Finance AI Index
Archive

Editions and raw data

Every prompt, every full response with timestamp, model version and search flag, every raw judge label, and the versioned vendor table. Editions are archived permanently at stable URLs.

Edition log

Each edition with the collection runs that produced it and the calibration repeat that set its noise floor. Monthly editions run on the first of each month at 06:00 UTC. Special editions follow major frontier model releases and are published as a comparison against the prior generation; they are added to the log when a release is announced.
EditionKindDateLabelsVendor tablethe change log is in the fileStatus
September 2026 EditionEdition2026-09-1746,778vv2026-09.2Published
Collection run, smoke, small business, mid-market and enterpriseRun2026-09-175,858vv2026-09.2Staged
Collection run, baseline, small business, mid-market and enterpriseRun2026-09-1740,920vv2026-09.2Staged
October 2026 EditionThen monthly on the first, through the October 2027 Edition. The full schedule →Edition2026-10-01pendingpendingScheduled
PublishedIn the public record at a stable URL, citable.MeasuredA calibration repeat, run and measured, feeding the noise floor.StagedThe data exists and is not public yet.ScheduledThe data does not exist yet.

The record

September 2026 Edition · vendor table vv2026-09.2. Every prompt, every full response with timestamp, model version and search flag, every raw judge label, and the versioned vendor table. The free files are the raw record as JSON. The pack is the same record resolved into the tables the site itself is built from, as CSV and one SQLite file. Nothing is withheld, because every figure in the pack can be recomputed from the free record.

Every file of this edition is free now.

ArtifactOne row of it isSizeWhere
The raw recordJSON, one row per label or answer. Free.
edition-2026-09-scores.jsonJSONOne row per named product, with the name exactly as written and a verbatim evidence quote.16.6 MBFree now
edition-2026-09-responses.jsonJSONEvery call with its full answer text, timestamp, version string, search flag, cited URLs, latency and token counts.53.0 MBFree now
README.mdMarkdownWhat each column of the record means, how to reproduce any share on the site, the license and the citation. The same file ships in the pack.4 KBFree now
edition-2026-09.mdMarkdownThe derived report for the edition, including the appendix on loose paraphrase stability.378 KBFree now
prompts.jsonJSONEvery prompt verbatim: six variants across 41 categories for each buyer segment, 246 per segment. Each prompt carries its set; the September 2026 mid-market set, asked as a B2B software company, is kept under its own ids beside the current one.347 KBFree now
models.jsonJSONPinned models, endpoints, search tools and list prices as of the edition.12 KBFree now
vendor_aliases.jsonJSONCanonical names, aliases, category-scoped readings, exclusions and the change log.523 KBFree now
The record, resolvedCSV and one SQLite file: the tables the site itself is built from. In the pack.
standing.csvCSVEvery category and product pair for every buyer segment: share, rank, negative rate, labels, quadrant.9,313 rowsPack only
labels.csvCSVOne judge label: raw name, canonical vendor, how it resolved, label, weight, evidence. The free labels file, resolved.46,965 rowsPack only
vendors.csv, vendor_by_model.csvCSVOne product: its totals, and how each model treated it.3,918 productsPack only
vendors/<product>.csvCSVEvery label for one product, for a vendor that wants only its own rows.3,918 filesPack only
responses.csv, prompts.csv, models.csvCSVOne answer with its version string, timestamp, search flag, cited URLs and tokens; every prompt; the model set.8,856 answersPack only
index.sqliteSQLiteEverything above in one SQLite file with a leaders view; one query reproduces any number on the site.all tablesPack only

Everything on this page is published under CC BY 4.0, free to cite. The data dictionary, what each column means and how to reproduce any share on the site, is published above as well as in the pack.

Cite this edition
Finance AI Recommendation Index, September 2026 Edition. finance-ai-index.com. Scored under vendor table vv2026-09.2.
September 2026 Edition pack36.6 MB

One zip: every row above, resolved, with the data dictionary. The same key opens a report for every category: the standing, the warnings, the sources and every answer, with a written summary.

$9 per edition, or $9.95 a month for every edition
Get the pack

Already bought it? Download it with the license key on your receipt, on the setup page that also connects an assistant. A subscription key opens the newest edition.

On request

The public index is a fixed scope, rescored every edition: 41 categories, three buyer segments, six framings, twelve models. Everything below runs on one category for that category's subscribers, as an expanded run, and never enters the public standing.
CutWhat it is
The flagship modelsClaude Opus 5, Claude Opus 4.8, GPT-6 Astra, GPT-5.6 Sol, Gemini 3.1 Pro, Perplexity Sonar Pro, on your category every month.
Job-to-be-done framingA seventh prompt per category that describes the work instead of naming the category.

Prices are on the Subscribe page. A cut that enough subscribers ask for can be promoted into the public index; that is announced in the edition log above and on the method page's change log, never done quietly.

Category roadmap

The addressable set across seven finance functions. The exceptions come first, because they are what the section is for.
41Categories coveredScored every edition, full record published.
8Out of scopeNamed below, with the reason.
7Finance functionsThe addressable set.
13Planned, not yet scoredAdditions are announced in the edition log.
Out of scopeNamed, with the reason. These are the rows the section exists for.
The adjacent functions domainAll 8 of its categories: Billing and subscription management, CPQ, Contract lifecycle management, Compliance automation, Third-party risk management, Payroll, Workforce planning, Cloud cost management. Products another index publishes or that a function other than finance evaluates. The buyer in the prompt decides which index a category is on.Out of scope
Products named as something other than a candidateA warehouse named as a data source inside a CDP answer, an ad platform named as a destination inside an attribution answer, a buyer's own tooling: counted only where the answer treats the product as a candidate in the category asked about, under the scoring rule on the method page.Excluded
Covered, by functionEvery covered category is scored the same way; open a function for its categories with the landscape reference and analyst coverage.
Accounting and closeThe ledger and the close: accounting software and ERP, close management, reconciliation, consolidation, revenue, leases, fixed assets and the AI assistants arriving in all of them.9Open 9 →
CategoryReferenceAnalyst coverage
Audit analyticsPlanned
Intercompany accountingPlanned
Planning and analysisForward-looking finance: FP&A, modeling and scenarios, reporting and dashboards, board reporting, corporate performance management and SaaS metrics.6Open 6 →
CategoryReferenceAnalyst coverage
Headcount and workforce budgetingPlanned
Spend and procurementMoney going out: expenses, cards, spend management, payables, procurement, sourcing, suppliers, travel and contractor payments.9Open 9 →
CategoryReferenceAnalyst coverage
Procurement analyticsPlanned
Purchasing cards and virtual cardsPlanned
Receivables and billingMoney coming in, after the invoice exists: receivables automation, invoicing, cash application, credit and B2B payment acceptance.5Open 5 →
CategoryReferenceAnalyst coverage
Dunning and payment recoveryPlanned
Treasury and cashCash and banking: forecasting, treasury management, business banking, cross-border payments and bill pay.5Open 5 →
CategoryReferenceAnalyst coverage
Bank connectivity and aggregationPlanned
Working capital financingPlanned
Tax and complianceObligations: sales tax, tax provision, e-invoicing mandates and the controls behind financial reporting.4Open 4 →
CategoryReferenceAnalyst coverage
Tax research and AIPlanned
Transfer pricingPlanned
Equity and corporateThe company itself: the cap table, its entities and its sustainability reporting.3Open 3 →
CategoryReferenceAnalyst coverage
409A valuationsPlanned
Business insurancePlanned
Investor relationsPlanned

A category enters the addressable set when it is a product a buyer asks an AI to recommend for going to market. Additions and removals are announced in the edition log and on the method page, never done quietly.

Who is asking

Every prompt set is one combination of four choices: how big the buyer is, what industry it is in, how the question is framed, and who answers. Each added value multiplies the prompt set, so these are taxonomy decisions first and scope decisions second.

Company size
Small business Prompts reframed for a small team and budget. Prompt ids cNN-s1 to cNN-s6.
Covered
Mid-market The scale prompt says 500 people. Prompt ids cNN-1 to cNN-6.
Covered
Enterprise Thousands of users, predictable total cost, SSO, SOC 2 and procurement review. Prompt ids cNN-e1 to cNN-e6.
Covered
Industry
B2B company Every prompt today says a B2B company.
Covered
Financial services Regulated, compliance-heavy stack.
Planned
Healthcare Privacy and consent constraints change the answers.
Planned
Public sector Procurement rules and data residency change the answers.
Planned
Manufacturing Operational technology beside the corporate stack.
Planned
Question framing
Category question The buyer names the category: "what is the best SIEM for…". The six variants on the methodology page.
Covered
Job to be done The buyer describes the work, not the category. The model names the category as well as the product.
Planned
Who answers
AI models Twelve models on the standard tier, fresh session per prompt.
Covered
Human buyers The same questions put to human buyers, published beside the model answers.
Planned
Cadence

A monthly edition on the same date. A special edition within days of a major frontier release, published against the prior generation. Quarterly written analysis, only once enough editions have accumulated to say something with substance.

Rescoring policy

When the vendor table changes, every edition in the series is regenerated under the new version, so the series is never a mix of table versions. A trend that appears only after a table change is tooling, not model behavior.

If the project ends

The historical series is the part that gets harder to replicate the longer it runs. Every edition stays at its stable URL, and the final edition is marked final on the site.