3 Model Errors That Will Make You Die a Little Inside

01. The assumption you can't actually explain

You're projecting revenue for the next five years. You build a churn model to estimate how many customers you'll lose each year. The churn rate you use is 5%.

Where did that come from?

You can't remember. Maybe you read it somewhere. Maybe you took it from a competitor's quarterly call. Maybe you just thought it sounded reasonable. The spreadsheet works fine. The math is clean. The output looks credible.

But you have no idea why you chose 5%.

This is the error that kills credibility the fastest. Not because the number is wrong, but because you can't defend it. In the room, when someone challenges your assumption and you can't explain it, you've just signaled that you didn't build this model with care.

Every number in your spreadsheet needs a source. A comparable company. Historical data. An expert opinion with a name attached. A published study. Something. If you can't defend an assumption, it doesn't belong in the model.

02. The sign that flipped

Your Cost of Goods Sold (COGS) is growing. You've estimated it based on unit volumes and cost per unit. The formula looks like this:

COGS = Volume × Unit Cost

Except somewhere in your model, you have:

COGS = (Volume × Unit Cost) × -1

Or maybe it's buried in a larger formula and it just looks like COGS is negative. The spreadsheet still calculates. Your gross margin is wrong, but everything flows through—your operating income is wrong, your tax is wrong, your valuation is wrong.

Sign errors are invisible because Excel won't complain. A negative COGS flows through as a negative cost, which makes your margins look amazing. Nobody questions amazing until they dig deeper.

The fix: sense-check every line item. Does this number make logical sense? Is the sign correct? If COGS is negative, something is wrong. Don't let Excel's silence convince you the model is right.

03. The input that wouldn't move

You build a sensitivity analysis. You want to show how your valuation changes if growth assumptions vary. You create a table where one input changes and everything else stays constant.

Except the input didn't change. It's hard-coded.

You forgot that you pasted a number instead of a cell reference. So in your sensitivity table, every cell shows the same result because the underlying input was a fixed number, not a formula that pulls from the table.

This is the error that survives the longest because it's not technically wrong. Your model runs. Your outputs look clean. But you're showing analysis that isn't actually analyzing anything.

Check every sensitivity table. Make sure that the input you're varying is actually varying. Use cell references, not numbers. And spot-check the results: does changing the input change the output? If not, something is hard-coded.