The Crystal Ball

Some financial models are built to be understood. Others, presumably, are built to overwhelm.

The second type has a logic of its own — or so it might seem. If the model is complex enough, the investment committee stops trying to understand it and starts assuming that anyone who built something this elaborate must know what they are doing. Shock and awe, applied to a spreadsheet.

Except it does not work. A seasoned investor once told me that it is almost standard practice — models that try to hide the real costs and risks somewhere in tab 86 on row 302. They know what they are looking at. Complexity is not a disguise. It is a flag. The more tabs, the more the experienced eye starts looking for what is buried in them.

The eighty-tab monster was not a strategy. It was a confession.

Whether the complexity was intentional or simply the accumulated result of building without a plan, the effect was the same: nobody could find where it started, which meant nobody could question whether it was right.

When the person who built it left, the model stayed. And with it, the task of turning it into something that could actually be used.

A colleague offered to help. The intention was genuine: take the eighty-tab monster, strip it back, produce something a Sales Manager could use to sketch scenarios, test assumptions, and present to investors without needing an archaeological expedition to find where the numbers came from.

The simplified version arrived. It looked good. Clean tabs, clear structure, numbers appearing where numbers should appear. At first sight, it looked like exactly what was needed, the labyrinth replaced by a corridor.

Then I tried to change something: what happens if the market price moves, what happens if the timeline shifts, what happens if the capital requirement changes.

Nothing moved.

I changed another input. Still nothing. I went looking for the formulas — the connective tissue that should have been linking assumptions to results, inputs to outputs, one tab to another.

There were no formulas. Not broken formulas. No formulas. Cell after cell of values, pasted in place, sitting there with the confidence of calculated results and the responsiveness of a photograph.

Later on, someone explained what had happened. The colleague who produced the simplified version does not really understand spreadsheeting. Faced with an eighty-tab monster they could not quite parse, they had reached for an AI tool — fed it the original, asked for a simplified version, and received something that satisfied the brief in the most literal possible sense. Fewer tabs. Cleaner layout.

The AI had done exactly what it was asked to do. And in doing so, it had removed the one thing that makes a spreadsheet a model rather than a table: the logic that connects the inputs to the outputs, that allows a number to change and have consequences, that makes the model breathe.

The model was now clinically dead. Outputs present, logic absent. The appearance of a financial model without any of the properties that make one useful.

The journey from shock and awe to clinically dead had produced, at both ends, the same result.

The original was a crystal ball — so complex that its opacity was the point. You could trust the outputs, or you could not because there was no way in. Whether by design or by accumulation, the complexity served the same function: it made the model unquestionable.

The simplified version was an even more perfect crystal ball. The original had formulas somewhere. You could, in principle, dig far enough to find the logic. The simplified version had nothing to find. It was all surface — outputs without mechanism, conclusions without reasoning, a financial model with no heartbeat.

At the moment when the proposal needed to be stress-tested, when scenarios needed to be modelled, when an investor in Paris asked a few simple 'what if' questions — there was nothing inside it that could respond to anything.

Two crystal balls. One built through too much. One built through nothing at all. Both equally useless when the room gets serious.

The fix, in both cases, starts with the same discipline: know what is an input and what is an output, and make that distinction visible. A model with clearly marked inputs can be questioned, tested, and trusted. A model that hides its assumptions, whether behind eighty tabs or behind pasted values, cannot be any of those things.

We wrote about that discipline in A Spreadsheet Is Not a Notebook — and why a subtle highlight on every input cell is the simplest possible signal that the person who built this understands what a model is for.

What a crystal ball costs when it reaches the wrong desk — and why auditability is not a bureaucratic nicety but the foundation of any serious financial conversation — is in the next layer: The Principal and the Agent.