Transforming procurement processes


3 min read

Procurement is an important strategic function within all businesses. An effective, streamlined process has the potential to hugely improve an organisation's profitability. In this Data Spark project, a team of students at Imperial College London successfully built a prototype model for their client which identified requisition requests suitable for fast-tracking. Using advanced data science techniques, they were able to demonstrate how this model has the potential to decrease the average cycle time of every requisition by 65%.

In this article Olivia Dalton-Hoffman (MSc Business Analytics), describes the project and techniques used to successfully build this model.

“Over the past few months, our Data Spark team has worked with KPMG and the procurement team of a global FTSE 100 organisation. We were asked to look at ways data analytics might be used to help reduce bottlenecks in their purchasing process. With a large volume of data to work with, we used unsupervised machine learning to gather insight on purchase orders (POs). We were able to identify key attributes for all POs.  This enabled us to build a proof-of-concept model which allowed us to identify those POs suitable for streamlining and those which required more detailed attention.  

POs with the potential to be streamlined were segmented as those requiring fewer iterations and having a low-level of complexity. Streamlined POs will typically have shorter cycle times. Non-streamlined POs are those that may create value. The latter requires more detailed attention and tend to have higher cycle times.

Through the application of this fast-track model, our analysis showed that the volume of purchasing and sourcing work could be reduced by 80%, resulting in a decreased overall average cycle time of 65%.

Our model can provide recommendations to determine which POs are high-value or have potential savings leading to faster decision-making. It presents the company with the ability to focus efforts on key requisitions to maximise savings on high-value POs and save time and effort on the streamlined POs.

Beyond this proof-of-concept model, we also offered the client recommendations for successful implementation.  We suggested practices to build the model at scale, along with how to enable employees and the machine learning model to work in conjunction.  Additionally, we provided data management recommendations that would be simple to implement, but have the potential to offer further insights either independently or when combined with the model. We believe the adoption of our model, combined with the data management suggestions, has the potential to create immense value for the client’s procurement process.”

Nick Whitfield (Partner, KPMG) explains

"I have been impressed with the speed in which the team has worked to unearth some impactful findings. The outputs of this project have validated a number of assumptions that the client had but have never been able to validate.

The team confirmed these assumptions with real data points. The finding will support in building a robust business case for investment in developing a data-driven solution to automate parts of the procurement function."

Written by: Olivia Dalton-Hoffman (MSc Business Analytics 2019/20)

Team members: James Chou, Olivia Dalton-Hoffman, Callum Fenn-Macalister, Thibault Lavallee, Lucy Vincent (MSc Business Analytics 2019/20).

Academic Mentors: Julio Amador, Jose Nunes Teixeira Vaz Moreno.