Nominal Coding
Whilst PO-based invoices typically derive their nominal or GL coding from the ordered items, overhead or expense-based invoices without a PO may require additional coding steps. Traditionally managed manually by the Finance team, there are now several more automated approaches available:
- Default supplier-related nominal codes.
- AI-driven coding:
- Based on previously coded invoices.
- By matching invoice line descriptions with nominal code descriptions.
AI Nominal Coding
The success of AI-driven options largely depends on the complexity of the nominal code structure and the level of variation in purchases. At a minimum, AI-based systems should provide users with a list of potential matches, ranked and sorted by accuracy.
When suppliers provide a wide range of goods and services, and the customer has a large nominal code structure, AI can be used to help speed up the process. By analysing how specific invoice lines have been coded in the past, AI can identify patterns and learn to apply the appropriate coding for future transactions, improving accuracy and efficiency.
Alternatively, the AI model can be trained on the entire nominal code structure, using natural language processing to match invoice line descriptions with nominal code descriptions. While this approach can be less accurate, it does not depend on previously coded invoices. To address ambiguity, the system may present a shortlist of potential matches, allowing the user to make the final selection.
Below is a simple example I created utilising an LLM to match the invoice line description to the correct nominal code:
{
“Items”: [
{
“invoice_line_description”: “PREMISES CLEANING FOR MONTH MAY 2024”,
“invoice_line_id”: “0”,
“ledger_code”: “730100”,
“ledger_code_id”: “366”,
“match_explanation”: “Matched due to similarity between ‘cleaning’ and ‘Cleaning contracts'”,
“match_accuracy”: “8.2”
}
]
}
More specialised AI models are being developed to function as full knowledge workers, offering a deeper understanding of the nominal coding process. Trained on large datasets and designed with a narrower focus than generic large language models, these agentic systems provide exceptional accuracy. They promise to deliver precise results with minimal human oversight and intervention.
Nominal Coding Requisitions
AI is becoming increasingly important in supporting nominal code allocation within the requisition process. It applies similar methods to those used for coding invoices, ensuring consistency and accuracy. This is particularly relevant in the education sector, where purchases are often ad hoc rather than sourced from predefined product catalogues. As a result, items must be manually assigned nominal codes based on their nature, rather than relying on predefined coding linked to specific products.
I spoke about this in my post – Streamlining Purchase-to-Pay in Education: Simplifying Nominal Coding with Large Language Models