Governments need to raise resources to be able to deliver essential services, but effective taxation is a challenge. Governments in the world’s lowest-income countries collect only 10 percent of their GDP in taxes, compared to 40 percent in high-income countries. Nigeria, for example, collects less tax revenue than Luxembourg, despite having 300 times as many people. Building an effective tax administration, figuring out how much taxpayers owe, and ensuring compliance are major challenges.

Simple, low-cost tax reminders can boost compliance and revenue without major policy changes. Evaluations from 32 countries show that sending reminder messages to taxpayers that use simple language and emphasize penalties of tax evasion, like fines and prosecution, are most effective. These interventions are often very low-cost to send and generate high returns.

New technologies have the potential to help make tax systems more effective and fair. AI-driven tools can boost efficiency and fairness by helping tax collectors identify taxable properties, quickly and accurately assess property values, reduce bias, and prioritize who to tax. Similarly, e-filing can save taxpayers’ time, reduce opportunities for bribery or misappropriation of funds, and increase revenue.

New technologies should be carefully tested and paired with broader tax reforms for lasting impact. Promising research is already in progress: In the DRC, a tax authority and researchers are building a property tax roll with machine learning to test progressive taxation. In Pakistan, researchers are working with a tax authority to evaluate an AI algorithm that predicts property values from photos.

Randomized evaluations can help identify the most cost-effective messages in each context. In Costa Rica, for example, sending email reminders to businesses that included specific examples of information known to the tax authority raised $18 in revenue for each email sent. In Guatemala, researchers estimated that a letter emphasizing that most taxpayers had already paid their taxes would have generated 36 times more revenue than the cost, if scaled up. In Uganda, a letter emphasizing penalties for non-payment generated 13 times more revenue than the cost. In Bangladesh, a hand-delivered letter that told businesses their tax compliance would be shared with peers led to five times more revenue than the cost.

Using AI to assess property values has the potential to be highly cost-effective, especially where traditional assessments are rare. In Senegal, researchers estimated that, assuming full tax compliance, the AI algorithm would cost about one-tenth of the manual process and raise more than double the revenue. While full tax compliance may not be realistic, this gives a sense of cost savings achieved through algorithms. However, AI models sometimes overestimated the cheapest properties’ values compared to human assessors—meaning human review and input is still important.