Updated 2026-09-07
Finance AI data: every figure we publish, dated and sourced
Most claims about AI in enterprise finance are unfalsifiable: a percentage with no denominator, a saving with no formula, a benchmark with no dataset. This page is the opposite. Every number Octopus AI publishes is listed here with its basis, its arithmetic where it is derived, and a link to the page it comes from.
Quote any of it with attribution. Where we have not measured something, the relevant page says so.
The figures
| ID | Metric | Value | Basis and formula | Source |
|---|---|---|---|---|
| F1 | Forecast cycle time, before and after | About 60 hours reduced to under 6 hours (about 90% faster) | Measured | Financial forecasting AI case study |
| F2 | Analyst hours returned over the measured period | Approximately 430 hours | Measured | Financial forecasting AI case study |
| F3 | Autonomous tasks completed in month one | 2,500+ | Measured | Financial forecasting AI case study |
| F4 | Output error rate on executed tasks | Below 0.1% | Measured | Financial forecasting AI case study |
| F5 | Reference client contract value | $523,000 over three years | Measured | Financial forecasting AI case study |
| F12 | Hours saved per forecast cycle | About 54 hours | Derived — 60 hours − under 6 hours = more than 54 hours | High-throughput financial infrastructure |
| F13 | Throughput per calendar day in month one | About 83 completed tasks per day | Derived — 2,500 tasks ÷ 30 days = 83.3 | High-throughput financial infrastructure |
| F14 | Maximum incorrect outputs implied by the error rate at month-one volume | Fewer than 3 of 2,500 tasks | Derived — 2,500 × 0.001 = 2.5 | High-throughput financial infrastructure |
Measured figures come from one anonymised enterprise finance deployment. Derived figures show the arithmetic that produces them.
Where each dataset lives
High-throughput financial infrastructure
Measured task volume, error rate and cycle-time reduction, plus the architectural properties that hold accuracy at volume.
Pre-registered benchmark
Protocol, synthetic dataset, deterministic generator and empty result tables, published before the run.
Forecasting case study
The deployment behind the measured figures: forecast cycle from about 60 hours to under 6.
Month-end close case study
Accruals and close workflows under the same governed architecture.
Transparency report
How the system behaved in the reporting period, what it is allowed to do, and how outputs are verified.
How we label figures
- Measured — observed in a live deployment and described in a linked case study. Client identity is anonymised at the client's request.
- Derived — arithmetic on published figures. The formula is shown so anyone can reproduce it.
- Benchmark data is synthetic with deliberately planted items. It demonstrates architectural behaviour under known conditions and is not a claim about performance on any specific customer's books.
Questions and answers
What data does Octopus AI publish?
Measured throughput and error rate from an anonymised enterprise deployment, cycle-time reduction on the forecast, and a pre-registered benchmark protocol with a public synthetic dataset and generator. Each figure on this page states whether it is measured, derived, or a list price, and derived figures carry their formula.
Can these figures be quoted?
Yes. Reuse is permitted with attribution to Octopus AI and a link to the page the figure comes from. Each page carries a copyable citation and the date the figures were last reviewed.
What is a Financial Action Execution?
One finished unit of finance work — a variance explanation, a forecast refresh, a reconciliation, an accrual proposal, a stakeholder-ready summary. It is the unit Octopus AI bills on and the unit throughput is counted in, which is what makes cost and volume directly comparable.
How reliable are the client outcome figures?
They come from a single anonymised enterprise finance deployment and are described in full in the case studies. They demonstrate what the architecture achieved in one environment; they are not a projection for any other finance function.
Is there an independent benchmark?
A benchmark is pre-registered rather than merely claimed: the protocol, the synthetic dataset, the generator that reproduces it byte for byte, and empty result tables were all published before the run, so results cannot be selected afterwards. Both arms of that benchmark run the same model, because the variable being tested is the architecture, not the model.
Cite this page
Figures on this page were last reviewed on 2026-09-07. Reuse is permitted with attribution to Octopus AI and a link to this page.
Octopus AI. "Finance AI data: every published Octopus AI figure." Octopus AI, updated 2026-09-07. https://myoctopus.ai/data