Skip to content
AI Finance & Analytics for Founders (Capstone)Lesson 5 of 10

5 · Building a simple budget and projection (and why AI projections can be wrong)

A budget is your plan for money in and money out over a period — usually a month. A projection (forecast) is your best guess at what future months will look like. Together they answer: "If I keep going, will I run out of money — and when?" Every real business plan has them, and the SBA expects financials in a business plan (SBA, n.d.).

Build a simple monthly budget (you can do this on paper or a spreadsheet):

  1. List money IN — expected revenue (be conservative; guess low).
  2. List money OUT — every cost: supplies, tools/subscriptions, shipping, ads, fees.
  3. Subtract: In − Out = your net for the month. Negative? You're burning cash — for how many months can you?
  4. Carry the balance forward so you can see cash flow over time, not just one month.

Make a projection — then distrust it appropriately:

A projection is a story about the future, and the future doesn't read your spreadsheet. Sales may come in lower and costs higher than you hoped. That's why founders build a conservative case (and often a "what if it's worse?" case too).

Why an AI's projection can be confidently wrong — and dangerously so:

  • It can do the arithmetic wrong. A model can produce a clean, professional-looking forecast with a math error baked in. Re-check every formula in the actual spreadsheet.
  • It invents optimistic assumptions. Ask AI to "project my first year" and it may assume a growth rate or a conversion rate out of thin air. Those guesses, not the math, drive the whole forecast — and they're often rosy.
  • It doesn't know YOUR reality. It has never met your customers or seen your real costs. A forecast built on made-up inputs is fiction with decimal points.

The discipline: let AI draft the structure of a budget or projection (the rows, the formulas, the layout). Then replace every assumption with your real, verified numbers, do the math in the spreadsheet, and label your guesses as guesses. NIST's whole point is keeping human judgment around the model, not handing it the wheel (NIST, 2023).

Trust DNA: an AI projection is a first draft of a guess. Its confidence is not evidence. Your verified numbers and honest, conservative assumptions are what make it useful.

Check yourself. Name two different reasons an AI-generated projection can be wrong — one about the math, one about the assumptions.

Sources

Sign in to track your progress