Skip down to main content

Policy Priority Inference

About PPI

Policy Priority Inference (PPI) is a research programme that emerged in 2018 from a collaboration between The Alan Turing Institute, the Centro de Investigación y Docencia Económica, and the United Nations Development Programme. Currently, PPI continues its development at The Alan Turing Institute as one of the research strands of the Computational Social Science Group, part of the Institute’s Public Policy Programme.  

PPI aims at modelling the causal link between government expenditure and policy outcomes while accounting for the multidimensionality and complexity of development. It employs computational methods to overcome the limitations of coarse-grained data on development indicators and public spending. One of PPI’s main outputs is an analytic tool that helps governments measure the impact of public expenditure on development outcomes. It supports evidence-based decision-making in an environment characterised by budget constraints, concurrent and competing targets, multidimensional development, imperfect governance, and context-specific interdependencies between policy issues. This tool uses a specific type of artificial intelligence called agent computing (or agent-based modelling), which allows modelling socioeconomic agents and their decision- making processes, providing a transparent alternative to black-box approaches. The PPI toolkit has been deployed in various countries through collaborations with multilateral organisations and governments (national and subnational) to support development planning and ex ante evaluation in multidimensional settings. You can find academic and policy publications associated with these applications in the publications section of this website.

The PPI model

You can read the details about the latest version of the PPI model in this publication. The original model has been written in Python, and the links below provide access to its source code as well as to a package that can be installed via PyPI. In addition, this website provides a series of tutorials to use the Python version of the PPI model.



While the Python version of the PPI model has been adopted by various users in public organisations, working with code still imposes a significant adoption barrier, especially when an organisation does not have the necessary capacity. For this reason, the research programme created an online version of the PPI with a graphical model and data templates for easy usage.