Inside Marginal AI.
Agentic intelligence and alternative data for economics and US-listed securities. How the extraordinary-event layer works under the hood, what it covers, who it's for, and what it costs.
The core
A proprietary ontology built on seven categories.
At the core, Marginal AI is built on a proprietary ontology of ~500 event types mapped to seven extraordinary-event categories.
Paired with multi-source ingestion from primary, credential-gated sources; multi-layer anti-hallucination safeguards; and citation-first provenance on every claim.
The result: fast, seamless access to event-driven intelligence in one platform, no coding, no manual workflows. Institutional-grade extraordinary-event coverage at a fraction of the cost, with agentic delivery that turns hours of analyst triage into seconds.
Short platform walkthrough.
Why events matter
They show up in every model, memo, and trade.
Whether you operate in private or public markets, tracking extraordinary events is essential for:
- Company Comparables & Benchmarking. Understanding how peer firms have reacted to non-operating events.
- Investor Memorandums, Due Diligence & Disclosures. Identifying and disclosing material risks and exposures, and stress-testing scenarios.
- Historical Evaluations. Assessing longer-term impact and conducting retrospective analysis.
- Risk Assessments. Pricing in external shocks across portfolios, counterparties, and concentrations.
- Investment & Budgeting Decisions. Evaluating event-driven market movements and incorporating them into capital allocation and planning.
- Origination & Research. Identifying trends and opportunities before they gain widespread attention, and complementing existing research workflows.
What this looks like in the data. Three decades of US systemically important universal banks (assets ≥ $100B), and the mark extraordinary events leave on the income statement.
Major expense categories and extraordinary events, in millions of US dollars, 1992-2024.
Extraordinary-event expenses by sub-category, in millions of US dollars, 1992-2024.
Extraordinary items and loan losses as a percentage of total revenues and net income, 1992-2024.
Composition of total expenses by category, 1992-2024.
Categories
Examples of extraordinary events.
Category 1
Accounting adjustments & other non-recurring events
- Restatements & material misstatements. Revisions to previously issued financial statements due to errors, fraud, or accounting standards violations.
- Impairments & write-downs. Goodwill impairments, asset write-downs, or inventory write-offs that materially affect reported earnings.
- One-time charges & gains. Severance and restructuring charges, divestiture gains, or insurance recoveries.
- Going concern & auditor issues. Going-concern doubts, auditor resignations, or qualified audit opinions.
- Changes in accounting methods. Shifts in revenue recognition, lease accounting, or other policies that alter reported financials.
Category 2
Force majeure & environmental events
- Pandemics & health crises. Outbreaks driving operational shutdowns, regulatory restrictions, or workforce and product shortages.
- Natural disasters. Hurricanes, earthquakes, wildfires, or extreme-weather events causing asset damage and operational disruption.
- Geopolitical conflicts & trade sanctions. Wars, embargoes, economic sanctions, or political instability affecting business continuity.
- Climate & environmental liabilities. Major spills, pollution events, or climate-related regulatory penalties.
Category 3
Litigation & legal disputes
- Class action & shareholder lawsuits. Securities-fraud claims, labor disputes, or product-liability suits.
- Contractual & commercial disputes. Court rulings or settlements on contract breaches, supplier conflicts, or partnership disputes.
- White-collar crime. Fraud, insider trading, money laundering, or corporate-misconduct cases leading to fines or criminal penalties.
- Intellectual property & patent disputes. Trade-secret leaks, copyright violations, or patent-infringement cases.
Category 4
Market & economic shocks
- Monetary policy surprises. Unexpected central-bank decisions on rates, quantitative easing or tightening, or forward guidance.
- Currency & sovereign debt crises. Sharp FX dislocations, devaluations, or sovereign default events.
- Commodity price shocks. Abrupt and sharp moves in oil, gas, metals, or agricultural commodities that hit input costs or revenue.
- Inflation & recession signals. Surprise inflation prints, yield-curve inversions, or recession declarations.
- Flash crashes & liquidity events. Sudden market dislocations, circuit breakers, or stress on funding markets.
Category 5
M&A and corporate restructuring
- Mergers & acquisitions. Takeovers, joint ventures, or consolidations that alter competitive dynamics.
- Corporate restructuring & reorganizations. Asset sales, spin-offs, divestitures, and business-unit consolidations.
- Bankruptcy & liquidation. Chapter 7 and Chapter 11 filings, receivership, or forced asset liquidation under financial distress.
- Activist campaigns & take-privates. Shareholder activist demands, board contests, LBOs, or going-private transactions.
Category 6
Operational & technological disruptions
- Cyber attacks & data breaches. Ransomware, hacking incidents, or unauthorized data access.
- Supply chain disruptions. Shortages, bottlenecks, or transportation failures affecting production or delivery.
- Product recalls & safety issues. Faulty or dangerous products pulled from market under regulatory or voluntary recall.
- Critical infrastructure & outages. Plant fires, refinery shutdowns, data-center outages, or network failures with material operational impact.
Category 7
Regulatory & legislative actions
- New legislation & draft laws. Tax laws, consumer-protection regulations, or financial-industry reforms that require businesses to adjust operations.
- Regulatory enforcement & penalties. Cease-and-desist orders, civil money penalties, prohibition orders, and enforcement actions tied to compliance failures, consumer-protection violations, or financial-reporting breaches.
- Antitrust & market-manipulation cases. Investigations into monopolistic behavior, collusion, price-fixing, or market manipulation.
- Government subsidies, bailouts & nationalization. Corporate bailouts, loan restructurings, explicit or implicit government guarantees, or outright nationalization of assets.
- Tax disputes & settlements. Tax-evasion cases, unexpected tax settlements, or changes in corporate tax rates impacting profitability.
Alongside extraordinary events, Marginal AI covers earnings and guidance in depth: quarterly and annual results down to the line item, performance broken out by business segment and/or geography, management guidance, and the analyst Q&A from earnings calls.
Vs incumbents
Why traditional solutions fall short.
No incumbent today combines full-suite extraordinary-event coverage with agentic reports, provenance-first guardrails, and multi-channel delivery at a sustainable seat price. The gap shows up in four ways.
Coverage is incomplete
Premium data platforms (Bloomberg, S&P Capital IQ, LSEG, FactSet, LexisNexis, AlphaSense) concentrate on M&A, executive changes, and litigation that reaches the newswires. The broader event surface (accounting restatements, regulatory enforcement actions, force-majeure events, supply-chain disruptions, sovereign and commodity shocks) sits scattered across regulator portals, court dockets, and specialized feeds with no common schema.
Pricing is prohibitive
Seat licenses run from ~$12,000 (FactSet) to ~$32,000 (Bloomberg Terminal), with API tiers reaching $250k+. You're paying for a full data terminal where extraordinary-event coverage is a small slice of what's offered, not the focus.
Delivery is manual and single-channel
Alerts route to a terminal or platform inbox. No SMS, no WhatsApp, no event-triggered agentic reports. Analysts still burn 1-2 full days triaging a single extraordinary event, pulling sources, reconciling them, and writing the narrative.
General LLMs aren't a fit
ChatGPT, Gemini, Copilot, and Claude lack a domain ontology, hallucinate on financial specifics, mangle charts and tables, and operate without consistent provenance or long-term memory. They're general-purpose tools; extraordinary-event intelligence isn't a general-purpose problem.
What Marginal AI gives you instead
- One platform, full coverage. A 500-type, 7-category proprietary ontology of extraordinary events, alongside earnings and guidance, unifying what incumbents leave fragmented, with deep alternative-data coverage across regulatory, litigation, macro, disaster, and market data sources.
- Seconds instead of days. Agentic reports turn 1-2 analyst-days of triage into a sourced, cited, auditable output, delivered the moment an event hits.
- Multi-channel delivery on your cadence. Push, email, SMS, and WhatsApp. Eight+ cadences from monthly down to near real-time. Significance scoring, deduplication, and frequency caps eliminate noise.
- Provenance-first by design. Every claim links to its primary source. Ontology-grounded retrieval anchored to the 500-type taxonomy, not generic semantic search.
- A fraction of incumbent pricing. $17-$250 per user per month, with agentic reports priced by company/contract count, frequency, and depth ($8-$875 per week per agentic subscription).
Marginal AI vs Claude Opus 4.7 (Adaptive Thinking)
Pricing
Marginal AI pricing.
Marginal AI is priced in Compute Units (CU). A CU measures the computation a request consumes, so longer and more complex requests use more CUs than short ones. Your plan includes a set number of CUs each period, and usage beyond that is billed at $0.08 per CU, up to a monthly overage limit. CU are metered per account for individuals, pooled across the organization for Business and API customers, and reconciled at the end of each billing cycle.
Pricing covers two independent product lines, available separately or together: platform plans (chat and core access) and Agentic Report subscriptions (automated monitoring and alerts).
Free Tier
Start with 75 CU per month to explore the platform, refreshed every 30 days from your registration date. Chart cards and financial metrics are included for a limited number of uses, which do not refresh.
Self-service · Individuals
Monthly platform plans for individual users. Annual billing: 10% off, with your CU allowance renewing every 30 days.
| Plan | Monthly | Annual (10% off) | Included CU / Month | Overage |
|---|---|---|---|---|
| Basic | $17 | $183.60 / year | 240 | $0.08 / CU |
| Intermediate | $25 | $270 / year | 355 | $0.08 / CU |
| Advanced | $60 | $648 / year | 850 | $0.08 / CU |
Self-service · Business
Per-seat platform plans, with CU pooled across the company. 5% off the per-seat price from 20 seats. Annual commitment: 10% off.
| Seats | Per seat / Month | Per seat / Year (10% off) | Included CU / Seat / Month | Overage |
|---|---|---|---|---|
| 1 - 19 seats | $250 | $2,700 | 3,400 | $0.08 / CU |
| 20+ seats | $237.50 | $2,565 | 3,400 | $0.08 / CU |
App store plans have no overage. If you subscribe through an app store, you top up with Compute Unit packs in the app instead.
UK VAT. Customers whose billing country is the UK pay 20% UK VAT on top of the plan prices shown (for example, Basic is $20.40 per month).
Enterprise API
Minimum $2,500 per month, billed annually. Includes 10 seats and 34,000 CU per month pooled across your organization. Additional seats $250 per month each. Overage $0.08 per CU.
| Onboarding | One-off fee | Best for |
|---|---|---|
| Quick Start | $1,500 | Teams launching on the API with standard volume needs |
| Custom Requirements | From $10,000 | Custom schema, agentic workflows, integrations, or resource augmentation |
Default limits. Up to 600 API calls a minute and 200,000 a day per company. Higher limits are available on request.
Agentic Reports & Multi-Channel Alerts
Configure agentic subscriptions that monitor up to 10 companies and 3 custom events on the cadence you choose, from one-time to every minute. Each subscription includes the initial report plus updates whenever a genuine material change is detected. Available standalone or alongside any platform plan.
Two depths
- Summary. Concise event briefs with citations. From $8 / week (one-time) to $160 / week (near real-time).
- Deep Research. Full agentic reports with deeper reasoning, longer context, and richer source coverage. From $40 / week (one-time) to $800 / week (near real-time).
Both depths include email and push notifications. Add SMS or WhatsApp alerts to any subscription for a small per-cadence uplift.
Representative pricing (per week, reports only)
| Cadence | Summary | Deep |
|---|---|---|
| One-time report | $8 | $40 |
| Once a week | $20 | $100 |
| Once daily | $28 | $140 |
| Every hour | $36 | $180 |
| Every 30 minutes | $38 | $190 |
Full cadence ladder and SMS / WhatsApp pricing visible in the agentic subscription page of the application.
Scope scales linearly. Tracking more than 10 companies or more than 1 industry group multiplies the price proportionally: 11-20 companies or 2 groups = 2× listed price; 21-30 or 3 groups = 3×; and so on.
Cancel anytime, effective at the end of the current billing cycle.
Ready to track what moves your markets?
Start free with 75 Compute Units a month, refreshed every 30 days.
Get started
