The share of statutorily-eligible individuals and households who do not receive social benefits – the non-take‑up rate – is significant for some countries and for some benefits. In Belgium, for example, 37% to 51% of eligible working-age people do not take up social assistance, and in the United States, around 20% of eligible families do not benefit from the Earned Income Tax Credit. These are sizeable gaps, with potentially large financial implications for households. Importantly, people with low resources are the least likely to respond to simple behavioural interventions encouraging them to enrol in social programmes.

Data‑informed, national strategies against poverty and social exclusion aim to identify people in need, often with the explicit target of increasing the reach of social protection. At least 29 of the 38 OECD countries have such frameworks in place. Most countries, including Ireland and Spain, identify groups and regions in need of social programmes based on probabilistic estimates of survey and administrative data. Once coverage gaps are identified, the policy response casts a wide net, including better communication and investment in new programmes. Public outreach and communication campaigns frequently target a particular benefit, a specific disadvantaged group or a geographic area. This approach is particularly useful for reaching people whose personal data may not be known by the public authorities, such as undocumented residents, informal workers or people experiencing homelessness.

A few countries, including Belgium, Chile, Estonia and France, increasingly link administrative data across sources to enable the identification of social benefit eligibility at the individual level. Data linking usually happens across different agencies or through a social registry.

Linked administrative databases can be used to measure non-take‑up; help close information gaps (e.g. eligible households can be encouraged to apply); and lower the administrative burden on users (e.g. by pre‑filling information from administrative sources). In a handful of programmes, including child benefits, linked data are even being used to enrol users automatically.

Most claims for benefits and services can now be made online in many OECD countries. While this presents barriers for some users, it should also enable agencies to focus human resources on people who find it difficult to access automated systems, like people with complex needs or with limited access to (or familiarity with) digital resources.

Governments are only at the beginning of digital transformation in social protection. Advanced uses of technology and data are less common in social programmes than, for example, in the healthcare sector. While government agencies are increasingly making use of administrative data, they are yet to exploit it in a systematic way.

Many uses of advanced technologies, in particular artificial intelligence (AI), continue to be small, ad hoc test cases to determine feasibility and scope for deployment. Countries are proceeding with caution, implementing and evaluating small scale projects to manage risks.

In social protection, AI is most often used to interface with clients via chatbots and for automating back-office processes. A few countries report using AI to facilitate fraud detection. Apart from these examples, the use of AI in social protection remains limited; so far other types of automated decision-making (with human involvement) remain more common. One reason for this is that significant challenges exist with the use of AI, and countries are proceeding with caution. Several high-profile cases highlight risks such as discrimination and exclusion and have threatened public trust and confidence in governments’ use of technology and data. Countries can work on identifying measures to address these risks, for example by applying the OECD Principles on Artificial Intelligence.

Leveraging advances in technology and data does not come without challenges and risks, particularly in the realm of data governance. Challenges include ensuring sufficient cross-governmental collaboration and data protection and privacy. There are also risks associated with discriminatory biases built into automated processes and decision-making, which have the potential to reinforce or create new sources of exclusion and disadvantage. These challenges require risk mitigation, with instruments like legal and regulatory frameworks.

• Strengthen national strategies to identify people in need and integrate them into social programmes. Linking data from different sources is useful for estimating non-take‑up, identifying potential beneficiaries and informing people of their entitlements. At the same time, continuing to identify groups in need via de‑identified data can help to inform outreach campaigns.

• Explore the feasibility of automatic payment of social benefits. Automatic enrolment using personally-identified, linked administrative data can increase the take‑up of benefits, as it relieves recipients from the burden of applying.

• Apply lessons from behavioural research to the digital transformation. In randomised control trials, treatments of simplified information and support in programme applications usually have positive effects on applications and eventual enrolment. Sending prompts or clear information about likely eligibility – for example, based on linked administrative data – can also help increase enrolment, though governments still face challenges reaching people disconnected from the state (e.g. those who do not file tax returns).

• Ensure an inclusive digital transformation. The digitalisation of services can save costs and simplify access for some users, but maintaining low-barrier, in-person support is essential for disadvantaged people who may lack the means to access services digitally. Good practices include combining digital offers with call centre and in-person options and providing training, intermediation and/or subsidies for devices. The digitalisation of social protection should follow much of the same guidance of digitalisation of governance more broadly: countries should prevent data breaches and manage them when they do occur, including through protective security frameworks, staff training, and data loss prevention tools. These efforts should be supported by a public office, such as a privacy or information commissioner. When using automated decision-making tools, including AI, governments must have appropriate accountability frameworks and transparent procedures in place to prevent and address discrimination and/or biases in automated systems.