AI in financial services and finance operations
Finance and accounting, financial services and insurance.
Studies
What we have measured in this sector.
No study specific to financial services and finance operations yet. The first one is in progress, and members hear first.
Tell us what you would want measuredAI vendors in this sector
106 vendors on our radar. The table groups them by area of work; the list shows a sample, tested vendors first.
Agentic AI platform for commercial due diligence and strategic advisory that sources and screens deals, sizes markets, drafts investment memos, CIMs and teasers, and analyses financials with traceable citations.Private marketsFinancial operations
AI workspace for consulting and finance firms that builds and audits Excel models, due-diligence databooks and valuations from project data, and turns them into cited decks through Excel, PowerPoint and Word add-ins.Capital marketsFinancial operations
Southeast Asian enterprise generative-AI provider offering agents for document processing, lending workflows and financial-statement analysis, plus a token-metered inference platform and custom LLM training and deployment.Credit and lendingFinancial operations
AI-native FP&A platform that connects finance and operating systems for budgeting, forecasting, scenario, headcount and revenue planning, financial reporting and multi-entity roll-ups, including plain-language model building.Planning and management accountingFinancial operations
AI-based core-systems platform for commercial and specialty insurance programs, covering underwriting, policy issuance, billing, claims, submissions intake and bordereaux processing.UnderwritingFinancial operations
AI actuarial platform for insurers covering P&C pricing and reserving and, via Akur8 Life, actuarial modeling for life insurers and pension firms, with explainable models and built-in governance.UnderwritingFinancial operations
AI-native research terminal for institutional equity investors that connects licensed market data, filings and firm research through a financial ontology, with agents drafting memos, models and earnings notes.Capital marketsFinancial operations
Self-clearing, FINRA-regulated broker offering developer APIs for stock, options and crypto trading, letting algorithmic traders trade programmatically and fintechs embed investing via its Broker API.Capital marketsFinancial operations
Binocs · AI for commercial due diligence · tested June 2026Does a specialist diligence agent beat general AI research tools?It finds far more sources than the general models, but cites fewer of its claims than GPT Deep Research.
source domains, against 64 to 96 for the general models
of claims carry a citation, against 75% for GPT Deep Research
repeated figures agree with each other, level with GPT Deep Research
The test. Six sector reports on satellite and Earth observation from each system: Binocs, Claude Opus 4.8 and GPT 5.5 (each with web access) and GPT Deep Research. We extracted 15,383 figures, checked 5,082 citations and scored 19 measures.
| Measure | Binocs | Claude Opus 4.8 | GPT 5.5 | GPT Deep Research |
|---|---|---|---|---|
| Source domains | 833 | 96 | 64 | 81 |
| Citations that still work | 84% | 70% | 77% | 79% |
| Claims with a citation | 56% | 48% | 52% | 75% |
| Firm point estimates | 82% | 34% | 59% | 72% |
| Companies covered | 1,062 | 131 | 202 | 344 |
| Forward-looking figures | 8% | 14% | 21% | 28% |
| Repeated figures that agree | 66.7% | 20.5% | 34.1% | 66.0% |
Held up
- Breadth: 7,722 sources across 833 domains, with the fewest behind paywalls.
- Specific numbers, with firm point estimates in 82% of claims.
Fell short
- 5,358 figures carry no citation.
- Little forward-looking content, and 284 contradictions across repeated figures.
PowerPoint output repeats figures across slides by design, which inflates repeats. One sector, one set of prompts. Request the full Binocs report
Tracelight · AI for financial models in Excel · tested July 2026Can an Excel AI add-in build a three-statement model you can trust?It catches errors and traces its sources well, but Claude's models balanced more often and handled a change in assumptions better.
of planted errors caught, level with Claude, against 67% for GPT
out of 100 for locating source cells, against 39.2 for GPT
models passed the change in revenue growth, against 4 of 4 for Claude
The test. Tracelight, Claude and GPT each built three-statement models from Apple and Nike 10-K data for fiscal 2023 and 2024. We checked history against the filings and 2024 forecasts against actuals, changed revenue growth by 5 points, and planted an error in cash.
| Measure | Tracelight | Claude | GPT |
|---|---|---|---|
| Workbook completion | 62.5 | 62.5 | 50.0 |
| Historical values correct | 75.8 | 77.7 | 78.1 |
| Source cells located | 75.0 | 73.1 | 39.2 |
| Model balances and closes | 87.5 | 100 | 68.8 |
| Revenue change flows through | 75.0 | 100 | 25.0 |
| Planted error caught | 100 | 100 | 66.7 |
| 2024 lines within 10% of actual | 64.6 | 66.7 | 45.8 |
Held up
- Every source cell it declared existed.
- Caught every planted error.
Fell short
- Lowest of the three on historical accuracy.
- Behind Claude on balancing the model and on the revenue test.
One run per system per company and year. Scores are out of 100. We do not combine them into one score. Request the full Tracelight report
Reports
Coming next: Finance roles tracker; Tax preparation: AI by task; Wealth management: AI by task.
Working on a deal or a deployment in financial services and finance operations?
Talk to us