accountable for the claim. It requires a main senior internal R&D contact and disclosure of all agents involved. For each project, it asks for the field of science or technology, the baseline capability, the advance sought, the uncertainties faced, and the work undertaken to resolve them. 

Behind the form is a simple test: would a competent professional accept that the uncertainty was real for the field, and that the work was systematic in trying to resolve it? The statutory DSIT guidelines set out this definition. 

Why ‘good R&D’ still fails: five avoidable breaks 

• Baseline drift: The baseline described in the AIF is written from the company’s perspective (‘we could not do X’) rather than the wider field (‘what was publicly known or readily deducible at the time’). 
  

• Uncertainty versus difficulty: Engineering challenge is framed as uncertainty even though the route to resolution was already established in the field. 
  

• Field-hopping: The project changes fields mid-narrative (for example ‘AI’, then ‘data science’, then ‘software engineering’) which makes the baseline impossible to anchor. 
  

• Cost pools that do not match the story: The claim includes a broad software or contractor population, but the narrative describes a narrow experimental thread. 

• Unexplained apportionments: Percentages are applied without a metric, a data source, or a reproducible method. HMRC expects costs to be attributable to the described R&D, not just broadly related. 
  

A ‘single story’ framework that holds together 

Claims that survive scrutiny usually have one thing in common: the technical narrative and the cost model can be cross-checked in minutes. A practical way to force that consistency is to build three artefacts from the same project register. 

1. Project register – A list of projects for the accounting period, each with a named competent professional, start and end dates, and the intended advance. 

2. Narrative pack – For each project, a short baseline, uncertainty and approach statement written in the language of the DSIT definition, then reused consistently in the AIF and any supporting report. 
   

3. Cost map – A one-page mapping from ledger accounts to qualifying categories, then to projects, with the apportionment method and evidence source noted for each category. 

Controls finance teams can implement quickly 

• Pre-submission reconciliation: project-level totals in the AIF match the computation outputs and the underlying ledger extract for the same accounting period. 
  

• Named-owner review: the senior internal R&D contact signs off the project narratives and confirms the competent professional list before filing. 
  

• Apportionment file: for each mixed-use cost, keep the metric, the raw extract (timesheet, usage report, invoice list), and a short note on why the method is reasonable. 

• Agent register: maintain a single list of all agents involved in the claim, including those who advised, analysed costs, or helped complete forms. 
  

AI drafting tools: consistency traps 

AI can produce fluent baselines and uncertainty descriptions, but it also tends to generalise and to change terminology between drafts. That is a problem because the AIF requires a stable field, baseline and uncertainty chain for each project. If AI is used, lock the vocabulary early and insist that every statement is backed by contemporaneous project evidence. 

FAQs 

What is the biggest reason an AIF-led claim is rejected? 

Internal inconsistency: the baseline, uncertainties and activities described in the AIF do not align with the cost pool and apportionment logic. 

How long should each project description be? 

Long enough to explain baseline, uncertainties and approach clearly, but short enough to be auditable. The AIF asks for specific elements rather than a full technical report. 

Do we have to disclose all advisers and agents? 

Yes. The AIF asks for contact details of all agents involved, including those advising, analysing costs, or helping complete forms and returns. 

What counts as R&D for tax purposes in the UK? 

Projects that seek an advance in science or technology through the resolution of scientific or technological uncertainty, as set out in the DSIT guidelines. 

Should we attach a separate technical report as well as the AIF? 

It can help where projects are complex. The AIF guidance allows additional supporting details to be submitted separately, but the AIF still needs to stand on its own.