Singapore is measuring its AI training by what officers change, not courses finished

Course completion is the easiest number in learning and development to report and the least useful one to act on. It tells you people turned up. It does not tell you whether anything changed at their desk the following Monday. Singapore's public service has now said, on the parliamentary record, that it will stop leaning on it.

In a written answer to a parliamentary question this week, the Ministry of Digital Development and Information set out how the Institute of Digital Government will judge its work. The institute aims to equip more than 150,000 public officers by the end of 2028. Rather than counting courses finished, it will look at three levels: whether officers have built competency, whether they apply it to real problems and whether workflows improve in a way that lasts.

The most interesting part is the AI Sprints. Officers bring a real problem from their own agency, learn product thinking and then prototype and test a solution with end users. More than 1,000 officers have been through them since May 2026. Foundational training sits underneath, and it includes critically assessing AI outputs and the ethics of using them.

The ministry has also declined to set one numerical target for the whole service. Each agency will use measures that fit its own work, such as time saved, turnaround time, accuracy, service quality, customer satisfaction and accessibility. That is harder to roll up into a single dashboard, and it is the right call. A tax office and a hospital do not get better in the same way, so a single number would end up measuring the training rather than the work.

There is a lesson here for any large organisation running an AI program. Measuring application and sustained change means the learning has to start from a real problem with a real owner, because otherwise there is nothing to measure afterwards. That is why sprints, rather than lectures, sit at the centre of Singapore's approach, and it is the logic behind our AI Design Sprints, which end with a tested prototype and a decision rather than a certificate.

The open question is follow through. Three levels of evaluation take effort to collect, especially the third, which only shows up months later. If the institute publishes what it finds at that level, it will have given every other public service a benchmark worth copying.

Sources:

• Ministry of Digital Development and Information, 3 November 2026: MDDI's response to PQ on measuring Institute of Digital Government's effectiveness beyond course completion rates