Timely, high-quality food price data is essential for shock responsive decision-making. However, in many low- and middle-income countries, such data is often delayed, limited in geographic coverage, or unavailable due to operational constraints. Traditional price monitoring, which relies on structured surveys conducted by trained enumerators, is often constrained by challenges related to cost, frequency, and reach.
To help overcome these limitations, the World Bank launched the Real-Time Prices (RTP) data platform. This effort provides monthly price data using a machine learning framework. The models combine survey results with predictions derived from observations in nearby markets and related commodities. This approach helps fill gaps in local price data across a basket of goods, enabling real-time monitoring of inflation dynamics even when survey data is incomplete or irregular.
Comparing three alternatives for price data collection
In parallel, new approaches—such as citizen-submitted (crowdsourced) data—are being explored to complement conventional data collection methods. These crowdsourced data were recently published in a Nature Scientific Data paper. While the adoption of these innovations is accelerating, maintaining trust requires rigorous validation.
A newly published study in PLOS compares the two emerging methods with the traditional, enumerator-led gold standard, providing new evidence that both crowdsourced and AI-imputed prices can serve as credible, timely alternatives to traditional ground-truth data collection—especially in contexts where conventional methods face limitations (Figure 1).
Figure 1. Reproduced from Adewopo et al. (2025) this figure compares AI-imputed and crowdsourced data with traditional enumerator-led surveys in northern Nigeria. It shows monthly data for maize (R = 0.99, R² = 0.98) and rice (R = 0.93, R² = 0.87) over a three-year period (2021–2023).
Figure 2: Reproduced from Adewopo et al. (2025), this figure compares crowdsourced data with enumerator-led survey data in the final year of the study, shown at daily, weekly, and monthly frequencies.
Setup of the ground-truth exercise
The study compares three distinct sources of price data to evaluate their accuracy and reliability:
- AI-imputed prices from the World Bank’s RTP platform.
- Crowdsourced prices submitted by local volunteers via a mobile application part of the Food Price Crowdsourcing in Africa (FPCA) project.
- Enumerator-collected prices gathered using conventional survey methods during the final year of the FPCA project.
The AI-imputed data are generated from decentralized survey inputs provided by partners such as the World Food Programme and the Food and Agriculture Organization, based on prices collected from local sources and field staff. The imputed prices are generated by machine learning models designed to estimate missing values by drawing on related commodities and exchange rates as well as observations from nearby locations or time periods.
These estimates are compared with results from the FPCA project that collected over 100,000 price entries across more than 150 market locations in northern Nigeria. Volunteers submitted prices from different points along the supply chain, including farm gate, wholesale, and retail. The locations are visualized in Figure 3.
Figure 3: Reproduced from Adewopo et al. (2025), this map shows the study area in the core northern region of Nigeria. It displays the georeferenced locations of enumerators and volunteer citizens (crowd) who submitted commodity price data under the FPCA project, overlaid with target market locations used for AI-based price imputation.
To provide a benchmark, these data were complemented by enumerator surveys conducted at fixed market locations using standardized protocols to ensure consistency over time.
Figure 4 summarizes the overall data flows used in the comparison study. This multi-source setup enabled a robust comparison of price information collected through diverse methods and under real-time field conditions.
Figure 4. Schema for validation of real-time data innovations to monitor commodity prices in data-scarce environments.
The comparison revealed a strong statistical alignment between the three data sources. For maize, correlation coefficients reached as high as 0.99 for some price pairs. Observed differences remained within the expected range of measurement variability and did not suggest systemic bias or methodological weaknesses. Importantly, the analysis found no evidence of time lag in the AI-generated estimates, indicating that these models can reliably reflect current market conditions.
The findings support the integration of non-traditional data sources into national market monitoring systems. In particular, they highlight the potential for cost-effective, scalable, and timely data collection that extends coverage beyond what is typically feasible through enumerator-led surveys alone.
Implications for policy and implementation
These insights are already informing practice. Encouraged by the pilot results, the National Bureau of Statistics (NBS) of Nigeria announced last year that they would deploy crowdsourcing and AI imputation methods to track local food prices in real-time. With technical support from the World Bank’s Development Data Group, they now launched a national crowdsourced price monitoring dashboard. This marks a shift in how official data systems are being designed and deployed.
AI imputation and crowdsourcing are both part of a broader effort by the World Bank to make food security data systems more scalable, timely, and cost-effective. The adoption of these faster methods of data production is also enabling the automated detection of food crisis risks. For example, the Joint Monitoring and Report (JMR) that is live in Yemen and Somalia uses the RTP data to raise alerts using a threshold model. This is initiative is part of national Preparedness Plans for Food and Nutrition Security Crises that define what constitutes a crisis, how it is monitored, and what actions should be taken with risks emerge. An example of this coupling of data and action is the World Bank’s Crisis Response Window Early Response Financing (CRW-ERF) which has been designed to rapidly disburse when early-warning indicators signal potential food crises.
📚 Further Reading
The newly published paper offers a robust empirical case for scaling the use of crowdsourced and AI-imputed price data. As global food systems face growing threats from conflict, climate change, and economic shocks, the need for timely, reliable, and cost-effective market monitoring has never been greater.
· 📄 Full Open-Access Article
Dive deeper into the methodology and findings in our paper published in PLOS ONE:
https://doi.org/10.1371/journal.pone.0320720
· 📊 Detecting Food Crisis Risk with High-Frequency Data
Explore how we use real-time price monitoring in the Joint Monitoring Report (blog)
· 🧠 Why Better Food Security Data Matters
Learn more about the role of better data systems in combating hunger: Five Alarming Stats on Global Hunger (blog)
· 🛠️ From Data to Preparedness
Watch how these initiatives support early action and planning: Preparedness Plans (YouTube)
· 🌍 Our Latest Work on Food Security Data
Watch an overview of our recent data innovations: Reshaping Food Security Analytics (YouTube)
This work was carried out with the support from Food Systems 2030.
Join the Conversation