Note: this site and the author does NOT have any financial relationship with the service provider being reviewed.

THREE POINT OVERVIEW:

INTRODUCTION:

Over the past week I tested out Shortcut AI (link), an AI agent for Excel. Overall I think it’s an ambitious tool that is off to a good start – as both of my write ups this week will show, while certain capabilities were impressive, some more work is needed to improve reliability. I am sure Shortcut will make these improvements - I plan on trying it again soon and posting updated reviews to keep readers up to date on progress.

This is the first of two write ups I will publish on Shortcut AI this week. Instead of focusing these reviews on trying a one shot three-statement DCF model that spreadsheet AI providers seem to like to put in their demos (I tried this and the result was a very high-level cookie cutter DCF model with some of the same reliability issues that we will touch on shortly), I focused this assessment on modeling work that a buyside analyst would actually need to do when trying to get up to speed on a new name.  If you have any experience in the industry, you know that a super high level three statement DCF model (even without accuracy issues) is of limited use and I can’t recall a single time when I’ve been asked to build one of those on the job.

This week’s first review focusses on the initial “table stakes” of financial modeling, while the second will focus on my attempt to build out parts of a granular model forecast. In general, I like Claude’s suggested approach to financial analysis with AI tools: 1.) Retrieve (the data), 2.) Analyze (the retrieved data), 3.) Create (outputs based on retrieval and analysis). I think it’s important to test AI tool capabilities in each part of this process - so this week’s first write up will focus on “Retrieve” while the second will focus on “Analyze” and “Create”.

I think “table stakes” of modeling include correctly laying out the capital structure and populating historical financial statements, segment financials, and any disclosed KPIs into the model. Of course, there are already existing (“non-AI” or “traditional”) providers that can be used for this - examples are Canalyst, Daloopa, BamSEC etc. – but it is still important to test AI tools’ abilities to pull accurate historical financials since these are marketed as one-stop shop solutions. Also, part of the value proposition of these AI tools is that they can be applied to private investing environments (where structured SEC data and public model libraries are unavailable), so evaluating their ability to correctly pull financial data from various provided materials in different formats is important.

So let’s dive in to how I tested out Shortcut - using various prompts, I asked the model to lay out the capital structure and import annual historical financials for an equity name I currently I have on my watchlist: Vail Resorts Inc. (ticker MTN) - just in time for ski season. First, I’ll go through my main takeaways from using Shortcut before showing you screenshots of the actual outputs. I tried ~10-15 reps of the same prompts so the takeaways below reflect multiple rounds of testing.

REVIEW OF RESULTS:

Below I show direct outputs from shortcut (aside from making formatting adjustments, I did not manually edit any data in the model outputs). Shortcut did a good job laying out the capital structure but ran into some accuracy issues in pulling historical financials.

Laying Out the Capital Structure:

Inputs: I wrote a prompt asking for capital structure to be laid out (full prompt in Appendix) + uploaded a PDF of the latest financials (FY 2025 10-K filed on September 29, 2025) and a header template (Figure 1) for a standard capital structure table that I would normally build out. I found that uploading a template for the capital structure was helpful – Shortcut was able to read the column headings without me needing to specify each set of data that I wanted for each tranche (ie. it recognized that I needed maturity dates and interest margins for variable debt, so I didn’t have to ask explicitly in the prompt).

Outputs – Capital Structure Table (Figure 2): Shortcut generally pulled correct data for each tranche of debt in the 10-K after a follow up prompt to address some formatting and layout issues (Figure 2). I was able produce decent first draft of the cap table. Shortcut pulled outstanding amounts, interest rates, and maturity dates correctly from the long-term debt section and related notes about each tranche. Only a few details were off - for example, revolver availability did not account for drawn letters of credit and Shortcut missed an undrawn $275M delayed draw term loan that is part of the Vail Holdings Credit Agreement. Also, certain nuances related to structural subordination, such as the HoldCo converts not being recognized as structurally subordinated to the rest of the debt stack for example, would need further prompting or manual updating - but this is not surprising given my prompt didn’t ask for structural subordination to be taken into account. Overall, aside from a few missing details I would say Shortcut did a decent job here.

Outputs – Diluted Shares Calculation (Figure 3): For the market cap calculation, I asked Shortcut to calculate fully diluted shares outstanding using the treasury stock method - the data used and structure of the calculations produced by Shortcut was almost all correct, except average shares during FY 2025 were used as the starting point instead of the end of period share count, so a follow up prompt was needed to correct this. But overall it was impressive that Shortcut was able to correctly take into account the outstanding converts, stock options, and RSUs (and pulled the right numbers for each of these) . I was pleasantly surprised here.  

Populating Historical Financials:

This is where we ran into persistent reliability issues.

Inputs: I wrote a prompt asking for four years of annual historical income statement and segment financial figures (full prompt in Appendix) + uploaded PDFs of the FY 2024 10-K and FY 2025 10-K

Outputs (Figure 4): I asked Shortcut to populate the financials for FY 2023-2025 first, and this is where the noticeable reliability issues started, with Shortcut mixing up numbers between years…

Not all pulled financials had these issues (for example, the EBITDA reconciliation pulled from the 10-K was correct for all years), but these issues of mixed up numbers were common in various parts of the pulled financials and segment KPIs in all 10-15 reps of pulling historical figures (each rep was us repeating the prompt to pull the financials, to make sure these errors were not one-offs). Below we show these errors in the segment financials (Figure 5) - note that these were not hallucinations, but numbers swapped between years…

When asked to trace where incorrect numbers were pulled from, Shortcut incorrectly cites 2024 figures as being 2025 numbers (Figure 6) …

KEY TAKEAWAYS:

I was pleasantly surprised by Shortcut’s ability to complete the cap structure table, but the simple errors in pulling historical financials and KPI’s were a glaring issue. For a modeling tool where accuracy is paramount, these simple mistakes impair trust in the tool and are a major offset to productivity gains.

Highlights:

1.) Shortcut did a surprisingly good job laying out the capital structure: I uploaded a cap table template and the model was able to pull correct outstanding debt balances, interest rates, and maturity dates for each debt tranche from the 10-K long-term debt schedule and notes describing each tranche. The treasury stock method calculation to calculate diluted shares outstanding was also mostly right, which was impressive.

2.) Shortcut successfully updated an additional year of historical results from another uploaded 10-K after model was built: Model was able to add an additional of historical results when I uploaded an additional 10-K without errors.

Shortcomings/Considerations:

1.) Accuracy issues: The mix up of numbers between historical years was an issue in almost every instance across our various reps of the same prompts (so not a one-time issue) – but it seems like this should be fixable. When I uploaded the same FY 2025 10-K to NotebookLM, it was able to pull the income statement and segment financials correctly without a problem - the number retrieval problems should be fixable for Shortcut (or perhaps there needs to be some improvement in the Anthropic models Shortcut uses)!

2.) Significant variability in formatting: several reps of the same prompts often produce different formatting despite the same instructions.  The formatting is not always consistent and the model does not always follow very clear specifications (for example, using font different from the Arial 10 that I specified in my constant preference settings, which also asked for universal modeling rules like blue font color for hard coded financials, formulas in black, links to other tabs in green, etc.)

APPENDIX:

a.) Prompts Used for Capital Structure Pull:

Initial Prompt: Complete the capital structure table on the Capital Structure tab for Vail Resorts Inc. (ticker is MTN) using only data from the FY 2025 10-K that I uploaded. Show exact figures without any rounding for all values. Show all dollar values in millions. Create sub-groups with secured debt (including finance leases) and unsecured debt, and within each sub group show earliest maturities at the top. To the right of the column which shows outstanding amounts, use the “x LTM EBITDA” column to calculate leverage through each part of the capital structure. To calculate the market capitalization, use fully diluted shares outstanding and a $135 share price.  Calculate fully diluted shares outstanding using the treasury stock method on another tab. Make sure to include any preferred equity and non-controlling interest in the equity portion of the table.  The enterprise value calculation should use debt net of unrestricted cash.

Follow Up Prompt #1: Group the EPR Secured Notes together into one line. Group the Employee Housing Notes together into one line. Only show leverage through totals in the capital structure. Clean up the formatting - make sure entire label is visible in the header and the total formatting lines go all the way across the table width. Below the total debt amount, show subtraction of cash and net debt, then the equity value, and then the enterprise value. I do not need the additional separate Enterprise Value calculation section that is currently included.

Follow Up Prompt #2: For the basic shares outstanding figure in the treasury stock method calculation, use end of period shares outstanding instead of weighed average shares outstanding during FY 2025.

b.) Prompts Used for Financials Pull:

Initial Prompt: Update the Income Statement tab. Use only exact values from the 10-K I uploaded without rounding. Show all dollar values in millions. Copy the historical income statement as shown in the uploaded 10-K. Below the income statement, in a separate section show the EBITDA reconciliation exactly as disclosed in the 10-K. Below the EBITDA reconciliation, in a separate section show segment financials exactly as disclosed in the 10-K for each segment. Show each segment separately and include any other metrics or KPIs disclosed for each segment. Make sure all totals on the Income Statement tab are formulas and not hard coded

Follow Up Prompt #2: Update the Income Statement tab with FY 2022 using only information from the FY 2024 10-K that I just uploaded.

c.) “Your AI Preferences” Used:

Note: these are the general guidelines that are uploaded for Shortcut to follow regardless of prompt:

Use size 10 Arial font for the entire model. Hard coded numbers in blue, formulas in black, numbers linked to other tabs in green. Show negative financial figures in a format that uses parentheses. Percentage numbers should be shown in a percentage format, and dates in a date number format. Show all historical figures with years increasing from left to right (with the latest year being in the right most column). Show the fiscal period end date under each fiscal year label. For all other formatting decisions, use professional formatting that would be used by an investment banking or private equity group.