Artificial Intelligence

The artificial intelligence (AI) behind SpendInsight emerged from a three-year project between the University of Reading, Goldsmiths College London and @UK plc.

When @UK plc bought Coding International Ltd., a company with 66 combined years of experience in coding and classifying products, a tantalising opportunity emerged. What if the expertise of Coding International could be combined with the knowledge and skill in AI of their partners at the University of Reading and Goldsmiths College?

The key to any artificially intelligent system is the amount of data. The team brought together huge product databases from the NHS and suppliers, and combined these with the unique data that Coding International has built up for hundreds of thousands of products.

State of the art algorithms use the entirety of this combined dataset to learn what makes a product or company unique. This forms the basis of a matching engine that can take a source of data, and link the products and companies therein to the central data repository. This means that the classification algorithms can use not only the data that you supply about a product, but also everything else that has been linked to that product.

@UK plc does not use a single classifier, but a group of five. Each uses a different model of how a product relates to a class, and are variously grounded in information theory, data mining, probabilistic modelling, and rule-based inference. When each classifier has processed an item, an arbitration algorithm assigns a final class.

Once your data are matched and classified, millions of comparisons are made to produce the various benchmarks and analyses. Because this scrutiny emerges from the data itself, it is simple to break it back down and identify the exact areas in the data that will give the biggest savings opportunities.

Paul Roberts and Matthew Brown are both submitting the research they have contributed to the novel artificial intelligence algorithms of SpendInsight as Ph. D. theses.