TGO-001

Novissi -- Geospatial Poverty Mapping for Emergency Cash Transfer Area Targeting (Model 2)

Download PDF
Togo Sub-Saharan Africa Low income Full Production Deployment Confirmed

Government of Togo -- Ministry of Digital Economy and Digital Transformation (MENTD); inter-ministerial steering committee

At a Glance

What it does Prediction (including forecasting) — Vulnerability, needs and risk assessment, including predictive analytics
Who runs it Government of Togo -- Ministry of Digital Economy and Digital Transformation (MENTD); inter-ministerial steering committee
Programme Novissi Emergency Cash Transfer Programme -- Model 2 (Geospatial Area Targeting Component)
Confidence Confirmed
Deployment Status Full Production Deployment
Key Risks Data-related risks
Key Outcomes For the area-targeting component itself, the strongest documented technical result is that the geospatial model explained a substantial share of variation in wealth estimates at grid-cell and canton levels in validation exercises reported by the World Bank paper.
Source Quality 4 sources — Report (multilateral / development partner), Academic journal article, Working paper / technical note

The Government of Togo used a geospatial poverty-mapping model as the area-targeting step of Novissi Model 2, the rural expansion of its COVID-19 emergency cash-transfer programme. The retained sources clearly support that this component ranked Togo's cantons using satellite-derived and other geospatial indicators so that the poorest areas could be prioritised for programme coverage during the pandemic. The deployment is best understood as a government use of a research-developed targeting component inside a broader emergency programme, rather than as a standalone state-built system with extensive direct technical disclosure.

Novissi was launched in April 2020 to support informal workers affected by pandemic restrictions. In Togo, a small country in West Africa where over 50 percent of the population lives in poverty, the programme was described as an exemplary case of social protection in response to the COVID-19 pandemic in Africa. The government built and deployed a completely contactless, digital cash-transfer system within weeks. Beneficiaries registered using their mobile phones via a USSD menu and received mobile money transfers of approximately 15 US dollars per month for three months. When the programme expanded into rural areas later that year, the Government of Togo faced a familiar social-protection problem: it did not have a dynamic social registry, universal unique identifiers, or sufficiently recent canton-level poverty data to support conventional targeting. The last census had been conducted in 2011, and existing nationally representative household surveys could only produce poverty estimates at the national or regional level, not at the granular canton level needed for geographic targeting. The geospatial model was introduced to help identify which cantons should be eligible for the second phase of the programme.

The strongest source base shows that the model relied on multiple non-traditional data inputs, including satellite imagery, nightlights, road density, elevation, precipitation, urban-form indicators, and population density, together with household-survey data used as ground truth. Specifically, the research team used survey-based estimates of poverty from the EHCVM 2018-2019 nationally representative household survey as ground truth to train a machine-learning algorithm to estimate the wealth of very small regions at the 2.4-kilometre tile level based on geographic characteristics. The algorithm learned that certain patterns in the satellite imagery are indicative of wealth, such as places with metal roofs and high-quality roads, while others are indicative of poverty, such as places with certain types of terrain and weather. The tile-level consumption estimates were then overlaid with high-resolution population density estimates to calculate average per capita household consumption for each of Togo's 397 cantons. The resulting estimates were used to rank cantons from poorest to least poor, and the government selected the 100 poorest cantons for programme coverage, with that number chosen based on the estimated distribution of wealth of the population living in those cantons. The coverage was later expanded to 200 cantons. Publicly available documentation also makes clear that academic and development partners, including researchers from the University of California Berkeley's Center for Effective Global Action and Innovations for Poverty Action, played a substantial role in the technical design and implementation, while government officials validated outputs against operational knowledge and used the resulting rankings in programme administration.

This matters because the decision at issue was consequential: being inside or outside the selected cantons affected whether residents could proceed to the next stage of eligibility screening, where approximately 580,000 citizens lived in the initial 100 poorest cantons and approximately 57,000 were to receive benefits within those areas. At the same time, the public record is stronger on the overall Novissi methodology than on the internal government operating details for this specific area-targeting component. For production purposes, the safest framing is therefore that Togo operationally used a machine-learning-assisted geospatial ranking method in Novissi Model 2, with the methodology and limitations documented mainly through academic and World Bank materials.

The sources also document important limitations. Validation was imperfect, the model depended on pre-crisis survey inputs, and exclusion risks existed for groups poorly represented by the data environment or excluded by downstream administrative requirements. The IPA documentation notes that the research team planned in-person surveys and qualitative interviews designed to determine whether the algorithm's predictions were inadvertently biased against particular types of beneficiaries such as women, illiterate individuals, and marginalised subgroups. This remains a robust and important case, but it should stay framed as a partner-heavy targeting component within a government cash-transfer rollout rather than a highly transparent autonomous government AI system.

Classifications follow the DCI AI Hub Taxonomy. Hover over field labels for definitions.

Social Protection Functions

Implementation/delivery chain
Assessment of needs/conditions + enrolment primaryOutreach/communications/sensitisation
SP Pillar (Primary) The social protection branch: social assistance, social insurance, or labour market programmes. Social assistance
Programme Name Novissi Emergency Cash Transfer Programme -- Model 2 (Geospatial Area Targeting Component)
Programme Type The type of social protection programme, classified under social assistance, social insurance, or labour market programmes. View in glossary Emergency Cash Transfers
System Level Where in the social protection system the AI is applied: policy level, programme design, or implementation/delivery chain. View in glossary Implementation/delivery chain
Programme Description Large-scale unconditional emergency cash transfer programme launched in April 2020 to support informal workers affected by COVID-19. Model 2 extended coverage to rural areas using geospatial and phone-based targeting methods. This case covers only the geospatial area-targeting component used to prioritise cantons for inclusion.
Implementation Type How the AI output is produced: Classical ML, Deep learning, Foundation model, or Hybrid. Affects validation, compute requirements, and governance profile. View in glossary Classical ML
Lifecycle Stage Current stage in the AI lifecycle, from problem identification through to monitoring, maintenance and decommissioning. View in glossary Monitoring, Maintenance and Decommissioning
Model Provenance Origin of the AI model: developed in-house, adapted from open-source, commercial/proprietary, or accessed via third-party API. View in glossary Adapted from open-source
Compute Environment Where the AI system runs: on-premise, government cloud, commercial cloud, or edge/device. View in glossary Not documented
Sovereignty Quadrant Classification of data and compute sovereignty: I (Sovereign), II (Federated/Hybrid), III (Cloud with safeguards), or IV (Shared Innovation Zone). View in glossary IV — Shared Innovation Zone
Data Residency Where the data used by the AI system is stored: domestic, regional, or international. View in glossary International
Data Residency Detail Additional detail on the specific data hosting arrangements and jurisdictions. Model development and CDR analysis conducted by US-based academic researchers (UC Berkeley, Northwestern University); data stored on secure servers managed by the research team
Cross-Border Transfer Whether data crosses national borders, and if so, whether documented safeguards are in place. View in glossary Without documented safeguards
Decision Criticality The rights impact of the decision the AI supports. High criticality requires HITL oversight; moderate requires HOTL; low may operate HOOTL. View in glossary High
Human Oversight Type Level of human involvement: Human-in-the-Loop (active review), Human-on-the-Loop (monitoring), or Human-out-of-the-Loop (periodic audit). View in glossary HOTL
Development Process Whether the AI system was developed fully in-house, through a mix of in-house and third-party, or fully by an external provider. View in glossary Mix of in-house and third-party
Highest Risk Category The most significant structural risk source identified: data, model, operational, governance, or market/sovereignty risks. View in glossary Data-related risks
Risk Assessment Status Whether a formal risk assessment, informal assessment, or independent audit has been conducted for this system. Formal assessment

Impact Dimensions

Systemic and societal
  • Data minimisation controls
  • Grievance mechanism
  • Human oversight protocol
  • Independent evaluation
CategorySensitivityCross-System LinkageAvailabilityKey Constraints
Geospatial and remote sensing dataNon-personalLinks data across multiple systemsCurrently available and usedHigh-resolution satellite imagery (Digital Globe), nightlights/radiance (VIIRS), road density (OpenStreetMap), elevation/slope (USGS), precipitation (NASA/JAXA), population density (HDX), urban/built-up (NASA MODIS). 112 features extracted from 14 data categories. Satellite imagery required CNN pre-processing and PCA dimensionality reduction. Total of 10,119 wealth estimates at 2.4 km-squared tile level produced for Togo.
Survey and census dataPersonalLinks data across multiple systemsCurrently available and usedEHCVM 2018-2019 nationally representative household survey administered by INSEED. Covered 6,172 households across 265 cantons (65 percent) and 922 unique tiles (9.1 percent). Provided ground truth consumption data with exact geo-coordinates for model calibration. Most recent population census was completed in 2011.
Telecommunications and mobile dataNon-personalSingle source (no linkage)Currently available and usedFacebook connectivity data: cell tower counts, WiFi access points, mobile device counts by type (Android, iOS). Aggregated at grid cell level to mitigate privacy concerns. Used as features in the geospatial canton-level poverty model (step 1).

Lawson, C., Koudeka, M., Cardenas Martinez, A. L., Alberro Encinas, L. I. and Karippacheril, T. G. (2023) Novissi Togo: Harnessing Artificial Intelligence to Deliver Shock-Responsive Social Protection. Social Protection and Jobs Discussion Paper No. 2306. Washington, DC: World Bank Group and Government of Togo.

View source Report (multilateral / development partner)

Aiken, E., Bellue, S., Karlan, D., Udry, C. and Blumenstock, J. (2022) 'Machine learning and phone data can improve targeting of humanitarian aid', Nature, 603, pp. 864-870. doi:10.1038/s41586-022-04484-9.

View source Academic journal article

Innovations for Poverty Action (2021) 'Using Mobile Phone and Satellite Data to Target Emergency Cash Transfers in Togo', 12 January. Available at: https://poverty-action.org/using-mobile-phone-and-satellite-data-target-emergency-cash-transfers-togo (Accessed: 27 March 2026).

View source Working paper / technical note

World Bank (2021) 'Prioritizing the poorest and most vulnerable in West Africa: Togo's Novissi platform for social protection uses machine learning, geospatial analytics, and mobile phone metadata for the pandemic response'. Washington, DC: World Bank. Available at: https://www.worldbank.org/en/results/2021/04/13/prioritizing-the-poorest-and-most-vulnerable-in-west-africa-togo-s-novissi-platform-for-social-protection-uses-machine-l (Accessed: 24 March 2026).

View source Report (multilateral / development partner)
Deployment Status How far the system has progressed into real-world operational use, from concept/exploration through to scaled and institutionalised. View in glossary Full Production Deployment
Year Initiated The year the AI system was first initiated or development began. 2020
Scale / Coverage The scale and geographic or population coverage of the deployment. Geospatial poverty estimates produced for all 397 cantons in Togo using 10,119 grid cell-level predictions; 100 and later 200 poorest cantons selected for programme coverage in Novissi Model 2
Funding Source The source(s) of funding for the AI system development and deployment. Government of Togo National Solidarity and Economic Recovery Fund; French Development Agency (AFD) grant; GiveDirectly (rural expansion); International Development Association (IDA) / World Bank (research and data collection financing)
Technical Partners External technology vendors, academic partners, or development partners involved. Academic researchers from University of California Berkeley (CEGA, Data-Intensive Development Lab), Innovations for Poverty Action (IPA), GiveDirectly, and Joshua Blumenstock's research group. Mobile network operators Togocel and Moov Africa Togo provided anonymised CDR data. NASA Harvest (UC Berkeley School of Informatics) contributed geospatial methodology.
Outcomes / Results For the area-targeting component itself, the strongest documented technical result is that the geospatial model explained a substantial share of variation in wealth estimates at grid-cell and canton levels in validation exercises reported by the World Bank paper. At programme level, Novissi reached more than 920,000 beneficiaries overall, but those broader programme outcomes should not be treated as attributable solely to this area-targeting component.
Challenges Togo lacked a dynamic social registry, universal unique identifiers, and recent canton-level survey data, making traditional targeting infeasible. Poverty maps were validated mainly on asset-based wealth rather than consumption-based poverty, and no alternative consumption data existed for further validation. The consumption proxy was calibrated on pre-crisis 2018-2019 survey data. Exclusion risks affected populations without mobile phones (approximately 15 percent of households), those with low digital literacy (33 percent illiteracy rate), and individuals excluded by the voter registry requirement (foreign residents, underage individuals). Around 40 percent of mobile phone subscribers were unaware of the programme at the time they attempted to register (Aiken et al., 2021).

How to Cite

DCI AI Hub (2026). 'Novissi -- Geospatial Poverty Mapping for Emergency Cash Transfer Area Targeting (Model 2)', AI Hub AI Tracker, case TGO-001. Digital Convergence Initiative. Available at: https://socialprotectionai.org/use-case/TGO-001 [Accessed: 1 April 2026].

Change History

Created 30 Mar 2026, 08:41
by v2-import (import)