Mark Graves PhD, R&D Technical Partner, AAB
Fintech and PE operations increasingly rely on software, machine learning and AI solutions. Software development work undertaken in these areas can be opportunities for potential R&D tax claims. A generous UK government scheme to encourage innovation, whereby companies can, based on eligible qualifying costs, receive either a reduction in their corporation tax liability or an increased tax loss surrenderable for cash.
However, for an eligible claim the development project must not simply “use advanced technology” but must seek to “advance technology” through the resolution of a “technological uncertainty”, for a field of technology as a whole, not merely make an advance for just the company themselves.
In Fintech and PE contexts, this often involves:
- limitations in existing data processing methods
- constraints of financial regulatory systems
- performance issues that cannot be solved by standard patterns
- integrating distributed systems with incompatible protocols
- developing new or pushing existing machine learning models beyond established frameworks
- safely deploying AI across highly heterogeneous data estates
Where engineering teams face obstacles that cannot be resolved through commercially available tools, established industry approaches, patterns or reference architectures, this may indicate qualifying R&D activity.
Fintech examples that may contain qualifying R&D
- real‑time fraud detection models requiring new techniques for feature extraction
- latency‑sensitive order routing platforms where existing technologies cannot meet throughput constraints
- cryptographic or secure compute architectures developed to satisfy regulatory requirements
- advanced risk‑modelling platforms integrating heterogeneous datasets with conflicting constraints
Private Equity examples that may contain qualifying R&D
- portfolio‑wide data unification where standard ETL methods fail due to incompatible legacy architectures
- developing a shared analytics or AI platform deployable across inconsistent environments
- resolving system‑level uncertainties around synchronising data across cloud/on‑prem hybrid estates
- designing scalable, secure mechanisms for portfolio companies to contribute to a consolidated data lake
The Private Equity portfolio challenge: standardising data across legacy architectures
Private Equity presents a particularly interesting scenario.
Most PE houses own multiple portfolio companies, often with:
- different ERP, CRM or finance platforms
- different data formats and schema definitions
- inconsistent data quality
- incompatible analytics and reporting systems
- a mix of cloud, hybrid and on‑premises infrastructure
A common PE objective is to create portfolio‑wide insights: standardised metrics, consolidated reporting, and comparable performance dashboards across all portfolio companies.
Routine data manipulation, visualisation and reporting when performed using established tools and methods, does not typically meet the technological advance requirement. These tasks may be highly complex, labour intensive and commercially valuable, but they do not typically require an overall technological advance.
However, qualifying R&D can occur where a PE house attempts to create a unified data layer, shared analytics platform, or cross-portfolio AI system in a manner that for example requires:
- new approaches to reconciling heterogeneous legacy systems
- non-obvious architectural solutions that address conflicting constraints across portfolio companies
- new methods of deploying AI models into incompatible environments
- non-readily deducible solutions that preserve data security and regulatory compliance across diverse systems
- techniques to integrate datasets of fundamentally different structure or semantics.
For example:
A PE house wishes to deploy AI‑based analytics across all portfolio companies, but the companies have incompatible data estates. Developing a deployment method that allows AI models to operate consistently across these heterogeneous environments, while maintaining regulatory compliance, latency requirements, and data‑governance controls, may involve resolving system‑level uncertainties.
This type of work may go beyond known approaches such as ETL (Extract Transform Load) and it could touch on genuine technological challenges around distributed computing, scalable AI deployment, and data architecture, areas that frequently fall within scope for R&D tax relief when uncertainties need to be resolved.
Identifying potential eligible R&D in AI and software projects
Ultimately to make an accurate determination of project eligibility and prepare a robust and defensible R&D claim for software / AI projects requires a deep dive into the technology developed on the project by someone experienced in the technical field and HMRC’s eligibility criteria and investigation process.
However, finance professionals in Fintech and PE can look for potential indicators such as:
a) Did the project overrun in terms of budget or timescale?
Whilst it doesn’t confirm qualifying R&D took place, a project which overran in terms of budget or timescale can be a good indicator that the technology being developed was not routine or readily deducible.
b) Limitations of existing frameworks or platforms
Did the team have to create new methods because existing tools could not meet accuracy, performance or security constraints?
c) Multiple technical attempts, iterations or prototypes
Did early approaches fail? Did different architectures need to be evaluated?
d) Evidence of system-level complexity
Did the project involve distributed systems, cross-domain constraints, or conflicting performance and regulatory requirements?
HMRC’s heightened scrutiny of R&D claims
Following widespread reports (in the national press) of non-compliance with R&D claims (especially prevalent amongst unregulated boutique advisors), after January 2023 HMRC dramatically increased levels of compliance with software-based claims frequently being investigated.
Many enquiries arise because claims lack technical detail or misinterpret the eligibility criteria. Common issues include:
a) Describing product features, not technological advances
A claim might say “we built a new risk‑scoring engine”, which only explains what was developed. HMRC needs to understand how it pushed the boundaries of software or IT.
b) Focusing on commercial novelty rather than technological advancement
Introducing a new system to the business does not mean it advances the state of the art.
c) Applying known technologies without resolving underlying uncertainties
Simply using AI libraries, cloud services or established frameworks is not R&D unless technological limitations had to be overcome.
d) Claims prepared without specialist input
HMRC frequently challenges claims that lack evidence of technical understanding or fail to provide a detailed assessment of the baseline technology (at the start of the project) by which a claim is judged.
Many companies adopting AI solutions have come under HMRC investigation and often this is a surprise, as they believed or were incorrectly advised that applying a state-of-the-art AI tool would be considered qualifying R&D. However, to be eligible a project needs more than just apply the latest LLM, it needs to show how it advanced some aspect of AI or some knowledge of how AI could be applied to solve particular types of problem.
In a fast-moving field like AI, a project which may have been state of the art at the start of the period, may be routine by the end, however such projects can be still claimed as the eligibility criteria is based on the technological baseline at the start of the project.
There is no requirement on the project being successful, indeed projects which “failed” are often fertile grounds for R&D claims since a project failure can often lead to new knowledge of the limitations of a particular technological approach.
Importance of an experienced advisor
An advisor with strong technical know-how, significant experience in preparing claims in similar fields of technology, as well as experience in how HMRC handle investigations, can be invaluable in both reducing the likelihood of an investigation in the first place and in efficiently resolving the investigation to a successful outcome.
A strong R&D advisor will proactively review and advise on the preparation of contemporaneous project documentation as part of the claim process and increasingly on larger projects this can be in “real-time” at periodic points within the development cycle rather than many months after the development which is when a claim will typically be submitted.
An advisor’s role should extend beyond drafting the technical narrative. They should:
- assess whether existing documentation adequately captures uncertainties and technical decisions
- identify gaps that may weaken the claim during an investigation
- advise engineering teams on simple practices to capture future documentation more effectively
- ensure the evidence aligns with DSIT guidance and the technical narrative prepared for HMRC
This reduces the risk of investigation and greatly improves the likelihood of being able to successfully resolve an investigation if one occurs.
Conclusion
R&D tax credits can lead to a generous tax saving / cash rebate reducing the overall cost of a software / IT / AI or machine learning project. Given the costs that many companies are spending, especially on AI deployments, this can be a very valuable incentive.
However, to qualify it is necessary to evidence that the project sought to make an overall advance in technology, not merely apply known solutions / best practice within the company. Whilst HMRC scrutiny of claims has increased in recent years, an experienced advisor should be able to quickly identify if a project is likely to qualify and help prepare a strong claim that can be robustly defended in the event that the claim is investigated.
Mark is a Partner in the Innovation Tax team at AAB, specialising in Software, AI and Machine Learning based R&D claims. He has a PhD in Machine Vision and spent 20 years working in industry before moving into the R&D tax claims in 2010. He has prepared over 2,000 claims and successfully resolved over 50 HMRC investigations.