AI can support affordable, nutritionally adequate household diets

Managing food costs while meeting nutritional needs has become a persistent challenge rather than a short-term adjustment. Digital tools exist to track spending or suggest meals, but they rarely address both problems at once, and almost never adapt automatically when prices shift.

New research proposes a unified solution built around agentic artificial intelligence, a form of AI designed for continuous planning, monitoring, and autonomous decision-making. The study presents a system that does not merely recommend meals or track expenses, but actively balances household budgets, nutritional adequacy, health conditions, and real-time food prices to generate adaptive meal plans that evolve as market conditions change.

The findings are detailed in the study FinAgent: An Agentic AI Framework Integrating Personal Finance and Nutrition Planning, presented at the IEEE International Conference on Computing and Applications and released as an arXiv preprint. 

A combined approach to budgeting and nutrition

Financial applications focus on tracking income and expenses but ignore dietary quality and health. Nutrition and diet planners emphasize nutrient intake but typically disregard household budgets, shared family constraints, and fluctuating food prices. According to the authors, this separation leads to suboptimal outcomes, especially for families operating under tight financial limits.

FinAgent addresses this gap by treating household food planning as a single, integrated optimization problem. The system takes household income, fixed expenses, family size, age distribution, health conditions, dietary restrictions, and cultural preferences as inputs. These are combined with authoritative nutritional guidelines and real-time supermarket price data to calculate a weekly food budget and aggregate household nutrient requirements.

Under the hood, the framework is a modular multi-agent architecture. Each agent is responsible for a specific function, such as budgeting, nutrition analysis, health personalization, cultural and dietary rule enforcement, price monitoring, substitution planning, procurement, and explanation. Rather than operating independently, these agents coordinate through a shared knowledge base that includes food ontologies, nutrient values, household profiles, and price histories.

Meal planning is formulated as a cost-minimization problem subject to strict constraints. The system seeks to minimize total food expenditure while ensuring that household-level nutritional requirements are met and the weekly budget is not exceeded. When food prices change beyond a predefined threshold, the price monitoring agent triggers re-optimization. The substitution agent then proposes nutritionally equivalent alternatives to maintain diet quality at lower cost, ensuring continuity rather than disruption.

This approach allows FinAgent to move beyond static planning. Instead of producing a fixed weekly menu, the system continuously adapts, re-planning meals in response to market volatility, health needs, or budget changes.

Agentic AI enables adaptive household decision-making

Traditional AI tools respond to user queries but lack long-term goals or autonomous replanning. In contrast, agentic AI operates through ongoing action loops that combine perception, memory, planning, monitoring, and execution.

In FinAgent, this capability allows the system to function as an autonomous household decision-support agent. The budgeting agent calculates the disposable weekly food budget based on income and expenses. The nutrition agent ensures macro- and micronutrient adequacy across all household members. The health personalization agent adjusts nutrient targets for conditions such as diabetes, sodium limits, or deficiencies. Cultural and preference constraints are enforced to respect halal rules, local cuisine, and seasonal foods.

Crucially, the price monitoring agent continuously tracks supermarket prices using APIs and verified scrapers. When significant price shocks occur, such as sudden increases in staple foods, the system automatically recalculates meal plans using substitution graphs that preserve nutritional balance while minimizing cost increases.

The inclusion of an explainer agent addresses transparency and trust. Rather than producing opaque recommendations, the system provides structured reasoning for its decisions, allowing users to understand why substitutions were made or how cost savings were achieved. This design choice reflects growing concern about explainability and accountability in AI-driven decision systems.

The study emphasizes that adaptability is not optional in household planning. Food prices can fluctuate by 10 to 30 percent over short periods, particularly in inflationary environments. Static meal plans, even when optimized once, quickly become unaffordable or nutritionally imbalanced. FinAgent’s architecture is explicitly designed to remain stable under such conditions.

Cost savings and nutritional adequacy under real-world conditions

To evaluate the framework, the researchers conducted extensive simulations alongside a real household case study. Synthetic experiments were run on hundreds of simulated households with varying incomes, family sizes, dietary preferences, and health conditions. These simulations tested the system’s performance under different price shock scenarios and planning baselines.

The results show consistent cost reductions compared with fixed menus, static optimization, and manual planning approaches. Across multiple trials, the agentic AI framework reduced weekly food costs by approximately 12 to 18 percent while maintaining nutritional adequacy above 95 percent for key nutrients such as protein, iron, calcium, and vitamin D.

Adaptability proved to be one of the system’s strongest features. When major food items experienced price increases of up to 30 percent, FinAgent successfully re-planned meals without exceeding the household budget or compromising nutrient coverage. In contrast, static plans frequently exceeded budget limits under the same conditions.

A four-week real household case study conducted with a Saudi family further demonstrated practical feasibility. The household, with a monthly income of approximately 10,000 SAR, experienced an average grocery cost reduction of around 17 percent while maintaining high nutrient adequacy and adherence to cultural dietary requirements. Price increases in commonly consumed items were offset through nutritionally equivalent substitutions, preserving both affordability and meal diversity.

Ablation experiments reinforced the importance of each agent within the system. Removing the price monitoring function led to significant cost increases. Disabling health personalization reduced nutrient adequacy for certain micronutrients. Eliminating preference enforcement resulted in repetitive menus and lower user satisfaction. These findings support the authors’ claim that the system’s performance depends on coordinated multi-agent interaction rather than isolated optimization.

The study also addresses ethical and practical considerations. Household financial and health data are sensitive, and the system is designed with privacy-first principles, including controlled data retention and secure processing. The authors acknowledge limitations related to data availability, noting that incomplete price feeds or nutrient databases can affect performance and should be addressed through regional datasets and improved data integration.

In its conclusion, the research positions FinAgent as a practical demonstration of how agentic AI can support everyday household decision-making rather than abstract optimization tasks. By unifying budgeting, nutrition science, and real-time pricing into a single adaptive framework, the system offers a pathway toward more resilient and equitable household planning.

The study’s implications extend beyond individual families. Rising food insecurity, diet-related health risks, and inflationary pressure are global challenges. Tools that help households balance cost and nutrition dynamically could support broader public health and sustainability goals without requiring constant manual intervention.

While the system remains a prototype, the authors outline future directions that include large-scale deployment, long-term user studies, integration with subsidy and loyalty programs, and enhanced governance mechanisms to ensure transparency and accountability. As agentic AI continues to move from theory into applied domains, FinAgent illustrates how autonomous systems can be designed to serve human welfare goals rather than convenience alone.

In a landscape crowded with financial trackers and diet apps, the study makes a clear claim. Effective household decision-making requires systems that understand budgets, health, culture, and markets as interconnected realities. Agentic AI, when carefully structured and evaluated, may offer a way to manage that complexity without sacrificing affordability or nutrition.