Shortcut

Shortcut has the strongest independent evidence among Excel agents. It ranked first in Wall Street Prep's three-statement model test (5.9 against Claude 5.5 and Copilot 4.4) and in Ankura's CFO-team comparison, and it reaches every spreadsheet surface: Excel add-in, Google Sheets, browser, desktop agent with VBA, CLI and API. Its own Pivot model, post-trained for spreadsheets, lowers cost, and it ships several releases a week.

The same tests show the limits. Every tool, Shortcut included, scored below Wall Street Prep's low-tier analyst, Shortcut hallucinated historical data, and Ankura found no production-ready model without meaningful review. Shortcut carries no licensed financial content, SOC 2 Type II is claimed without a visible report, and it is one product inside a small multi-product lab competing with Microsoft and Anthropic, which also supply its models.

We initiate at 5.89, Emerging, with an Improving outlook. Modelling and delivery surfaces score among the highest in our coverage; Data & Content, at 3.17, reflects a tool that brings no content, and the framework's research orientation weighs on it. Buyers evaluating Excel agents should treat Shortcut as a leading candidate on modelling, alongside a content source.

Strengths

• Third-party results. First in Wall Street Prep's model test and Ankura's comparison against Claude for Excel and Microsoft Copilot.
• Surface coverage. Excel add-in for Windows and Mac, Google Sheets, a browser spreadsheet, a Windows desktop agent with VBA and Goal Seek, a CLI, an API and a coding-agent plugin.
• Modelling depth. DCF, LBO and three-statement models, sensitivities, banking-standard formatting and cell-level source notes.
• Own model. Pivot, post-trained for spreadsheet tasks and run at a fraction of frontier-model cost.
• Deployment options. Private model gateway, customer-managed keys, customer-owned storage and a local desktop agent; SAML SSO and SCIM.
• Shipping cadence. Several releases a week since launch in July 2025.

Cautions

• Output reliability. Third-party tests found hallucinated historical data and no production-ready model without review.
• No content. No licensed filings, transcripts, estimates or research; data comes from the user's documents and API keys.
• Security evidence. SOC 2 Type II is claimed, but the report period and auditor were not seen; no ISO 27001; information-barrier controls were not located.
• Training data. Free-tier usage trains its own model; paid and enterprise data are excluded by vendor statement, not located in contract terms.
• Vendor scale. A small team inside Fundamental Research Labs, whose stated mission extends beyond finance.
• Platform competition. Microsoft and Anthropic supply models and sell competing Excel agents.
• Evidence independence. Shortcut published the Ankura study and lists Wall Street Prep as a case study; no commercial terms are disclosed.
• Pricing changes. Prices roughly doubled between late 2025 and 2026.