AI Models Can Now Overhaul Policies Like GST. But Should They Be Trusted?

For years, public policy and governance were evaluated using simple economic tools that, while providing a detailed look at a single economic component, usually examined an industry in isolation. Also, traditional economic forecasting relies on historical data, econometric models, and periodic reporting, often resulting in time lags and reduced accuracy during periods of economic volatility.

In contrast, a Computable General Equilibrium (CGE) model quantifies the interrelatedness of the real macro and micro economies, viewing the economy as deeply interconnected and cognizant of the reality that a policy may have unintended consequences.

The Indian government and economists are now using the tool to understand different sectors’ responses to policy interventions. These models simulate interactions between sectors, consumers, industries, taxation systems, trade flows, and labour markets.

For example, if a government increases GST on tyres, economists can use CGE models to estimate how the decision could affect manufacturing costs, household spending, inflation, transportation, and downstream industries.

With AI taking over every sector, the Indian government has also made its first move towards incorporating AI in its economic policymaking.

Recently, the Ministry of Electronics and Information Technology constituted the AI Governance and Economic Group, a high-level inter-ministerial body that will serve as India’s central institutional mechanism for AI governance policy development and coordination.

Till now, policymakers have been relying on external CGE tools to assess policy impact. Researchers in India, particularly the National Council of Applied Economic Research (NCAER), have collaborated with Australia’s Victoria University to develop and utilise CGE models tailored for the Indian economy.

“AI enhances the ability of CGE models to predict the impact of shocks, such as free trade agreements, global supply chain disruptions, or Union Budget proposals, on different sectors, employment levels, and household incomes,” Raaj Kumar Banerjee, Co-founder of Isource, tells AIM.

However, the strategic need to develop a homegrown CGE model is becoming more evident.

This is what Isource is attempting by bringing AI into the complex and traditionally manual domains of economic policymaking. The Delhi-based company is developing AI-powered CGE models to help governments, researchers, and institutions simulate the impact of policy decisions more quickly and efficiently.

From GST revisions and agricultural subsidies to welfare schemes and manufacturing incentives, the startup’s platform aims to analyse how economic shocks ripple across industries, households, employment, pricing, and production systems in near real-time.

“We are letting economists focus on the analysis and interpretation rather than doing all the clerical work manually,” Banerjee states. The broader ambition is to move policymaking toward faster, data-driven decision-making powered by AI-assisted simulations.

AI-Assisted Policymaking

Governments today are under pressure to respond quickly to inflation, unemployment, supply chain disruptions, climate risks, and welfare demands. Traditional economic modelling methods often struggle to match the speed at which policy decisions now need to be made.

Banerjee believes Isource’s AI-assisted CGE systems could eventually help governments make more cautious and evidence-based decisions. The startup sees potential use cases across taxation policy, agricultural planning, subsidy optimisation, industrial policy, welfare distribution, and trade strategy.

However, building CGE models remains highly labour-intensive and dependent on extensive manual data preparation.

According to Banerjee, one of the largest bottlenecks lies in converting Supply Use Tables (SUT) into Input-Output Tables (IOT), which form the foundation for policy simulations.

“Earlier, this conversion had to be done manually. A process that could take seven to 10 days, and sometimes even months, can now be completed in around 10 minutes using AI,” the co-founder explains.

The startup claims its AI engine automates much of the repetitive economic modelling workflow. Once the IOT is generated, economists typically build what is known as a Social Accounting Matrix (SAM). The SAM acts as the baseline dataset for CGE analysis, mapping relationships between industries, households, governments, taxes, and trade flows.

Traditionally, creating a SAM requires significant technical expertise and manual adjustments, which AI can simplify. The startup believes this could dramatically improve the speed at which governments evaluate policy trade-offs.

A Homegrown Stack

Isource argues that domestic policy systems require local context, localised datasets, and India-specific economic assumptions.

The company has already initiated discussions with several state governments, including Maharashtra, and is also exploring partnerships with policy institutions and research organisations. “Many stakeholders want homegrown solutions for policymaking and economic analysis,” he states.

The firm is additionally working on patent filings for its proprietary algorithms.

Structural Limitations

Despite the excitement around AI-driven economic modelling, several experts remain cautious about its practical reliability.

Sanjib Pohit, Professor and Thematic Lead for CGE Modelling and Policy Analysis at NCAER, believes the biggest challenge is that most AI systems are trained on economic assumptions rooted in developed economies. “The CGE model literature and its application in developing countries are still narrow,” he mentions.

Meanwhile, developing economies such as India function under conditions very different from those in advanced economies. “Market rigidities, product segmentation, shortage of capital, and absence of exit policy for labour are hallmarks of developing countries. These are not the characteristics of developed economies on which an AI-based CGE model draws their knowledge,” he explains.

Pohit also raises concerns around transparency and quality control in AI-generated economic models. “In a product, there is a quality check like the ISI standard. However, where is the quality check that AI has done correctly? You just have to believe it,” he states.

Economic modelling often involves multiple intermediate calculations, assumptions, balancing steps, and corrections before releasing a final model. Pohit also explains that AI systems frequently hide those intermediate processes. “If one asks the intermediate steps of data construction, one can see the error it commits,” he notes.

He further warns that AI systems may produce outputs even when the underlying economic logic is flawed. “The beauty of an AI-based CGE model is that it would always give you an answer. It will never fail. Whereas, in reality, a CGE model may not always give a result or may not solve,” Pohit explains.

Can AI Replace Economists?

While AI can automate repetitive workflows, Pohit argues that economic modelling still depends heavily on human judgement and contextual understanding. “There is no standard CGE model. Modellers build a model depending on the issues to study and the underlying conditions of the economy. That knowledge comes from the modeller’s knowledge,” Pohit says.

He believes economists must first understand how to build models manually before relying on AI-assisted systems. “Just like a driver needs to learn to drive, it is best to first build a CGE model without an AI assistant before taking recourse to an AI-based CGE model,” he adds.

This is among the reasons why AI-based CGE systems have not yet seen widespread global adoption. But that is now beginning to change.