ai-in-finance

[s35 candidate] Compared to what? The board question that comes next

By Samer Azar, Fractional CFO · 2026-11-03 · 5 min read

Key Takeaway: When a number misses, don't start with what went wrong. Start with which assumption broke, and when you knew. Write your assumptions down next to the budget the day you set it, and the board meeting becomes a check, not a hunt.

Dear reader,

Every time I present financials to a board, someone asks "compared to what?"

Compared to last month. Compared to last year. Compared to budget. That one is easy. Any spreadsheet answers it.

The question that catches people out comes next: which assumptions broke, and when did we know?

Most finance teams can't answer it. The assumptions live in a meeting that happened four months ago. The target lives in a spreadsheet nobody updated. The strategy lives in someone's head.

So the number shows up naked, with no link to the decisions that created it.

If you'd rather talk through how your own board pack explains a miss, book a 30-minute call with me.

Compared to what we assumed

A revenue target doesn't exist on its own. It's built on assumptions.

Here's a made-up example. A business sets its target for the half on three assumptions: a new market open by the third quarter, a product launch in April, and a summer peak like last year's.

The half comes in 27% below target. The usual conversation is "what went wrong?", and it goes round in circles.

Now take the assumptions one at a time instead:

1. The new market opened on time. That one held.

2. The launch slipped to June. The team knew in March.

3. The summer peak came in lighter than last year. Nobody wrote down how much lighter would hurt.

The 27% is no longer a mystery. It's two broken assumptions, and one of them was known three months early.

The board's question changes too. It stops being "what went wrong?" and becomes "why didn't the forecast move in March?" That's a harder conversation, and a much more useful one.

When a number misses, how fast can you say which assumption broke: the same day, within a week, or we never really find out? Reply with your answer.

Write the assumption next to the number

The fix is boring, and it works. The day you set the budget, write every assumption behind it on one line:

• What we assumed. "Launch in April."

• By when. A date, not a quarter.

• Who owns it. One name.

• How we'll know early. The first sign it's slipping.

Then every month, before you report the number, mark each assumption: holding, wobbling or broken, and the date you first saw it.

If you can't name the owner of an assumption, nobody is watching it.

Before you explain a miss, check the number

There's a step before "which assumption broke?": making sure the number that missed is real.

When a figure is wrong, it's almost always wrong in one of three ways. It was invented. It was true once and nobody updated it, like a budget from before the reforecast. Or two places show two different numbers for the same thing.

The second one is the same problem as the assumptions: something true four months ago, still sitting in the pack. So I ask one question for each:

1. Invented? Where exactly did this figure come from?

2. Stale? Is it from the latest close, and which version?

3. Two versions? Does every mention of it agree with the ledger?

Copy them into your review checklist. Nothing goes out until every number has an answer to all three.

The gap is where the value is

Most of the work in a board pack goes into the number. Most of the value sits in the gap between the number and the story behind it.

Three moves for Monday:

1. Pick your biggest budget line and list the three assumptions behind it.

2. Put a date and an owner next to each one.

3. Before your next board pack, mark each one holding, wobbling or broken, with the date you knew.

Don't ask what went wrong. Ask which assumption broke, and when you knew.

Where AI helps here, and where it doesn't

Files from building the AI CFO. Why I never let AI produce the number in a board pack, only the words about it.

The assumption log above is exactly where people want AI to help. It can, if you give it the right job.

My rule: we never ask the model for a number, we ask it about a number. Code computes the figures from the books. The model writes the sentence that explains them, using only the numbers it was given.

So for a board pack, the AI can draft "revenue is 27% below target; the launch assumption broke in March and the forecast didn't move." It can't decide whether 27% is right, and it can't decide which assumption to blame. A person signs that, with the three questions above in front of them.

Steal it:

• Inputs: the budget, the assumption log and the actuals, nothing else.

• The test: every sentence in the commentary points to one number and one assumption. If it can't, it comes out.

• The human: whoever signs the pack reads every sentence against its number.

If you run a finance team, forward this to whoever writes the board commentary.

Not subscribed yet? Join the CFO Lab

Which describes you best? One tap. It tells us what to send you.