I Have Trust Issues With Beautiful Financial Models

I have trust issues with beautiful financial models.

Not because I don’t like them.

I do.

Give me clean tabs, consistent formatting, sensible colors and a model where I don’t need an archaeological expedition to find the assumptions.

I’m happy.

But I’ve been around Finance long enough to know that a beautiful spreadsheet and a good financial model are not necessarily the same thing.

Sometimes they’re not even particularly close.

There’s something about opening a perfectly formatted workbook that makes your brain relax.

Everything is organized.

The colors mean things.

There are dropdowns.

Maybe a dashboard.

Someone clearly knew what they were doing.

Then you find this:

12.0%

Blue font.

Assumption cell.

Okay.

Why 12%?

And suddenly the room gets a little less comfortable.

Excel can make almost anything look official

This may be one of Excel’s greatest talents.

Someone types 12% into a cell.

It flows through six worksheets.

It changes revenue.

Revenue changes gross profit.

Gross profit changes EBITDA.

Eventually 12% appears in a beautifully formatted chart in a PowerPoint deck.

By the time it reaches the executive meeting, 12% has acquired a suit and a LinkedIn profile.

It looks extremely employed.

There’s just one problem.

Nobody really knows where it came from.

Maybe someone made an educated estimate six months ago.

Maybe Sales suggested it.

Maybe it came from last year’s budget.

Maybe the person who originally entered it left the company.

My personal favorite explanation is:

“That’s what we’ve always used.”

Nothing makes me want to keep asking questions quite like discovering an assumption has apparently become company folklore.

And this is why I spend a surprising amount of time looking at the cells without formulas.

That’s where the humans are hiding.

The formulas aren’t always what worry me

Obviously, formulas matter.

I’ve seen enough broken formulas to maintain a healthy level of paranoia about those too.

But Excel will calculate a terrible assumption perfectly.

That’s the problem.

Tell Excel customer growth will be 18%, and Excel says:

Fine.

Tell it gross margin improves 300 basis points next year.

Sure.

Tell it we’re hiring 47 people.

Absolutely.

It will calculate salaries, payroll taxes, benefits and start dates down to an impressive level of precision.

Excel has absolutely no opinion about whether finding and onboarding 47 people in six months is remotely realistic.

That’s our job.

And I think Finance occasionally gets seduced by precision.

$4,728,391

looks very serious.

It may have started with three assumptions someone more or less guessed.

That’s the part I care about.

I have a few annoying questions

If you’ve ever built a model and had someone like me review it, you probably know where this is going.

I start asking questions.

Where did this assumption come from?

Who owns it?

What would have to happen operationally for it to be true?

What evidence supports it?

What happens if we’re wrong?

And occasionally my favorite:

Does anyone actually believe this?

That question doesn’t technically appear in most FP&A textbooks.

I still recommend it.

Because sometimes everyone involved knows an assumption is optimistic.

Sales knows it.

Operations knows it.

Finance knows it.

The CFO probably knows it.

Yet there it is in the model, enjoying all the rights and privileges of a real number.

That’s when the spreadsheet has stopped helping us think and started helping us avoid an uncomfortable conversation.

I’d rather have the uncomfortable conversation.

Strip it down

One of the simplest things you can do with a complicated model is temporarily make it boring.

Ignore the dashboard.

Ignore the charts.

Ignore the formatting.

Pull out the handful of assumptions actually driving the result.

Revenue growth: 15%

Price increase: 4%

Churn: 7%

New hires: 32

DSO: 45 days

Now look at them.

Suddenly there’s nowhere for them to hide.

This is where I want to spend time.

Why 15%?

Why 4%?

Why 32 people?

Why 45 days?

Not because every assumption needs a 40-page research paper attached to it.

Please don’t do that either.

But somebody should be able to explain why we believe it.

And “because that’s what the model says” is not an explanation.

The model says it because we told the model to say it.

Sometimes I think we forget that.

A model should show me where we’re uncertain

There’s another thing that makes me nervous.

A model where everything appears equally certain.

Businesses don’t work that way.

We might have signed contracts supporting part of next quarter’s revenue.

Pretty high confidence.

We might have a sales pipeline supporting another piece.

Less confidence.

Then somewhere farther out we’re estimating customer growth, hiring, pricing, retention and seventeen other things while pretending December is just as knowable as next Tuesday.

It isn’t.

That’s fine.

A forecast isn’t supposed to eliminate uncertainty.

If it could, I’d be writing this from a much larger house.

The job is to understand where the uncertainty lives.

Which assumptions could materially change the outcome?

Which ones are weak?

Which ones depend on several other things going right?

Which ones would actually change a decision?

That’s much more useful to me than pretending every cell deserves the same confidence because they’re all formatted the same way.

AI is going to make this problem more interesting

We’re entering a period where making finance work look impressive is getting very cheap.

AI can help format.

Build charts.

Write commentary.

Create presentations.

Summarize variances.

Eventually we’ll probably ask it to sit through the budget meeting for us.

I’m waiting patiently for that feature.

This doesn’t make Finance less valuable.

But I think it changes where our value sits.

If everyone can produce polished analysis quickly, polished analysis stops being particularly impressive.

The scarce thing becomes judgment.

Someone still has to ask:

Why do we believe this?

Does this make sense given what we’re hearing from the business?

What would prove us wrong?

Are we seeing a real trend or just timing?

What decision does this actually change?

AI can make a bad assumption look fantastic considerably faster than we used to.

That’s progress, technically.

Finance still has to recognize that it’s a bad assumption.

And yes, I still want the spreadsheet to be pretty

Every time I make this argument, I can practically hear someone preparing to send me the ugliest spreadsheet ever created as proof that formatting matters.

It does.

Please format your models.

I don’t want 14 tabs of black numbers on white backgrounds named Sheet1 through Sheet14.

I’m not an animal.

Good formatting makes thinking easier to follow.

It helps someone understand the model without having the creator sitting beside them narrating every click.

That’s valuable.

I just don’t want presentation doing work that the underlying thinking hasn’t earned.

So my order is pretty simple:

Get the thinking right.

Then make the thinking easy to understand.

Because I will happily admire your beautiful financial model.

I’ll compliment the dashboard.

I’ll appreciate the clean assumptions tab.

I’ll even respect the color scheme.

Then I’m probably going to click on that blue 12.0% cell and ask where it came from.

I can’t help myself.

My husband would probably tell you I do the same thing with good news at home.

Finance just gave me Excel so I could make it everybody else’s problem.

I start with the business architecture before the workbook architecture

Before I admire tabs, I want to understand the business the model is supposed to represent.

How does revenue happen? What creates cost? What consumes cash? What constraints matter? Which decisions is management trying to make?

Then I can judge whether the workbook architecture makes sense.

A model can be beautifully organized around the wrong mental model of the company.

That is a harder problem than a broken formula.

Consistency can hide a consistently bad assumption

One reason polished models feel trustworthy is that everything ties together.

Revenue flows into receivables. Headcount flows into payroll. Depreciation flows from CapEx. The balance sheet balances.

Excellent.

If the revenue assumption is fantasy, the model can still be internally consistent all the way to the bank balance.

That is why reconciliation is necessary but not sufficient.

A model can agree perfectly with itself.

I would also like it to have a working relationship with reality.

I look for false precision

There is something deeply comforting about a forecast that says next year’s EBITDA will be $18,437,219.

Very specific.

We apparently know the future down to the dollar.

I am suspicious.

Precision in formulas is not the same as certainty in assumptions.

If revenue depends on customer behavior, hiring dates, pricing and a product launch, I would rather see management understand the range than become emotionally attached to the seventh digit.

Sometimes the most honest model is the one willing to admit what it does not know.

A good model makes change explainable

When the forecast moves, I want to know what moved it.

Revenue timing. Hiring. Margin. Collections. A capital project.

If the model is so interconnected that nobody can explain why EBITDA changed $2 million without tracing formulas for an hour, beauty has not rescued us.

The architecture should help Finance build a forecast-to-forecast bridge.

Management needs the story of the change more than it needs admiration for the workbook.

I test the boring parts too

Beautiful models tend to draw attention toward the clever parts.

I also look at dates, signs, units, opening balances, cash flow treatment and whether annual and monthly views actually reconcile.

Some of the most damaging errors are spectacularly uninteresting.

A formula points one column too far. A percentage is entered as 12 instead of 12%. A monthly assumption gets treated as annual.

No dramatic modeling philosophy required.

Just an expensive Tuesday.

I want to know who can maintain it

A model can be excellent and still be wrong for the organization.

If only the builder can operate it, the company has acquired a dependency.

Sometimes that complexity is justified.

Often it is not.

I like models that a capable finance person can understand, update and audit without receiving oral history from the creator.

Documentation helps. Clear inputs help. Sensible formulas help.

Restraint helps more than modelers like to admit.

Scenarios reveal whether the model actually behaves

One of my favorite tests is to change something meaningful.

Revenue down 15%. Hiring delayed. Gross margin compressed. Collections slower.

Does the model react the way the business should?

Do costs that are supposed to be fixed remain fixed? Does cash respond? Do capacity constraints appear?

A model can calculate the base case correctly and behave strangely the moment management asks a real question.

Scenario testing is where some very attractive spreadsheets begin confessing.

The best-looking models make the thinking visible

This is where aesthetics genuinely matter to me.

Good formatting can show inputs, calculations and outputs clearly. It can make review faster. It can help someone understand the logic without clicking through every cell.

I am not anti-beautiful-model.

I am anti-beauty-as-evidence.

The presentation should reveal the thinking, not distract from whether the thinking makes sense.

Once the model earns trust, make it gorgeous.

I will enjoy that part too.

I also ask what the model refuses to show

Every model has boundaries.

Maybe it shows the P&L beautifully but barely touches cash. Maybe it models revenue in detail and treats working capital as a percentage. Maybe it has a perfect base case and no useful downside.

That is not automatically a flaw.

No model should contain the entire known universe.

But the boundaries should match the decision.

If management is deciding whether it can afford a growth plan, a model that stops at EBITDA is missing a fairly important ending.

I want to know what the model was built to answer—and what it was not.

Beautiful models become dangerous when presentation makes their boundaries invisible.