One should not forget the importance of risk models that are more useful from a policy prescript and diagnostic perspective. There are several risk models with heat maps providing diagnostic capability that could be complemented with measurement models, creating a composite indicator. The reason these models don’t perform well is that they are risk models that require experts within countries to run focus workshops to address the highly subjective ranking. Moreover, the questionnaires are detailed and drill down to very specific problems within organisations that have IFF touch points. The model is easy to navigate, use and populate and has the distinct advantage of being used regularly (i.e. annually or biannually), to track progress in implementing the risk mitigating, contextually-oriented solutions. While the tool does not offer precise solutions, its diagnostic capability is prescriptive enough to provide insights and guidance to government officials. One of the most powerful models is the IFF CRP risk model and the national risk assessment on beneficial ownership. The latter model is probably the most powerful lever in curbing IFFs.Another equally important step, is to pilot of an Artificial Intelligence and Machine Learning model using big data analytics and Neural Networks. This is because AI and ML are not only methodologies but also technologies and tools. AI and ML approaches using big data analytics, the various algorithms and neural networks is a critical ingredient in developing a measurement indicator rather than an estimate. This requires close collaboration with the relevant government authorities to obtain the necessary high-level political approvals to use the transactional, administrative data. Several case studies have been proposed and these could be used, especially the tax evasion and trade mis-invoicing cases. The UNECA needs to run a pilot study modelling an IFF tax evasion indicator using AI and ML applied to, for example, transactional tax declarations, customs declarations, cross border flow payments, etc. The data should not just be limited to what is available in the public domain. These findings should form part of the composite indicator. This could be piloted for South Africa and possibly Namibia who are currently keen to take the AI work forward.In the absence of granular data to facilitate measurement, estimates have to be sufficient since this is a clandestine activity being measured. IFFs and their various channels are illegal and are by nature hidden, clandestine activities. From a policy perspective it is more important to provide insights to developing and developed sovereign governments to proactively curb IFFs and estimates are sufficient if they are consistently tracked over a period of time. National governments struggle to get accurate national accounts data recorded, how much more difficult will it be to record and track a hidden one?Officially published data is king, especially national accounts data. In the public sector, any models that do not use officially published statistics, are generally discounted by governments. The advantage of SNA MACRO data sources, is that they are standardized, and reporting is managed through stringent guidelines with large pools of resources both in the government and at the UN for example to manage and ensure correct and more accurate reporting.It is evident, that models that use SNA-MACRO data have been allocated with higher scores. This is largely due to the fact that these are national official statistics with significantly more resources to collect, collate, clean and mine this data for reporting to international bodies such as the UN, the IMF and the OECD, to mention but a few.While it might be impossible to eliminate IFFs entirely, efforts to measure them coupled with improved transparency measures, will go a long way in understanding the problem and the channels used to facilitate illicit outflows, making it easier to curb IFFs in developing and developed countries alike. It is however possible to use proxies and innovative models, methods and tools to try and quantify the scale of IFFs. IFFs are clandestine activities that are virtually impossible to measure accurately and to do so would not only be foolish but futile exercise and a complete waste of precious scarce fiscal resources. It is possible however to use proxies and sophisticated models to quantify IFFs as best possible, cascading from top-down macroeconomic models and data sources through to more meso-level measures and estimates, and finally down to very granular level administrative, survey or transactions-level data provides far more insight into the extent and scale of IFFs while highlighting key governance, legislative and administration loopholes. As we move towards more innovative AI and ML techniques, in an ever-changing digital world, with rising levels of e-commerce, new tools will become available to better quantify IFFs, using big data analytics, machine learning, artificial intelligence and fuzzy logic as well as neural network programming. While we move towards more sophisticated artificial intelligence, in the interim however, a baseline measurement is required in order to monitor and track IFFs if this is meant to be a useful evidence-based policy tool for developing countries on the African continent. Such an indicator for IFFs should therefore consider not only the total quantum, but also provide details enabling governments with the information to motivate and drive the desire to curb IFFs from its different sources and channels
