Better data first, then AI: Making aid work harder for people

By Gallina Andronova VinceletteHisham Abdo KahinLisandro Martin and Haishan Fu

Imagine a minister of finance preparing for a budget review. They pick one line, say primary health care, in a single province, and see every activity financed there: a multilateral loan, two bilateral grants and a foundation-funded program. For each one, they can follow the money further: which donor or country provides it, which agency channels it, which ministry or NGO implements it, which districts receive it, and how much was already committed and spent. They notice that three funders are active in one district, while none in the neighbouring one, and take that evidence to the next donor coordination meeting.

Parts of that picture can already be assembled from open data and the aid management systems some governments run. But for most governments, the complete picture is still out of reach. Closing the gap matters now more than ever. Artificial intelligence (AI) is often pitched as the shortcut; and indeed, it could help, but only if the data underneath is good enough.

A crowded system with less money

Over the past two decades, the foreign aid system has become crowded and fragmented. Between 2002 and 2021, the number of official finance providers grew from 62 to 112, while the number of donor agencies more than doubled, from 215 to 565. Over a similar period, the average official development assistance (ODA) grant size halved, from USD 1.7 million to USD 0.8 million. For countries with limited capacity to carry out foreign-funded projects, that means more donors, more projects, more reporting and higher costs just to manage it all.

Now, the system is also shrinking. ODA from members of the OECD Development Assistance Committee fell by over 23 percent in real terms in 2025, according to preliminary data, the largest annual drop on record. A crowded system with less money leaves no room for duplication or waste.

Why the full picture matters

An integrated view of development financing helps governments identify synergies, overlaps, and gaps across programs, enabling better coordination and more effective allocation of resources, whether for skills development, rural roads, or support to small businesses.

With this fuller picture, governments can match projects to national priorities, plan investments in the right order, and negotiate with partners based on evidence rather than guesswork. That is where transparency meets jobs: aid that goes where it counts is more likely to fund the services, infrastructure, and businesses that create them.

This is work the World Bank has taken on. For the 2026 Aid Transparency Index, a biennial ranking by the independent organisation Publish What You Fund of how openly aid organisations publish their data, we published the names of the organisations implementing our activities, along with transaction records showing who received each payment, so users could trace financing at both the project and the transaction level. We are also building shared reference lists with the governments we support. These are standard lists of codes for things like government ministries and agencies, so that each one is named the same way in every dataset, which makes data easier to connect. The results show: the International Development Association (IDA), the World Bank Group’s fund for low-income countries, ranked second among 34 organisations in the 2026 Aid Transparency Index. IDA scored 97.1 out of 100, up from 92.2 in 2024, with its biggest gain on networked data, meaning data that links funders, implementers and recipients. The ranking is not an end in itself, but it shows that openness can be put into practice.

Data comes first, then AI                                                                                

Most ministries cannot build this picture on their own. Two things can help, and the order matters.

The first is networked data. Donors and financiers need to publish who they fund, who carries out the work, and who receives the money, using shared codes that link related activities. Then aid stops being a pile of separate files and becomes a trail that can be followed from start to finish.

The second is AI. Today’s tools can already query raw data, tell funders apart from implementers, and process large volumes with far less technical effort than in the past. They work best on open, machine-readable data. For ministries with limited capacity, this could turn a specialist task into a routine question: which partners are funding primary health care in this province, and how much have they spent?

But AI also has limits. It cannot compensate for missing, inaccurate, or outdated information, and it does not replace expertise. Given poor data, it can produce confident answers that are simply wrong, and decisions based on them affect real people. Experts still need to validate and interpret what it produces. So the sequence matters: better data first, followed by AI to make it usable.

The work still ahead

Better data is one part of making every dollar work harder; speed in implementation is the other. An approved project cannot help anyone until it is ready to start, and delays between approval and start-up are a cost of their own. Through its Development-Ready (D-Ready) initiative, the World Bank is putting legal, operational, and capacity conditions in place before projects are approved, with the aim of reducing the time from Board approval to effectiveness, so that jobs, services, and other benefits reach people faster.

And there is more work to be done. Closing that gap will require coordination among financiers, donors and partner governments. Once those identifiers are in place, a ministry’s own budget data will be able to connect to the World Bank’s published data and that of others. The minister in our budget review would no longer need to piece the picture together. They would simply see it. We will keep publishing in ways that help every country see, and steer, the aid it receives toward jobs and opportunities.