The most consequential software ever written

I have spent most of my working life inside a grid. Working papers behind a trial balance, unifying the account structures of newly acquired companies, payroll for a few hundred people, and later the estimates and resourcing models behind delivery work. Different jobs, different industries, same rectangle.

For a long time I thought of it as a container. Somewhere to put numbers that already existed. It took me an embarrassingly long time to notice that the grid was not holding my thinking, it was shaping it.

The innovation was not arithmetic

Calculators did arithmetic. Ledgers did storage, and did it well enough that we still use their vocabulary. Neither of those is what changed.

What changed is that a formula recalculates. Change one input and every figure downstream of it rewrites itself, immediately, for nothing. That single property converts a forecast from a document into a function.

Before that, a projection was a thing you produced. It took days, it was correct as of the assumptions you started with, and revising it meant redoing it, which meant you revised it rarely and defended it strenuously in between. Planning was an event. Somebody went away, came back with a document, and the document was then argued about as an object.

Afterwards, planning is continuous, because the marginal cost of asking a different question fell to nearly zero. “What if the volume is fifteen percent lower” stopped being a request for a week of work and became something you do in the room, while the person who asked it is still speaking.

What that did to the room

The authority in a meeting moved to whoever is holding the file.

Not the most senior person and not the person who knows the business best. The person who can answer a follow-up in ten seconds, because everyone else’s alternative is to wait a day and be answered by the same file anyway. I have been that person often enough to know it is a strange amount of power for what is essentially a clerical position.

The second effect is subtler and I think it is the real one. When scenarios were expensive you had to decide which question was worth asking, and that constrained the discussion productively. When scenarios are free, the constraint moves from producing the number to choosing which number to believe, and nobody was trained for that. Groups now generate more possible futures in an afternoon than they can form a view about in a quarter, and then pick one by a process nobody would write down.

The part I had wrong

I used to think the danger of the spreadsheet was error. A wrong formula in a corner cell propagating quietly into everything, and there are famous cases, and I have found a few unfamous ones.

Error is not the danger. Error is at least the kind of problem the discipline knows how to hunt. The danger is that cheap recalculation makes it very comfortable to keep asking and never commit, and it dresses that up as rigour. A model with nine scenarios looks like diligence and is often the opposite, because at some point the work of deciding has been replaced by the work of elaborating.

There is a smaller thing too, which is that a rectangle wants everything to be a rectangle. Whatever you are describing gets forced into rows and columns whether or not it has that shape, and the parts that will not fit tend to be dropped rather than represented, quietly, by whoever is building it.

The thing I keep noticing

Almost every consequential decision I have been near was made by a group of people looking at a rectangle that one person built and nobody else opened.

And the person who built it always knows where it is weakest. There is a cell somewhere they are not comfortable with, an assumption they would defend less strongly than the confidence of the output implies. Early on I said so out loud. After a while I noticed that saying so cost me something and changed nothing, so I said it less, and I am not sure I ever decided to.