Probability-Weighted > Defensible > Reasonable

I have heard some version of this many times in FP&A.

"This forecast seems reasonable."

But is that actually enough?

Reasonable usually means the number passes the sniff test. Nothing obviously broken. If revenue grew about 10 percent the past few years, forecasting 10 percent again next year feels reasonable.

The problem is that reasonable is a low bar.

AI models can now produce forecasts that look perfectly reasonable. Clean charts. Logical sounding assumptions. Numbers that fall in a believable range.

That does not make them correct.

The next step up is defensible.

A defensible forecast changes the conversation from opinion to methodology. Instead of saying "this seems right," you can show why the assumption holds up.

🛠️ What makes an assumption defensible

1. Evidence-based. Start with a baseline grounded in reality. Historical trends, internal benchmarks, or external market data.

2. Correlated to the business. Assumptions should connect to operating drivers. If revenue grows, something else must move with it. Sales capacity, marketing spend, implementation teams, COGS. Growth rarely appears in isolation.

3. Tested under pressure. A defensible model survives basic stress tests. Change the driver. Break the assumption. Run sensitivities. If the model collapses under small changes, it was never defensible.

But even defensible assumptions are not the end goal.

The real goal is probability-weighted scenarios.

Instead of asking "Is this forecast reasonable?" the better question is:

"What are the realistic outcomes, and how likely is each one?"

📉 What that looks like in practice

Base case built on defensible assumptions. Upside case tied to specific operational wins. Downside case tied to identifiable risks. Probabilities assigned to each scenario.

Now the model becomes a decision tool.

Leadership can see the distribution of outcomes, not just a single point estimate. Capital allocation, hiring plans, and risk management all get clearer when the forecast reflects uncertainty instead of hiding it.

AI will make it easier to produce numbers that look reasonable.

That raises the bar for FP&A.

The job is not just to generate forecasts. It is to build models that hold up to scrutiny and reflect the range of outcomes the business might actually face.

Reasonable is the starting point.
Defensible is better.
Probability-weighted is the standard.