Common Modelling Mistakes Analysts Make

The most common modelling mistakes are not exotic. They are hardcoding numbers into formulas, showing false precision, building on restated history instead of point-in-time data, straight-lining growth with no driver behind it, letting circular references creep in, and running a single case with no scenarios. Each is easy to make, easy to miss, and quietly corrosive, because a model that looks polished can be wrong in ways that never show up until real decisions ride on it.

A financial model is just a chain of assumptions dressed up as arithmetic. The danger is that the arithmetic makes the assumptions look like facts. A clean spreadsheet with confident numbers earns a trust it has not necessarily paid for. Most of the mistakes below share one root cause: the analyst stopped being able to see their own assumptions clearly. Here is what goes wrong, and how disciplined modellers avoid it.

Hardcoding numbers that should be inputs

The first and most frequent mistake is typing a number directly into a formula. You are projecting next year’s cost and you write a calculation that multiplies this year’s cost by 1.08 inside the cell. That 8 percent is now invisible. It is not labelled, it cannot be audited, and if you decide costs should grow at 6 percent instead, you have to go find every place you buried it.

Every assumption in a model should live in its own cell, labelled, in an inputs area you can read at a glance. Formulas should reference those cells, never contain raw numbers. This sounds pedantic until someone asks you the only question that matters about a forecast: what are you assuming, and why. If your assumptions are scattered inside formulas, you cannot answer, and neither can anyone reviewing your work. A model you cannot explain is a model you cannot defend.

The habit is simple. If a number is a fact from a filing, it can be an input. If a number is a judgment about the future, it must be a visible, isolated input. Nothing that represents a decision should ever hide inside an equals sign.

False precision, or pretending you know more than you do

The second mistake is presenting a forecast with more decimal places than the inputs deserve. An analyst feeds in a rough revenue growth guess, a rough margin guess, and a rough tax rate, and the model dutifully returns a profit figure that reads as an exact number down to the last digit. The precision is real in the spreadsheet and fake in the world.

Output cannot be more precise than input. If your revenue driver is a genuine estimate, your earnings forecast is an estimate too, and dressing it up as a precise figure misleads whoever reads it, including your future self. The fix is to round outputs to a sensible level and, better still, to show a range rather than a point. A forecast that says profit will land somewhere in a band is more honest, and more useful, than one that implies a false bullseye.

False precision is dangerous because it feels rigorous. The more decimal places on the page, the more the reader trusts the work, exactly when they should trust it less. Real rigour is knowing how uncertain your number is and saying so.

Building on the wrong version of history

The third mistake happens before the model even starts. You pull a company’s ten-year history to anchor your forecast, and you pull today’s version of that history, complete with every restatement the company has made since. A demerged business has already been stripped out of the early years. A discontinued division has already been separated. The numbers you are extrapolating from were not the numbers that were actually reported at the time.

This quietly poisons both trend analysis and any backtest. Growth rates computed off restated figures describe a business that only exists in hindsight, and a model tuned on that history is fitting a past that was reshaped after the fact. This is the exact reason why point-in-time data matters: a serious model is built on the numbers that were knowable on each date, not the tidied-up version sitting in the database now. Skip that discipline and you have imported lookahead bias into your foundation without touching a single formula.

Straight-lining growth with no driver underneath

The fourth mistake is the lazy forecast: revenue grew 12 percent last year, so it grows 12 percent every year forever. A single growth percentage stretched across a forecast is not a model of a business. It is a guess wearing a trend line.

Real revenue is a product of drivers. It is price times volume, or subscribers times revenue per user, or stores times sales per store. A forecast is honest only when it forecasts those underlying quantities and lets the topline fall out of them. This is where revenue mapping does its work: instead of one number growing at one rate, you have a small set of drivers, each with its own logic, some accelerating and some fading. Straight-lining hides the fact that a company adding stores fast and a company squeezing more out of existing stores are completely different businesses with the same headline growth. The drivers show it. The straight line buries it.

The same trap applies to margins and working capital. A margin held flat across a forecast assumes nothing about the business changes, which is almost never true. Every projected line should trace back to something you can defend, not to a copy-paste of last year.

Circular references and fragile plumbing

The fifth mistake is structural. Interest expense depends on debt, debt depends on cash flow, cash flow depends on profit, and profit depends on interest expense. Left unmanaged, this loop makes a model circular, so it either errors out or, worse, quietly settles on an unstable answer that shifts every time you change an input.

Circularity is a modelling problem, not a business one, and there are clean ways to handle it, such as using opening balances for interest or a controlled iterative calculation you fully understand. What you must never do is leave a model silently circular and trust its output. The broader lesson is that a model needs to be built to a clear structure, with a defined flow from drivers to statements. A disciplined three-statement model links the income statement, balance sheet, and cash flow so they stay consistent and traceable, which is what keeps small errors from propagating invisibly. Fragile plumbing produces confident nonsense.

Running one case when the future has many

The final mistake is committing to a single forecast as if it were the answer. One case, one set of assumptions, one number at the end. The future does not work that way, and a model that offers only one path gives the reader no sense of what could go differently or how much it would matter.

The fix is scenarios. Build a base case, then a downside and an upside, changing the two or three drivers that actually move the outcome. The point is not to predict which case happens. It is to see how sensitive your conclusion is to the assumptions you are least sure of. If a small change in one input swings the whole result, that input deserves far more of your attention, and often more of your reading, including going back to what management has said about it. Grounding scenarios in real signals rather than round numbers is why management guidance belongs in the process. A model with no scenarios is a strong opinion pretending to be a calculation.

What to take away

None of these mistakes require advanced technique to avoid. They require discipline and a willingness to keep your own assumptions visible. Before you trust any model, run a short checklist:

• Isolate every assumption: no judgment about the future should live inside a formula.
• Match precision to certainty: round outputs and prefer ranges over false bullseyes.
• Use point-in-time history: build on what was knowable then, not today’s restated version.
• Forecast drivers, not trend lines: every projected number should trace to something you can defend.
• Keep the structure clean and stress-test it: control circularity, and always run more than one case.

A good model is not the one with the most tabs or the cleverest formulas. It is the one whose every number you can explain, question, and change. Get that right and the model becomes a tool for thinking. Get it wrong and it becomes a very persuasive way to fool yourself.

This article is educational. Altys Labs is not a registered research analyst or investment adviser, and nothing here is investment advice or a recommendation to buy, sell, or hold any security.