evidoria

← Back to browse

Good practice Imported

Pulse Lab Jakarta — AI nowcasting of food prices for Indonesia's national policy monitoring

Indonesia · Jakarta · See the Indonesia profile · See the Jakarta profile

Evidence: Observational / pre–post Top 66% 53/100 · Ask Evidence Copilot about this practice

Pulse Lab Jakarta, a joint Bappenas/UN Global Pulse facility, built a machine-learning model nowcasting Indonesian food prices from Twitter data, validated over 15 months and extended via a 200-reporter, 65,000-datapoint citizen-pricing pilot in West Nusa Tenggara.

15 months
Validation period (2018)
4 commodities
Commodities tracked
200+ volunteers
Citizen reporters recruited (NTB pilot)
65,000+ data points
Price data points collected (NTB pilot) (10 weeks)

Details

Promoter
Pulse Lab Jakarta (Bappenas & UN Global Pulse), with WFP and FAO
Period
{'end': '2018 validation; lab ongoing as UN Global Pulse Asia Pacific', 'start': '2014'}
Keywords
machine learning, nowcasting, food security, social media analytics, policy monitoring

Context

Pulse Lab Jakarta, a joint facility of Indonesia's Bappenas and UN Global Pulse (rebranded UN Global Pulse Asia Pacific in 2023), built a machine-learning model that mines public Twitter posts to estimate near-real-time food prices for beef, chicken, onion and chili. The initiative, running since 2014 in collaboration with WFP and FAO, was designed to provide an early-warning input alongside official statistics for national food-security and inflation policymaking.

Objectives

The goal was to complement Indonesia's official price statistics with a faster, social-media-derived nowcasting signal that policymakers could use for food-security and inflation monitoring.

Activities

The core system mines Twitter data with machine-learning models to nowcast prices for four commodities. A secondary citizen-reporter pilot in West Nusa Tenggara (NTB) recruited 200+ volunteer reporters online, compensated with mobile-phone credit, who submitted price data via a mobile app.

Results

Over roughly 15 months of validation in 2018, the Twitter-based nowcasts were reported as 'closely correlated' with official government price data, though an exact correlation coefficient is not independently verifiable from accessible sources. The NTB citizen-reporter pilot collected more than 65,000 individual price data points over 10 weeks.

Conclusions

The approach was validated only for four commodities nationally and piloted as an extension in a single province, so scalability evidence is limited; it is unclear from available sources whether the original nowcasting tool remains operationally embedded in government workflows or served mainly as a prototype that informed later systems under UN Global Pulse Asia Pacific.

Implementation
Indicative cost
Low (< €50k) — No budget figures are available in accessible sources; the core system was primarily a research/data-science effort (social-media mining), while the NTB pilot's main disclosed cost was mobile-phone-credit compensation for 200+ volunteers — together suggesting a low direct programme cost, though total lab operating costs are not disclosed.
Time to results
Long (> 3 years) — The initiative began in 2014, was validated over roughly 15 months in 2018, and the hosting lab continued operating (rebranded UN Global Pulse Asia Pacific in 2023), indicating a long, decade-plus lifespan even though the original tool's current operational status is unclear.
Staffing & skills
Pulse Lab Jakarta / UN Global Pulse Asia Pacific team (model development), Indonesia's Ministry of National Development Planning (Bappenas) (government partner), World Food Programme and FAO (technical/food-security collaborators), 200+ volunteer citizen reporters recruited for the NTB pilot extension

Conditions for success

  • Access to public Twitter data for price-mention mining
  • Official government price statistics available as a validation benchmark
  • Mobile-phone credit incentive model to sustain citizen-reporter participation
  • Standing institutional partnership between Bappenas and UN Global Pulse

Common failure modes

  • Validation covered only four commodities and relied on social-media signals that may not generalize to other goods or regions
  • Citizen-reporter extension was a single-province, 10-week pilot, not a national rollout
  • Available sources do not confirm whether the tool is still operationally used in government decision-making today

Where it fits

Governance type
joint government-UN innovation lab
Scale
national nowcasting (4 commodities) plus single-province citizen-reporter pilot
Income level
lower-middle-income (Indonesia)

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Similar practices you may find useful