Everyone is talking about AI taking over jobs. I wanted to cut through the noise and find out what was actually true — from the desk of someone who has been doing this work for 25 years.
So I ran a clean experiment. A real company, real financials, a real deliverable, and a timer. The question: could Claude + Excel take over the job of a financial analyst?
The Setup: Pick a Real Company, Not an Easy One
I deliberately avoided the obvious choices. Testing Claude on Apple or Microsoft is almost too easy — there's so much written about those companies that any AI would have rich data to draw from. I wanted a genuine test.
So I went back to my mining roots and picked Pan American Silver (NYSE/TSX: PAAS) — the world's largest primary silver producer, operating ten mines across seven countries, with complex multi-entity financials, operational metrics by mine site, and the kind of nuanced investment story that takes a skilled analyst days to properly understand.
I downloaded their 2024 Annual Report directly from their website. 180+ pages. Full financial statements, mine-by-mine operational data, management discussion, and risk factors. Then I started the clock.
Prompt 1: Extract and Structure the Financials
Five minutes later, Claude had built exactly what I asked for. A properly structured multi-sheet workbook — income statement, balance sheet, and cash flow each on their own tab with formulas calculating margins, ratios, and year-over-year changes. A mine-by-mine operational summary with AISC, production volumes, and country breakdowns. Formula architecture throughout, so every derived number traces back to a source cell.
"Anywhere there can be a formula should be a formula" — that's the instruction I'd give a junior analyst on their first day. Claude followed it precisely. Every margin, every YoY change, every ratio: a live formula, not a hardcoded number.
Prompt 2: Build the Board-Ready Executive Summary
This is where it got genuinely interesting. Because I told Claude what I wanted to use this for, it didn't just organize the data it already had. It pulled Pan American Silver's February 2026 earnings release — released just two days prior — incorporated FY2025 actuals alongside the FY2024 model, and built a much fuller picture of the company's current position.
But what impressed me most was the bull and bear case. I asked for no BS — and it delivered both sides without pulling punches.
Record $3.6B revenue, $980M earnings, $1.15B FCF — all in one year. 14% silver production growth guided for 2026. Fortress balance sheet with $2.1B liquidity. La Colorada Skarn represents generational upside. Juanicipio acquisition already exceeding expectations.
"Record results are largely driven by $40+ silver and $3,400+ gold. At FY2024 price levels, earnings collapse 85%+. This is a commodity bet first, a company bet second. Stock up 126% in one year — much of the good news is already priced in."
Every number had a verifiable source reference. Every external data point was linked. The bottom-line recommendation read like a CFO wrote it — conditional, honest, and specific about what macro view had to be correct for the thesis to hold.
Prompt 3: Build the Board Charts
What came back was a dedicated charts sheet with eight fully linked visualizations — revenue and earnings trajectory, cash flow generation, balance sheet strength showing the shift from net debt to net cash, silver production trajectory with 2026 guidance, realized prices vs. AISC showing margin expansion, mine portfolio production by asset, EPS and dividend growth, and a silver price sensitivity table.
Every chart cell linked back to the source data sheets. Pull up the sensitivity table showing estimated revenue at $20, $28, $41, $50, and $58 per ounce silver — every number is a formula referencing the production data. Another analyst could pick up this file, trace every number, and verify every assumption. That's what happens when you tell Claude why you need it.
Tasks might change. But the job will be transformed.
What Claude Got Right — and What Still Needed the CFO
Structure, comprehensiveness, speed. The discipline to include a bear case unprompted. Proactively pulling live earnings data two days after release. Formula-linked charts. A bottom-line recommendation written in plain language, not consultant-speak.
Verifying key numbers against the source PDF. The Escobal context — suspended since 2017, ILO 169 consultation with no end date — needs years of industry knowledge to properly frame. And the conditional recommendation needs a CFO to walk the board through it, not just leave it on a slide.
So — Can Claude + Excel Replace the Analyst?
No. But the job just changed.
The grunt work — hours of researching, extracting, formatting, and building a first-pass model from a 180-page PDF — is now 15 minutes. That time gets redirected. Instead of spending the day building the model, the analyst spends the day on validation, context, and the kind of judgment that actually changes a board's decision.
The analyst who learns to work this way isn't being replaced. They're being elevated. The floor of what they produce in a day just got raised by about 70%. What they do with the other 30% is where the real value lives.
Interactive animation showing the before and after of using Claude plus Excel to analyse Pan American Silver's annual report in 15 minutes
