A model worth questioning

A company reports that sales have risen by eight per cent. The presentation is handsome. Management speaks about a stronger brand and customers willing to pay more. A spreadsheet extends the growth across the next five years. By the time the valuation appears, the explanation has begun to feel like an observation.

This is the moment in investment research that interests me: the passage from something reported to something believed.

My earlier finance essays approached it through different doors. One looked at the discipline of making investment principles explicit. Another asked what a financial model could tell us without an understanding of the business. Later pieces tried to describe economic mechanisms and combine uncertain signals. The recurring ambition was to make judgement more deliberate and more open to correction.

I still want that. A useful model should give a question somewhere precise to land.

Follow the eight per cent

Consider an invented manufacturer that sells one product. Everything in this example is fictional; the figures are in Canadian dollars and cover two comparable quarters. We will use only units sold, selling price, production cost and other operating expenses.

The sales increase is real within the example. So are the fewer units sold, higher production costs and lower operating profit. Higher prices more than offset the fall in volume at the revenue line. They do much less for the amount left after costs.

The arithmetic lets us be specific. Revenue increased by $80,000. Production cost increased by $75,000, leaving only $5,000 more gross profit. Other operating expenses rose by $25,000. Operating profit therefore fell by $20,000.

Nothing in those calculations tells us whether customers love the brand. That remains an explanation to investigate.

They also do not prove that the business is deteriorating. Perhaps a supply interruption reduced deliveries, a temporary material shortage raised costs, or a worthwhile expansion increased expenses. Each possibility changes what the quarter might mean. The accounts give us a place to begin asking.

There is another question beyond this small table: did the company collect the money? Profit and cash flow are related, but they are different views of the business. Receivables, inventory and the timing of payments can matter to that difference.¹ The fictional table does not contain enough information to calculate cash flow, so I would leave that answer blank.

Give the explanation a fair test

The original mental-model essay drew me towards a business before its spreadsheet: who buys, why they return, what they can substitute and what it costs to serve them. Buffett’s account of a circle of competence gives that curiosity a boundary. We can concentrate on businesses we can evaluate while being honest about the limits of that understanding.²

In this example, pricing power is a useful idea because it produces questions. Were existing customers willing to reorder at the new price? Did some leave? Were competitors facing the same cost increase? Did the manufacturer deliberately give up low-margin sales?

I would keep at least two explanations alive. The company may have improved its ability to charge for something customers value. Or it may be passing through costs while losing demand. A supply constraint could explain part of the same pattern. These possibilities need not be mutually exclusive.

A good next observation is one that helps distinguish them. Repeat orders at the new price would tell us more about continuing demand than another confident description of the brand. Delivery records would help separate an inability to supply from an inability to sell. Changes in product mix would matter in a real manufacturer with several products; our one-product example deliberately leaves that complication out.

The aim is to give the explanation a fair chance to fail. If every possible result can be made to confirm it, the model has become difficult to learn from.

Change one assumption

Keep the later quarter’s price at $120, production cost at $75 per unit and other operating expenses at $275,000. Each additional unit sold contributes $45 towards those expenses and then operating profit, within this simplified model.

At 8,000 units, operating profit would be $85,000. At 9,000 it is $130,000. At 10,000 it would be $175,000. The range makes the importance of volume visible. It does not assign a probability to any outcome.

The entire calculation fits in a few lines of Python:

# Fictional CAD example: fixed price, unit cost and other expenses.
def operating_profit(units):
    return units * (120 - 75) - 275_000

for units in (8_000, 9_000, 10_000):
    print(f"{units:,} units: ${operating_profit(units):,}")

Someone can inspect every assumption and reproduce every result. They can also point to what is missing. At a different level of production, unit costs or staffing requirements may change. Competitors may respond to the price. A simple calculation earns its place by showing which questions are consequential.

This is what I now want from the mechanism language in the earlier writing. Describe how a result is produced, then examine whether those relationships hold. A formula can be deterministic while the business it represents remains uncertain.

The same distinction matters when combining signals. A percentage produced by an estimator depends on the starting assumptions, the observations and the rules used to connect them. Coding those rules makes them repeatable; it does not make the assumptions objective. Several indicators may also reflect the same underlying event. More inputs do not necessarily mean more independent evidence.

I would rather have an intelligible uncertainty than a precise number whose meaning I cannot defend.

Let AI sharpen the question

AI is useful here because it can help me take another pass through an argument. I can ask it to separate a reported figure from a proposed explanation, generate competing accounts, check arithmetic and identify which missing fact would matter most.

For the fictional manufacturer, a worthwhile prompt is: “What else could produce this combination of higher revenue, lower volume and lower operating profit? What would distinguish those explanations?” The response gives me research leads. I still need to find the records and decide whether the suggested distinction is sound.

For a real company, I would attach each material figure to its reporting period, source page and definition. A fluent summary cannot resolve a mismatch between a quarterly figure and an annual one. Code can check the arithmetic; someone still has to check whether the inputs belong together. If the evidence does not support a probability, I do not want the assistant to supply one just to complete the table.

I also want the original judgement kept. Write down what the evidence seemed to imply, what remained uncertain and what would change the view. When the next report arrives, compare it with that dated account before rewriting the explanation. A disappointing outcome may reveal a weak assumption, an omitted factor or ordinary uncertainty. The distinction is easier to examine when the earlier reasoning survives.

This is the thread I want to carry from the old finance work into the Ledger: careful attention to how a business works, the freedom to use mathematical and computational tools, and a willingness to question what they produce.

The eight per cent increase has become more interesting by the end. It is still eight per cent. Now it leads to customers, costs, capacity and a decision about what to investigate next. That is the kind of model I find worth building.