A small business can buy access to an impressive AI service and still struggle to use it. The connection drops. The procedure it needs is scattered across old documents. Nobody has time to check the output. The problem begins before anybody types a prompt.
The World Bank’s Digital Progress and Trends Report 2025: Strengthening AI Foundations puts those conditions at the centre of the discussion. Its four foundations are connectivity, compute, context and competency. In plain language: access, computing capacity, useful data and skills. The report describes uneven conditions across countries, rather than one universal path to adoption.
What the report is saying
Connectivity includes the infrastructure needed to get online, with reliable electricity part of the foundation. Compute is the processing capacity behind AI services. Context means data that help systems fit local needs. Competency covers the skills needed to participate. The report argues that gaps in these foundations can limit how widely people benefit from AI.
That does not mean every organisation needs a data centre. A national infrastructure question and a small firm’s purchasing decision are different questions. The useful business question is narrower: which missing foundation is blocking this particular task?
Start with the failure you can see
Consider an illustrative repair business that wants help drafting customer updates. If staff cannot reliably access the service from the workshop, another prompt template will not fix the connection. A fallback that keeps the approved information available offline may matter more.
If the connection is sound but the draft invents a completion date, examine the source. Is the job record current? Does it distinguish an inspection from a completed repair? Are the latest details stored in one place? Better access to an outdated record still produces a weak foundation for a decision.
If the source is clear but nobody can check the answer confidently, the bottleneck is people’s capability and time. Give the reviewer a defined checklist, a route for uncertain cases and permission to reject the result. Calling that person the “human in the loop” does not create the time or authority they need.
Why smaller applications matter
The World Bank’s accompanying material highlights “Small AI”: comparatively accessible applications aimed at particular development problems. The practical lesson is to consider a bounded use case rather than assume that value requires building a frontier model. That is not a guarantee that a smaller system is cheaper or safer in every setting.
For a community organisation, the first useful experiment might be preparing a draft checklist from a public event procedure. It might be comparing two versions of a document and flagging changes for review. Neither task requires handing over decisions about who receives support or exposing private case records.
Local language is part of usefulness
A tool that handles a standard English demonstration well may still misunderstand the language people actually use with it. Test the wording, terminology and situations that matter in your setting. Use consented or fictional examples and ask people who understand the context to review the result. Do not turn “localisation” into a cosmetic change of place names.
Keep the comparison specific. Can the tool preserve the meaning of an instruction? Does it ask when a phrase is ambiguous? Does it quietly replace a local term with an inaccurate assumption? A small test cannot establish performance for every speaker, but it can expose a poor fit before a wider release.
What to do with this tomorrow
Choose one task and write four short notes: how people will access the tool; what computing service it depends on; which approved information it needs; and who can judge the result. Add a manual fallback and a stop condition. Then test the full workflow, including review time.
Find the missing foundation before you buy another layer of AI.
This is an original application of the report’s ideas, not a World Bank endorsement of a product or a forecast for your organisation. The report helps frame the questions. Your own evidence has to support the decision.
Sources & further reading
World Bank: Digital Progress and Trends Report 2025 — report overview ↗World Bank: Strengthening AI Foundations — key findings and practical opportunities ↗Sources checked 2026-10-19. Examples and recommendations are original editorial analysis.