Model a financial strategy

Step 1 — Load the data AND the situation

Numbers without a situation produce generic analysis. Upload the CSV and paste this with it (fictional, like the data):

THE SITUATION: Meridian Coffee Roasters (fictional) has three lines of business — wholesale to restaurants, a retail café, and online bean sales. The attached CSV is 24 months of actuals. The owner has $60,000 to deploy next year on ONE of three moves: (a) expand the café, (b) push online harder, or (c) hire a wholesale sales rep. Help me get to a defensible recommendation.

The decision is context the window needs: analysis without a decision attached is a book report; analysis pointed at a choice is strategy.

Step 2 — Interrogate before you ask for answers

Before any conclusions: profile this data. For each revenue line — trend over the 24 months, seasonality, and gross margin as best the cost columns allow. For each cost — trend, and whether it scales with revenue or steps on its own. Flag anomalies: one-off spikes or level steps, separated from trends. Show me the three most decision-relevant patterns as small tables. No recommendations yet.

“No recommendations yet” is load-bearing: make the model show its reading of the data before it starts persuading. Then check it — pick any two numbers it cites and confirm them against the raw CSV yourself. And a tell to watch for: somewhere in this data is a one-off cost event. If the profile treated it as a trend, you’ve just seen why this step exists.

Step 3 — Test scenarios, don’t collect opinions

Model the three options over the next 12 months, using ONLY patterns present in this data. For each: state every assumption as a number I can change (growth rates, margins, ramp times, added costs), then show a conservative and an expected outcome for revenue and profit. Where the data cannot support an estimate — a ramp time, a hire’s pipeline — write UNKNOWN and tell me what real-world fact I’d need. Do not invent one.

Assumptions-as-changeable-numbers is the whole spreadsheet mindset in one rule. An analysis whose assumptions are prose can only be argued with; one whose assumptions are numbers can be tested.

Step 4 — Attack the recommendation

Ask for the recommendation, then open a fresh window and paste the profile, the scenarios, and the pick:

You are a skeptical CFO. Attack this: which assumptions are doing the most work? What in the data contradicts the recommendation? How small a change to which number flips the answer? What did the analysis ignore?

Back in the original window, run the flip test:

What would have to be true for [the runner-up option] to be the right answer? Are any of those things checkably true today?

A recommendation you haven’t tried to kill is an average wearing a conclusion’s clothes. The CFO pass and the flip test are how you find out whether it survives contact.

Step 5 — The one-page decision memo

Write the memo: THE DECISION being made. THE RECOMMENDATION, two sentences. THE CASE — the three numbers from the data that carry it. THE ASSUMPTIONS — each one a number, each one changeable. WHAT WOULD CHANGE THE ANSWER. Top line: “Based on data through [last month], checked [today’s date].”

Put it where the decision gets made, next to the dataset. And the habit that makes it strategy instead of a one-off: next quarter, fresh data in, same five steps — the memo is living, the assumptions are numbers, and you’ll know within minutes which ones reality has already voted on.

---

Check your understanding: why does the profile come before the recommendation — and why must every assumption be a number you can change?

Use this process for other financial decisions. Data + decision → profile → scenarios with changeable assumptions → CFO attack → dated memo. That’s the budget allocation, the pricing change, the hire/don’t-hire, the lease renewal — every choice where the numbers exist but nobody’s made them argue.