Tax and Customs Administration books good results using risk models to check returns
Data-driven selection of high risk tax returns benefits the public purse
The Tax and Customs Administration has successfully developed electronic risk models and is putting them to good use checking tax returns. It is using data to automatically identify tax returns with the highest risk of errors so that they can be corrected by a tax inspector.
The Netherlands Court of Audit has investigated two risk models in practice. Most errors were detected by a relatively simple model to check VAT refunds. The new approach increased the proportion of returns detected with errors from 17% to 34%, good for additional tax receipts of approximately €60 million a year. A risk model developed to detect errors in personal income tax returns was not as effective because it was unable to generate concrete indications of where errors were made and tax inspectors still had to check the returns in detail by hand.
The audit was published on 11 June 2019. It reveals that the risk models select only returns suspected of declaring too little tax. Returns are not selected if taxpayers overstate their liabilities. This is not in keeping with the Administration’s supervision policy. In his response to our audit, the State Secretary for Finance said that the risk models would be adapted in this respect.
Staff capacity determines how many returns are checked
The Tax and Customs Administration cannot check every return it receives. On an annual basis, it finds approximately one million potential errors in the returns; of which about 350,000 can be physically checked by a tax inspector. The risk models identify the returns with the highest risk of error so that they can be checked by hand. How many returns are checked, however, depends on the available staff capacity. The deployment of additional staff is not weighed up against the additional revenue that more checks would generate. The Court of Audit recommends that the State Secretary for Finance clearly explain staff deployment decisions to the House of Representatives. He does not do so at present. The Administration is investing heavily in data-driven initial checks by means of risk models. It expects the new instruments and risk models to increase tax revenues by €750 million a year in due course.
In the models we audited the Administration treats personal privacy with respect. The Court’s Board member Francine Giskes commented: “Our audit did not find any signs that the Tax and Customs Administration’s models were unethically profiling taxpayers’ characteristics or that certain groups of taxpayers were being checked more often than others. The Administration is aware of the public sensitivity of the use of algorithms and data in the tax return processes we audited.”
To protect personal and corporate data, we recommend that senior management be more closely involved in monitoring the red lines. The Court of Audit also recommends that a set of standards be developed on the use of models and data analysis. The standards should in any event apply to the Tax and Customs Administration and perhaps even to all of central government.