Blab AI
AI Shortcuts to State Capacity: Lessons from Togo and India
When a satellite image and a mobile‑phone call record can decide if a family receives a cash transfer, the line between data science and social welfare blurs.
In Togo, the Novissi program fused high‑resolution satellite imagery, mobile‑phone call‑record data and machine‑learning models to pinpoint households living in extreme poverty during the COVID‑19 pandemic. The system identified roughly 140,000 people who would have been missed by conventional targeting lists and routed cash transfers through mobile‑money platforms.
India’s more recent initiative, the Local Economic Intelligence Platform, aggregates tax filings, business registrations and other public sources to give local authorities a clearer view of economic activity in their districts. An IMF working paper notes that when businesses and governments digitise together, tax‑to‑GDP ratios can rise by up to three percentage points—a gain that could be decisive for states that collect less than 15 % of GDP in taxes.
Digital identity and payments infrastructure further illustrate AI‑enabled data processing. India’s Aadhaar‑based identity system, coupled with a national payments network, reportedly saved the state about $37 billion by eliminating duplicate and fictitious beneficiaries from welfare rolls. In Malawi, biometric identification has expanded formal credit access to borrowers who were previously invisible to the system.
These cases show that AI can solve the information bottleneck that has long constrained state capacity. By drawing on data generated outside the state—satellite images, phone records, payment flows—AI substitutes for the large‑scale information‑processing tasks that are the core of many public services.
However, the technology does not automatically confer legitimacy or trust. Togo researchers found the algorithm performed best when used to supplement, rather than replace, existing targeting mechanisms. In a system already plagued by dysfunction, AI can increase throughput without improving fairness—a situation scholars describe as "premature load‑bearing."
Political risks also arise when foreign vendors build and run systems that decide who receives welfare or pays taxes. History shows that foreign‑led revenue collection can become a symbol of compromised sovereignty, as with China’s Maritime Customs Service of the 19th and early 20th centuries. Donor‑run projects can deliver efficient services but may bypass domestic institutions, eroding the consent of the governed.
In short, AI can compress the timeline for building administrative capacity, but it is not a substitute for the slow work of institution building. States that deploy imported capability transparently and under domestic control, while continuing to strengthen accountability and the rule of law, are more likely to gain both efficiency and legitimacy. Those that mistake an automated interface for a functioning state risk automating only the appearance of governance.
The coming decade will test whether AI can be harnessed to build, rather than bypass, the institutional foundations that sustain long‑term development.
In Togo, the Novissi program fused high‑resolution satellite imagery, mobile‑phone call‑record data and machine‑learning models to pinpoint households living in extreme poverty during the COVID‑19 pandemic. The system identified roughly 140,000 people who would have been missed by conventional targeting lists and routed cash transfers through mobile‑money platforms.
India’s more recent initiative, the Local Economic Intelligence Platform, aggregates tax filings, business registrations and other public sources to give local authorities a clearer view of economic activity in their districts. An IMF working paper notes that when businesses and governments digitise together, tax‑to‑GDP ratios can rise by up to three percentage points—a gain that could be decisive for states that collect less than 15 % of GDP in taxes.
Digital identity and payments infrastructure further illustrate AI‑enabled data processing. India’s Aadhaar‑based identity system, coupled with a national payments network, reportedly saved the state about $37 billion by eliminating duplicate and fictitious beneficiaries from welfare rolls. In Malawi, biometric identification has expanded formal credit access to borrowers who were previously invisible to the system.
These cases show that AI can solve the information bottleneck that has long constrained state capacity. By drawing on data generated outside the state—satellite images, phone records, payment flows—AI substitutes for the large‑scale information‑processing tasks that are the core of many public services.
However, the technology does not automatically confer legitimacy or trust. Togo researchers found the algorithm performed best when used to supplement, rather than replace, existing targeting mechanisms. In a system already plagued by dysfunction, AI can increase throughput without improving fairness—a situation scholars describe as "premature load‑bearing."
Political risks also arise when foreign vendors build and run systems that decide who receives welfare or pays taxes. History shows that foreign‑led revenue collection can become a symbol of compromised sovereignty, as with China’s Maritime Customs Service of the 19th and early 20th centuries. Donor‑run projects can deliver efficient services but may bypass domestic institutions, eroding the consent of the governed.
In short, AI can compress the timeline for building administrative capacity, but it is not a substitute for the slow work of institution building. States that deploy imported capability transparently and under domestic control, while continuing to strengthen accountability and the rule of law, are more likely to gain both efficiency and legitimacy. Those that mistake an automated interface for a functioning state risk automating only the appearance of governance.
The coming decade will test whether AI can be harnessed to build, rather than bypass, the institutional foundations that sustain long‑term development.
