🛟 Why Top Banks Are Investing in AI to Predict Markets (and Winning)

👋 Welcome, thank you for joining us here :)

The AI Check In is your weekly power play to navigate AI governance and strategy in banking and finance.

What to expect this edition:

• 🛟 Need to Know: Govern AI-Driven Market Forecasts

• 🥷 Deep Dive: How Banks Are Deploying AI Forecasting

• ⚔️ Vendor Spotlight: Kensho AI from S&P

• 📈 Trends to Watch: What’s Emerging in AI Forecasts

Rewrite the rules. Outmaneuver competitors. Stay two steps ahead of regulators.

Let the games begin.

🛟 Need to Know: Govern AI-Driven Market Forecasts

AI forecasting is expanding across interest rates, FX, equities, credit, commodities, and liquidity markets. Yet governance frameworks, especially in the U.S., are not keeping pace with deployment.

1. Model Oversight: Regulators, including the Federal Reserve and OCC, are applying SR 11-7 to AI tools that influence credit, trading, or pricing. JPMorgan applies its Model Risk Governance and Review (MRGR) framework to forecasting models, requiring formal validation, tiering, and approval.

2. Data & Decision Integrity: Forecasting models ingest earnings calls, rate paths, geopolitical risk, and market sentiment at scale. Exposure to volatility-based errors, esp. forecasting FX and rates, should be noted. Boards must ensure data provenance and define override controls, particularly when outputs support client portfolios or capital allocation.

3. Global Adoption Outpacing Regulation: RBC launched an AI unit inside its capital markets group. Bank of America uses Glass—an AI-powered market analysis platform—to help identify trends and guide client advice. In China, High Flyer Quant integrates predictive AI into end-to-end trading using only machine learning and trading data.

Together, these signal a strategic shift: AI-powered market forecasting is operational, not theoretical. Oversight must now catch up.

🥊 Your Move:

1. Extend SR 11-7 protocols to market-facing models
   Ensure AI-driven forecasting tools—especially those influencing trading or pricing—are formally tiered by market impact, validated, and override-capable under existing model risk guidelines.

2. Benchmark against global leaders—and bring the board in
   Compare your AI forecasting posture to institutions like JPMorgan (governance), RBC (capital markets AI deployment), and Bank of America (Glass platform). Establish board-level visibility into any market-facing AI use cases—especially where forecasts shape client portfolios or asset pricing.

3. Audit data inputs and narrative risk
   Require model validation teams to document all unstructured data sources (e.g., earnings calls, sentiment signals, geopolitics), and log where AI-generated forecasts are integrated into client-facing or regulatory narratives.

🥷 Deep Dive: How Banks Are Deploying AI Forecasting

AI forecasting has moved beyond experimentation. In leading institutions, these systems are now embedded into core workflows, from treasury and advisory to trading and market risk.

While each bank chooses its own architecture, what separates scalable forecasting from pilot noise is not only the algorithm. It is how the system is governed, validated, and wired directly into decisions that affect clients, capital, and control.

Case Study 1: JPMorgan Chase - Forecasting Built for Treasury Intelligence

JPMorgan’s Cash Flow Intelligence (CFI) is one of the bank’s flagship AI platforms in treasury and forecasting. It provides corporate clients with forward-looking cash flow estimates based on internal transaction data, ERP feeds, behavioral patterns, and market signals.

The platform combines multiple models, including neural networks, random forests, and ensemble methods. Natural language processing is used to integrate signals from unstructured sources such as economic briefings and research notes.

Forecasts are delivered in real time and adjust as new data flows in. Clients using CFI have reduced manual forecasting efforts by up to 90 percent and accelerated their treasury decision-making cycles.

CFI sits within the firm’s Model Risk Governance and Review (MRGR) structure. Models are formally tiered, validated, and monitored. Output is designed to be explainable to users and challengeable by supervisors. This combination of real-time prediction and institutional oversight has positioned CFI as both a productivity engine and a compliance-ready forecasting layer.

Case Study 2: RBC - Dynamic Execution through Reinforcement Learning

In May 2025, Royal Bank of Canada launched an AI and digital innovation team in its capital markets division. Their platform, Aiden, designed in partnership with Borealis AI, is used for electronic trading and macro signal interpretation.

Aiden uses deep reinforcement learning to train itself on market feedback and trading execution outcomes. It processes more than 32 million calculations per trade order and selects strategies based on constantly shifting liquidity, volatility, and price conditions. Tools such as Aiden VWAP and Aiden Arrival have been tuned to outperform rule-based benchmarks in terms of execution quality and cost efficiency.

Post-trade explainability reports are provided to clients and internal supervisors. While Aiden operates at speed, the bank retains governance oversight through validation protocols and model performance reviews.

RBC expects its AI initiatives to deliver up to C$1 billion in new business value, a signal that forecasting is no longer an experiment but a commercial advantage.

Case Study 3: Bank of America - Forecasting at Scale Across Banking Channels

Bank of America is directing $4B of its 2025 technology budget into AI, with forecasting tools deployed in wealth, commercial, and global banking. Its Glass platform supports client-facing teams with sector and trend predictions using supervised machine learning models. These models analyze asset class performance, market data, and behavioral indicators to support portfolio positioning and investment recommendations.

For treasury teams, BoA’s CashPro Forecasting tool uses AI to generate cash position estimates across geographies and currencies. It accounts for payment cycles, receivables, and external market factors. These forecasts feed into liquidity planning and FX exposure management for global clients.

Both tools are governed internally. Forecasts are validated through historical backtesting and ongoing performance checks. Glass includes narrative layers to explain model outputs. CashPro Forecasting aligns with operational risk controls and is embedded into the bank’s product governance structure.

Governance in Focus

All three banks share a common thread: they view forecasting models as institutional tools subject to model risk controls, regulatory alignment, and clear decision rights.

Each platform includes oversight processes for data provenance, output explainability, and escalation of anomalies.

None of these banks assume that accuracy alone is enough.

Governance, validation, and political defensibility are treated as part of the deployment plan. Human override mechanisms are embedded, and forecasting outputs are treated as advisory signals that must pass institutional filters before action is taken.

🥊 Your Move

1. Operationalize Forecasting
   If your forecasting models are still stuck in a sandbox, the window for advantage is closing. Integrate forecasting outputs into advisory, pricing, or capital workflows with clear oversight and ownership.

2. Audit the Governance Perimeter
   Document model inputs, outputs, and supervisory controls. If a model affects pricing, portfolio, or treasury posture, it must be formally tiered and validated.

3. Secure Forecasting as Strategic Infrastructure
   Treat forecasting models as a source of competitive leverage to continuously extract value. Govern them tightly. And ensure that board-level visibility exists before model-driven actions scale further.

⚔️ Vendor Spotlight:  Kensho AI

Kensho Technologies, an S&P Global subsidiary, provides AI-powered tools for market intelligence, forecasting, and risk modeling. U.S. banks use Kensho’s platform, including tools like NERD, Scribe, and Link, to extract insights from earnings calls, analyst reports, and macroeconomic narratives.

A Tier 1 bank used Kensho to process over 12,000 earnings calls per quarter, cutting review time by 80%. Another institution applied Kensho Link to model third-order exposures during geopolitical disruptions.

As a natural language processing (NLP) extractor, rather than a generative AI, Kensho is designed to retrieve and structure data from source materials, minimizing the risks of hallucinated content often associated with large language models.

In April 2024, S&P launched Kensho Benchmarks, enabling banks to evaluate LLM performance across financial tasks.

🥊 Your Move:

1. Evaluate Kensho Benchmarks to compare LLMs used in financial forecasting.

2. Prioritize vendors offering transparent NLP pipelines and integration with regulated datasets.

3. Review Kensho Link to model indirect exposure risk in global macro scenarios.

📈 Trends to Watch: What’s Emerging in AI Forecasting

The first generation of AI forecasting tools is now embedded in treasury, advisory, and macro strategy. But a new wave is forming—one that reshapes not just how banks forecast, but how they govern the machines doing the work.