Boosting Efficiency in Tax Analysis

GIC uses AI to classify tax update notifications, reducing the need for manual screening of reports

Every country has a tax code used by employers to calculate the taxable portion of each person’s income. Tax codes change frequently and this has implications across investment portfolios, causing a huge burden for tax teams which have to keep abreast of these developments in order to provide consolidated advice to stakeholders.

At GIC, a sovereign wealth fund established by the Singapore government, the tax team faced this same time-consuming task.

As Prateek Prakash, Senior Vice President and Head of Private Markets Solutions at GIC’s Technology Group explained, “The GIC tax team monitors tax and regulatory changes across the world. The information comes in multiple forms (such as) email notifications, newsletters, portals and the web.”

Analysts had to track and screen all these reports manually for updates based on topic and region of interest. This involved sifting through mountains of documents in their inbox to identify information specific to tax categories that they specialised in, before they could even get started on the analysis.

The AI transformation

Wanting to transform this laborious process, GIC looked to artificial intelligence (AI) to help the tax team streamline the reports and data that they receive. It worked with AI Singapore under the 100Experiments (100E) programme to build a document labeling algorithm that could parse a document and identify the specific tax topics it related to, as well as the geographical region that it covered.

The AI system developed by the project team uses an ensemble of natural language processing models to automatically classify reports based on topics and country of relevance. With the AI solution doing the first round of classification, GIC’s tax specialists are able to spend more time on value-added tasks such as analysing the impact of the changes and providing expert insights.

The system passed GIC user acceptance tests with 82 percent accuracy and its subsequent adoption by the organisation led to improved performance in analyst-level classification, resulting in more streamlined workflows and greater efficiency in tax analysis.

Author