Page 8 - IOTA Good Practice Guide

2. Predictive modelling

Many tax administration think that predictive modelling is complicated and requires a high level of
expertise to start with. A good practice it is to start small and easy and to grow iteratively.
Predictive modelling is an optimization process, data science and big data offers new opportunity to
improve the efficiency of risk modelling.
Predictive modelling is the traditional area application of advanced analytics and big data in tax
administration, because it is a great tool to increase revenue, detect fraud and support reorganization.
Predictive modelling is like cooking
You need some ingredients before starting with predictive modelling. These ingredients are your risk
indicators. It could be raw material like business rules or prepared ingredients like the output of a
clustering model or results of a network analysis.
With that, the state of the art is to create a tasty recipe, much better is to improve the procedure by
iteration, like a chief who test a new dish and improve his recipe through the time. Mixing unusual
ingredients needs some technics and is risky, and could create unexpected results. Of course, new
advanced analytics and machine learning methods offer a great opportunity to find the ultimate recipe
by optimizing the combination of the ingredients.
Risk indicators – Business rules
Level: basic
Contact person: Jean-Luc Wichoud, Switzerland
Problem:
How to capture good ideas and apply it in a systematic way in
a risk model.
Solution:
Using rules defined by the business as risk indicators is the
old fashion and easiest approach, used for more than 15
years in tax administrations. It is one of the best ways to
capture business knowledge in a risk model. In the attachment on the GPG website, you find an
example from Switzerland, showing the integration and the evaluation of business ideas as new risk
indicators and the challenge to find efficient rules fitting with the aims of the tax administration.

6