learn-finance-with-phoebe / Practitioner session 10 of 10
Learn Finance with Phoebe · Practitioner session 10 of 10

The model bench

Nine sessions built a model one line at a time. This one puts the whole thing in front of you and lets you drive it. Change what customers pay in, change how fast the company grows, change the window you look through, and watch the answer change with it. Then break the balance sheet deliberately and see what a modelling error looks like from the outside, which is the thing worth practising, because from the outside it looks like every other run.

🔴 Hardest Anyone who will own a forecast Anyone who has to check somebody else's 45 min
0-3 · Welcome 3-18 · Drive it 18-34 · Break it 34-45 · Size the gap and close
Part 0

Where this lands

Session 1 wired the three statements together. Sessions 2 to 5 built each one and the check that holds them honest. Sessions 6 and 7 drove the model and measured the eight ways out. Session 8 sized the number to ask for and session 9 recorded a claim that died when the window changed. Everything below runs in this page, from the same assets/fin-live.js every other session used.

Live = worked through in the 45 minutes Self-study = the full depth, after class 🔍 Every diagram zooms

★ What you walk out with today. A model you have driven yourself across the whole assumption space, so the growth ladder is something you have seen rather than something you were told. The experience of watching a balance sheet stop tying and noticing that nothing else on the page looks any different. And the four-part answer that goes to whoever asks how much money the business needs.

LiveWhat the bench is actually doing4 min▶

Each press of the button runs the model month by month for the horizon chosen. For every month it computes the income statement from the assumptions, works out the three working capital balances from that month's trading, turns the change in them into a cash movement, and lands the results on the balance sheet through the two wires from session 1. Then it checks that the sheet ties.

None of the figures below are stored. Change a control and the previous answer is discarded and a new one computed, which is why an assumption you thought was minor can move the headline more than you expected.

Part 1 · the bench

Drive it

15 min hands on

Four controls. Start with the base case as it stands, then change one thing at a time and predict the direction before you press the button. Predicting first is the whole exercise: an assumption you cannot predict the sign of is one you do not yet understand.

LiveThe four runs worth doing first5 min▶
  • Base case, 24 months. Cash goes negative in month 10. Note the net income column beside it and confirm it never turns negative.
  • Base case, 48 months. Same assumptions, longer window, and the company comes back. The trough is minus £115,501 in month 19 and it is above zero again in month 28. Nothing changed except how far you looked.
  • Growth at 0%, 24 months. The gap disappears entirely and twelve month net income falls to £165,627. Growing less makes the company safer and poorer at the same time.
  • Customers pay in 45 days, 24 months. The gap disappears and twelve month net income stays at £445,631, exactly. Not approximately: the timing levers do not touch the income statement at all.
Self-studyTwo combinations worth finding4 min▶

First, set growth to 6% and the horizon to 48 months. The trough is minus £897,962 in month 31 and the company does not recover inside the window. Between 4% and 6% the problem stops being a bridge and becomes something the business cannot outgrow, and there is no warning label on the control to tell you that.

Second, put growth at 8% and leave the horizon at 12 months. Everything looks survivable, because at twelve months it is. The window is doing the reassuring, not the business.

Part 2 · the check

Break it on purpose

8 min live

Press "Break the tie". The button drops one side of a single transaction in month 7, the way a real modelling mistake does, and leaves everything else alone. Look at what happens to the page.

The same model, one wire cut Both wires connected net income into equity change in cash into cash Assets = liabilities plus equity worst error 2.24e-8 Across the 50 run canon set. That is floating point noise, not a mistake. Equity wire cut at month 7 net income goes nowhere change in cash into cash Assets exceed the claims against them by month 7 net income The gap then persists in every later month, because nothing puts it back. What the broken run looks like from the outside Revenue still grows. Net income still rises. The cash column still falls and still crosses zero in a plausible month. Every figure renders, every chart draws, and the whole thing is wrong from month 7 onward. The check is the only thing on the page that knows.
🔍 Click to zoom - the point is not that the model broke, it is that nothing else changed when it did.
LiveWhy a check you never see fail is not a check4 min▶

A tie test that has only ever reported success tells you nothing, because a test that cannot fail and a test that is passing look identical from the outside. The habit worth keeping is to break the thing deliberately, once, and confirm the check notices. Session 5 did that to the model; this button does it on demand.

The same applies to any check you inherit. Before relying on somebody else's validation, make it fail once on purpose. If you cannot make it fail, you have not found a robust check, you have found a decoration.

Self-studyWhat ties and what does not3 min▶

A tying balance sheet says the bookkeeping is internally consistent. It says nothing at all about whether the assumptions are sensible. A model can tie perfectly every month and still assume a growth rate nobody can deliver and a collection period nobody has agreed to.

That is the honest limit of the check and it is worth stating whenever the check is quoted as reassurance. Internal consistency is necessary and it is nowhere near sufficient.

Part 3 · build-along

Size the gap and write the ask

11 min hands on

Six steps, on the bench above, ending in four sentences you could send to somebody who controls money.

Find the trough, not the first bad month

Base case, 48 month horizon. Write down the first negative month and the deepest point. They are nine months and £108,069 apart, and only one of them is the number to finance.

Find the end

Read the month the cash position comes back above zero. That month is what turns the request into a term rather than an open-ended overdraft, and it is the single fact that makes the ask answerable.

Test the assumption that sized it

Move growth to 6% and run it again. The trough moves by an order of magnitude and the recovery disappears. Write down what that means for the confidence you can honestly attach to the first figure.

Price the alternative

Put growth back and set customers paying in 45 days instead. The gap closes without borrowing anything. Note what that change would actually cost in the real world, which the model does not know and will never tell you.

Break it and look away

Press the break button, then read the table as if you had just been handed it. Decide how long it would have taken you to notice without the check. That number is why the check exists.

Write the four sentences

How much, over what period, resting on which assumption, and what you will do if that assumption turns out wrong. Four sentences, no spreadsheet attached. If any of them needs a paragraph, the model has not finished telling you something.

The habit to keep. Report the horizon next to the answer, every time. A forecast without the window it was run over is not a forecast, it is a number, and the same model over two windows supported two opposite decisions in session 9.

Check yourself

Three questions

The base case shows cash first going negative in month 10 at minus £7,432. What should the funding request be built on?

The first negative month understates the need by more than a hundred thousand pounds. Funding it covers month 10 and nothing after: the balance is -£28,360 in month 11 and -£47,261 in month 12. The trough is the depth of the hole and the recovery month is what gives the facility a term.

You press "Break the tie" and the check reports a failure. What is the status of the other figures on the page?

The figures all still render and still look plausible, which is exactly the danger. An imbalance means the bookkeeping is internally inconsistent, and the check is the only thing on the page reporting it.

A model ties in every month across every run you have tried. What does that establish?

Internal consistency is necessary and nowhere near sufficient. A model can tie perfectly while assuming a growth rate nobody can deliver and a collection period nobody agreed to.

After the course

Homework

One hour, on a business you actually know.

Build the smallest version

Revenue, gross margin, fixed and variable costs, and the three working capital dials. Six inputs is enough to produce the effect this whole course has been about, and anything more can wait.

Put a tie check in before you put a chart in

Then break it once on purpose and confirm it fires. A check you have never seen fail is not yet a check.

Run it over two horizons

Whatever window feels natural, and then double it. If the two disagree, the longer one is usually the one telling you something, and either way the horizon goes next to the answer.

Covered today

What this session covers

✓Driving the whole modelGrowth, collection period, payment period and horizon, all live.
✓The trough against the first negative monthMinus £115,501 in month 19 against minus £7,432 in month 10.
✓Breaking the tie deliberatelyOne side of a month 7 transaction dropped, and what that looks like from outside.
✓Why a check you never see fail is not a checkMake it fail once, on purpose, before you rely on it.
✓The limit of a tying sheetInternal consistency says nothing about whether the assumptions are sensible.
✓The four sentence askHow much, over what period, on which assumption, and what you do if it is wrong.
◐What the timing levers cost in the real worldNamed in session 7; the model prices the cash and nothing else.
○Valuation, and what a business is worthDifferent tools, different subject; nothing here is investment advice.
Take this with you

Cheat sheet

Finance the trough. Not the first negative month. At Marchmont they are nine months and £108,069 apart.
The recovery sets the term. Month 28 is what makes it a bridge rather than a rescue.
Report the horizon. A forecast without its window is a number, not a forecast.
Break your check once. A test that has never failed is indistinguishable from a test that cannot.
Tying is not validity. Internally consistent, with assumptions nobody can deliver, is a real state.
Timing levers move no profit. Twelve month net income stays £445,631 across all four of them.
The model prices cash only. Customer goodwill and stockout risk are real costs it cannot see.
The four sentences. How much, how long, on what assumption, and what you do if it breaks.