Key takeaways
- RIAPA.ai makes economywide analysis more accessible. A conversational interface allows users to pose questions in plain language, run simulations, and explore results.
- AI simplifies the workflow, not the economics. The platform helps formulate scenarios and explain results; the underlying model produces the analysis.
- Testing found similar accuracy to experienced modelers, with simulations completed in minutes rather than weeks.
A policy shock rarely stops where it begins. A shift in crop productivity, import prices, public spending, or export demand can spread through production, trade, employment, incomes, and household welfare. Economywide models trace and assess these relationships, giving researchers and policymakers important insights into how countries are faring in a volatile global economy. But using these models has typically required specialized training and substantial time.
RIAPA.ai is designed to lower that practical barrier. Building on IFPRI’s Rural Investment and Policy Analysis (RIAPA) modeling system, this new artificial intelligence platform allows a user to describe an economic scenario in ordinary language, refine it through a guided conversation, run a country model, and explore the results through the same chat interface.
The objective is wider access to structured economic analysis—using AI to expand and accelerate research grounded in established models and expert assessments.
A more accessible route into economywide analysis
RIAPA.ai does not simplify difficult policy questions, nor does it eliminate the need to understand assumptions and uncertainty. What it changes is the cost of engaging with a rigorous model: users can move from a question to a reproducible simulation quickly, inspect how the scenario was formulated, and continue asking questions as they interpret the results.
That creates new possibilities for policy dialogue, training, rapid scenario exploration, and collaboration between modelers and the people who use economic evidence. The conversation becomes easier; the economics remains anchored in a formal model.
The model beneath the conversation
RIAPA uses country-level computable general equilibrium (CGE) models to simulate how producers, workers, households, government, and the rest of the world interact across an economy, casting light on these complex dynamics. (It is not a forecasting model.)
Because sectors and institutions are connected, the model can examine both direct and indirect economic effects. A simulation can show how an initial shock affects production and prices, how those changes alter factor incomes and household consumption, and how outcomes differ across population groups. This makes it possible to look beyond an aggregate result such as gross domestic product and examine distributional indicators, including poverty, food consumption, and rural and urban household welfare.
What the AI changes—and what it does not
RIAPA.ai adds a conversational and agent-based workflow around the economic model. When a user poses a question for potential modeling, the system helps formulate it as a model-ready scenario, checks whether key details are missing, executes the simulation, and explains the numerical results.
The underlying economic relationships are still solved by the CGE model. The large language model (LLM) powering the AI user interface does not generate the simulation results itself. This division of labor matters: the AI makes the workflow easier to navigate, while the formal model provides the accounting consistency and structural constraints needed for economywide analysis.
The current public platform focuses on the core RIAPA economic workflow. Users can ask about available data and model capabilities, formulate economic shock scenarios, review each before execution, and then interrogate or download the resulting analysis.
From a question to an interpretable result
Here is how the process works:
- Describe the question. State the policy, shock, country, and magnitude you want to examine in your own words.
- Refine the scenario. RIAPA.ai checks the formulation and asks follow-up questions when information needed by the model is unclear or incomplete.
- Review and run. The platform presents the model-ready scenario for confirmation, then executes the full economic model in the cloud.
- Explore the results. Ask follow-up questions, request tables or charts, and download the generated artifacts for closer inspection and reporting.
The system works in any web browser, including on a phone, and requires no local installation or knowledge of the model’s programming syntax. Complex scenarios still benefit from economic judgment, but users no longer need to operate the underlying software manually.
How did it perform against expert analysis?
To test the system’s reliability, IFPRI evaluated RIAPA.ai against experienced human CGE modelers and a general-purpose language-model. An independent panel of senior economists assessed responses to the same sets of economic scenarios. The system passed various benchmarks:
RIAPA.ai is as accurate as an expert modeler. There was no statistical difference in accuracy between RIAPA.ai and the human experts—both achieved similar scores on the evaluated tasks.
Run-time reduced from weeks to minutes. The automated AI workflow substantially reduced the time needed to formulate, run, and report a simulation.
Far more accurate than generic chatbots. Ordinary LLMs, without the full CGE model, failed to produce policy-ready answers to more than 60% of the study’s policy questions.
The study also observed manual processing errors in 7% of human-modeler submissions. RIAPA.ai avoided these workflow errors by automating the transfer of a scenario through the model pipeline. This does not mean that an AI-assisted system is error-free: scenario assumptions, model applicability, data limitations, and interpretation still require scrutiny.
Yet, these findings do not establish that RIAPA.ai is equivalent to experts in every setting. They do show why connecting an LLM to a formal economic system can be more reliable than just asking a general chatbot to emulate economywide analysis.
The full methods and results are reported in this IFPRI Discussion Paper.
Explore RIAPA.ai
- Visit RIAPA.ai.
- Register using your email address and a personal password.
- Use the chat interface to ask about available data or model capabilities, or directly describe a simulation you would like to run.
- If the system asks follow-up questions, continue responding. These validation steps help turn the request into a coherent, model-ready scenario before execution.
- Explore the Resources section and download the artifacts generated during the analysis. They provide useful detail for checking, interpreting, and communicating the results.
Building bridges from non-economic evidence to economic scenarios
Development of RIAPA.ai is continuing beyond the core economic workflow currently available to users. IFPRI and partner research centers are developing modules for creating scenarios for specific domains such as crops and climate, pests and diseases, livestock, and aquatic foods.
These specialist modules are intended to help users translate evidence from outside economics—for example, a crop-yield response or a livestock-production shock—into transparent, aggregated inputs that a RIAPA country model can simulate. The specialists will guide scenario formulation, document the evidence and calculations used, and identify gaps instead of inventing unsupported parameters.
These modules are under development and not yet part of the publicly accessible RIAPA-AI workflow. They will be introduced progressively after testing their data, scientific assumptions, and connections to the economic model.
Further reading and resources
- RIAPA.ai project page
- Mukashov et al. (2026), IFPRI Discussion Paper 2424
- RIAPA Data and Modeling System
- RIAPA.ai platform
Askar Mukashov is a Research Fellow with IFPRI’s Foresight and Policy Modeling (FPM) Unit; Eleanor Jones is an FPM Program Manager.
This work was supported by the Gates Foundation and the CGIAR Program on Policy Innovations.
