NSW Procurement
General Ledger data is collected by NSW Procurement from over 150 NSW Government Agencies.
This data covers $30 billion in annual spend and more than 2 million transactions per quarter.
The Question
The rule-based system in place utilises over a million mapping rules, which is difficult and time-consuming to maintain. The DAC has been engaged to develop an automated and robust Spend Categorisation Tool to replace this process.
Our Solution 
- Use AI to classify semi-structured and unstructured information into categories based on taxonomy
- The tool can learn mapping rules from previously categorised spend data and apply insights to new raw data
- Self-learning: accuracy will improve over time with use
- Accuracy for Whole of Government spend data has exceeded proposed 90% benchmark
- Use of AI to remove use of ‘other’ category (~10-15% of total) and replacing these uncategorised invoices with correct classification
Impact
- Greater transparency of government spending
- Demonstrates spending patterns through categorisation at a point in time, or over months or years
- Allows elimination of silos where different departments purchase same assets
- Encourages interdepartmental collaboration and economies of scale Identification of areas where resources are over or under-utilised
The DAC has engaged really motivated scholars and used cutting-edge technology to deliver an outstanding result
Amol Chavan, NSW Procurement
What Next? 
- Improve accuracy to greater than 95%
- Use tool for classifying whole of government spend data from multiple perspectives
- Potential to commercialise the model into a solution that would be offered to other agencies across government, and even the private sector