Problem

Zoa Research addresses the limitations of conventional quantitative forecasting, which is typically built for one narrow domain and requires highly specialized people to spend years testing features, tuning hyperparameters, and iterating models. That makes forecasting labor-intensive, slow to adapt, and poorly suited to cross-domain or event-driven shocks. The company’s near-term “killer” use case is proprietary trading: using short-horizon forecasts to identify mispriced risk and supply liquidity where markets need it. The same capability could eventually help organizations anticipate supply-chain volatility, energy imbalances, and other uncertain outcomes without building a large internal data-science team.

Product / Service

Zoa builds large-scale, domain-agnostic forecasting models that learn from broad collections of time-series and event data. Its technical approach combines cross-domain forecasting engines with LLM-driven optimization loops that automatically build, test, and improve models against quantitative metrics. The intended benefit is to discover patterns that specialized models or human intuition miss, while reducing the time and expertise required to create useful forecasts.

The company currently deploys its predictions in proprietary trading, while its external delivery model is Forecasting-as-a-Service: a secure API and dashboard for companies, NGOs, and public agencies. This would let customers consume high-accuracy forecasts and integrate them into planning or decision-making without hiring a full forecasting research organization; the public company materials indicate that this external productization is still an expansion path rather than a mature, widely documented product.

Market

Zoa operates at the intersection of AI foundation models, quantitative forecasting, predictive analytics, and financial technology. Its potential buyers include proprietary trading desks and hedge funds, supply-chain and operations teams, energy and commodity traders, laboratories and academics, and public-sector or nonprofit organizations. Named adjacent competitors include RavenPack and Kensho, which provide event, sentiment, data, and analytics infrastructure; Numerai, which turns machine-learning signals into trading strategies; and o9 Solutions, which provides enterprise supply-chain planning and forecasting. General-purpose forecasting-model providers such as Nixtla’s TimeGPT and Amazon’s Chronos are also adjacent alternatives, although Zoa differentiates itself by emphasizing cross-domain event forecasting and proprietary trading applications.

The company appears to have early, mixed traction rather than a scaled SaaS business. It was founded in 2024, joined Y Combinator’s Summer 2024 batch, is listed as active with five employees in New York, and describes its models as already trading; a secondary company profile characterizes that proprietary-trading operation as revenue-generating. Zoa’s site says it is backed by Y Combinator and others, while PitchBook reports $500,000 raised. However, the retrieved public materials do not establish the number of external Forecasting-as-a-Service customers or the scale of recurring software revenue, and PitchBook’s current-revenue field is blank, so the most defensible characterization is an active, funded, trading-led startup with external SaaS commercialization still developing.