🐾 Building TaxCat: AI-Powered VAT Classification for UK SMEs 🐾

TaxCat Interface

As someone working in tax technology, I’ve witnessed firsthand the challenges small and medium enterprises face when navigating UK VAT compliance. Complex regulations, countless exceptions, and the fear of getting it wrong create real barriers for business owners who just want to focus on growing their core business. This inspired me to build TaxCat - an AI-powered application that makes VAT categorisation as simple as possible.

The Problem TaxCat Solves

UK VAT rules are notoriously complex. A children’s book might be zero-rated while an eletronic one is standard-rated. Digital services could be subject to reverse charge mechanisms or standard VAT depending on whether they’re B2B or B2C transactions. Cross-border scenarios add another layer of complexity, especially post-Brexit.

Currently, SMEs face three unsatisfactory options: spend hours researching HMRC guidance documents, pay for expensive professional advice on routine questions, or make their best guess and hope for the best. TaxCat offers a fourth option: instant, AI-powered classification based on official HMRC guidance.

How TaxCat Works

TaxCat employs several modern AI techniques, including Retrieval-Augmented Generation (RAG) among others, to combine the reasoning capabilities of large language models with authoritative knowledge sources. Rather than relying on general AI knowledge, the system draws from a carefully curated database of official HMRC VAT notices and guidance documents through sophisticated processing and analysis methods.

Users simply describe their product or service in plain English and TaxCat provides the VAT treatment along with clear explanations and references to relevant guidance. I’ve also added some common trip-ups if the scenario is particularly complex with edge cases or unknown nuances. The system is still under development but it aims to handle complex scenarios like reverse charge mechanisms, cross-border digital services, and industry-specific exemptions.

Technology Foundation:

  • Backend API built with modern Python frameworks
  • AI/ML Pipeline using semantic search and document retrieval
  • Responsive Web Interface optimized for business users
  • Cloud Infrastructure ensuring reliable availability

Key Learnings from Building TaxCat

1. Starting is Easier Than You Think

The barrier to entry for AI applications has dropped dramatically. With modern frameworks and APIs, you can build a functional prototype faster than you might expect. The key is starting with a clear problem and iterating from there.

2. Getting it Right is Harder Than Expected

While building a basic system is accessible, creating something reliable enough for business-critical decisions requires significant refinement. Consistency, accuracy, and handling edge cases demand careful attention and extensive testing.

3. Problem Definition is Critical

If you can’t articulate the problem you’re solving across different layers of complexity, AI will amplify that confusion rather than solve it. I spent considerable time mapping out VAT scenarios, balancing features, and outlining and refining user contexts before the technical implementation became effective.

4. New Skills for the AI Era

Many fear that AI will replace jobs. I still see it differently. My role evolved from traditional tax research to making strategic decisions about AI system behavior, designing comprehensive testing benchmarks, and navigating complex technical trade-offs. Interestingly, the actual coding now takes significantly less time than before - but this means everything that remains represents genuinely hard problems requiring deeper domain expertise. These hybrid skills - combining specialized knowledge with AI system design - are increasingly critical in our professional services in my view.

5. Human-Centered Design Matters

Even with perfect tax advice, the system fails if users can’t understand or apply it. For example, reverse charge guidance must consider whether the user is issuing or receiving an invoice. Small businesses think in terms of practical actions, such as tax codes in their accounting software, not abstract tax principles.

6. This is Just the Beginning

TaxCat represents an exciting start in applying AI to tax compliance, but there’s enormous potential for expansion - other tax areas, different jurisdictions, and deeper integration with business workflows. I am truly thrilled to be on this journey.

Current Status and Next Steps

TaxCat is in active development with a functional system achieving high accuracy on standard VAT classification scenarios. The core infrastructure is deployed and working, with ongoing refinements to handle edge cases and improve user experience.

Why This Matters

TaxCat demonstrates how AI can democratize access to professional-grade tax guidance. Instead of replacing human expertise, it makes that expertise more accessible to those who need it most - small business owners who shouldn’t have to become tax experts to run their companies.

Note: TaxCat provides guidance based on official HMRC sources but should not be used as the sole basis for VAT compliance decisions. Always consult official guidance or qualified tax professionals for definitive advice.

The project has reinforced my belief that the most impactful AI applications solve real business problems rather than showcasing technology for its own sake. By combining domain expertise with modern AI capabilities, we can build tools that genuinely improve how businesses operate.

Building TaxCat has been challenging but incredibly rewarding. It’s shown me the potential for AI to transform professional services and reinforced the importance of understanding both the technical possibilities and the human needs they serve.

TaxCat is currently in development. For updates on the project or to discuss AI applications in tax technology, you can follow the progress at taxcat.ai or drop me an email at hi@taxcat.ai