Oxygen Finance uses GenAI to provide more insights to clients

Seeking a way to identify public sector spending cues

Founded in 2004 and based in Birmingham in the United Kingdom, Oxygen Finance helps public sector organisations and their suppliers to manage procurement and payments for goods and services. It offers a cloud-based service called Oxygen Insights that helps businesses learn about opportunities to sell to the public sector. A small Oxygen Finance research team identifies these opportunities by reviewing public sector websites, meeting minutes and other public records for discussions about procurement matters.

For example, the team might find a local council conversation about plans to update an ERP system, which could signal a potential bidding opportunity for companies that provide such systems. Every month, the research team reviews thousands of public sector records to update the information that Oxygen Insights provides to customers.

Since Oxygen Insights was launched in 2013, it has tracked more than GBP7.8 trillion in public sector spending data and supported the awarding of more than 65,000 government contracts. But the company knew there were additional sectors and more procurement opportunities that it could bring into scope that were currently restricted by the laborious manual research required to find relevant information. After learning of SoftwareOne’s AI and data expertise and deep experience in working with Microsoft Azure, Oxygen Finance decided to work with SoftwareOne to develop an automated system that could help to identify and gather public sector procurement information that the research team might be missing.

Building a proof of concept using OpenAI’s GPT-4o

Working in collaboration with SoftwareOne and Microsoft, Oxygen Finance developed a plan to create an AI-powered tool that could help identify procurement discussions in public sector records and automatically generate reports for customers based on that information. SoftwareOne assigned a data team based in Europe to lead the project. Microsoft also provided financial support for the project through its Azure Innovate fund.

Using a decade-plus worth of data from Oxygen Insights, the team began training a model using Microsoft Azure and OpenAI’s GPT-4o to intelligently automate the research process. They developed the proof of concept over a period of two-and-a-half months, with repeated rounds of training to enable the model to recognise potential buying signals in public sector data.

The solution was designed to use an advanced web crawler to search a set of targeted online sources and extract relevant data from PDFs, Word documents and other types of information. It uses Azure Blob Storage to securely and scalably organise and manage that data, and Azure OpenAI to summarise and categorise information. To ensure that data is easy to retrieve and act on, SoftwareOne used prompt engineering to fine-tune categories to align with the existing templates used for Oxygen Insights.

The team designed the system to intelligently integrate new information that it found with the existing data in Oxygen Insights. This provides a clear timeline of discussions and publications as a procurement process progresses and matures. Generative AI is then used to automatically generate reports that are tailored to users’ needs and specifications. The solution was designed to enhance data accuracy as much as possible while reducing the need for manual intervention. To ensure trust and transparency for customers, Oxygen Finance manually reviews every report that’s generated using AI and clearly labels each one to distinguish these reports from human-researched data. This checking ensures that data quality is maintained and also provides feedback to further train the AI solution.

After testing showed that the solution was performing according to expectations, Oxygen Finance went into production in December 2024.