Your tax team is exploring AI. Why hasn’t anything changed?

How tax functions are using AI 

AI handles ambiguous inputs to produce structured outputs. PDF invoices, legal agreements, or a set of meeting notes can be turned into something usable. 

Across the tax lifecycle, AI is being used alongside automation to deliver the most value for tax functions across the three areas below: 

• Tax data entry and processing: extracting information from unstructured documents, cleaning and formatting data, linking across systems, and populating forms that standard tax software does not support 

• Tax data analysis and classification: analysing large volumes of tax data, surfacing patterns and anomalies, and classifying transactions and assets to support faster, sharper review 

• Intelligent tax guidance and decision support: helping tax professionals make informed decisions using data, context and precedent, with AI grounded in company-specific data and historical outcomes 

Most tax teams sit between ‘Foundational’ and ‘Emerging’ on our AI maturity scale and are using AI tools primarily for productivity. 85.3% of attendees told us they already use general AI tools (e.g., Copilot or ChatGPT) informally. None reported AI being embedded across multiple parts of the tax lifecycle. Over the next 12 – 24 months, we expect most tax functions to experiment and try to move up the scale; however, we see a gap between intent and progress, with few likely to move past ‘Emerging’

Data is holding tax functions back  

Upgrading AI tools won’t move functions to ‘Advancing’ on its own; reviewing data is key before this.

Without consistent, complete and accurate data – regardless of the AI tools used – outputs will be unreliable. Most heads of tax in the room stated that they have work to do in at least one of these areas. Looking ahead 12 months, data quality and infrastructure topped the list of constraints (47.1%), ahead of budget and resource (29.4%). Then, when asked about what their primary data hurdles were, 54.5% pointed to data quality and consistency issues, with a further 24.2% citing siloed data. 

Data is holding tax functions back  

Upgrading AI tools won’t move functions to ‘Advancing’ on its own; reviewing data is key before this. 

Without consistent, complete and accurate data – regardless of the AI tools used – outputs will be unreliable. Most heads of tax in the room stated that they have work to do in at least one of these areas. Looking ahead 12 months, data quality and infrastructure topped the list of constraints (47.1%), ahead of budget and resource (29.4%). Then, when asked about what their primary data hurdles were, 54.5% pointed to data quality and consistency issues, with a further 24.2% citing siloed data.

Practically this means that data readiness must come before AI adoption, not alongside it. Data cleansing cannot be treated as a one-off project – it must be an ongoing control, because AI amplifies errors as much as it amplifies insight. 

Governance and controls can often be seen as slowing adoption down; however, these are what make long-term adoption sustainable. Functions getting the most from AI put these in place early:  

• Assign clear ownership  
• Ensure human review  
• Build testing into your process  
• Monitor on an ongoing basis  
• Maintain an audit trail  
• Ensure appropriate data access controls  

The tax landscape is moving forward 

It is worth zooming out from the room itself. HMRC's 2026 Transformation Roadmap confirms it is investing heavily in AI, automation and the data platforms that underpin them; its own reporting states that AI and analytics contributed to protecting and recovering £10 billion in tax during 2025/26¹. This means data quality is not just becoming an inefficiency issue but a risk one.  

Where to start  

Embedding AI and automation is more resource- and time-intensive than it may first seem, and there is a lot of ground to cover to get to ‘Advancing’ – we can help you get there. A few practical starting points drawn from the room and our own work with tax functions: 

• Diagnose first: Know where you sit on the maturity scale as this will guide your next steps
• Prepare data foundations: Map data sources and define requirements before committing to a build
• Prove using pilots: Pick one or two priority use cases, focusing on areas of highest repetitive effort or greatest data volume for quick wins 

_{[1] HMRC, Transformation Roadmap: update 2026; HMRC, Annual report and accounts 2025 to 2026}