What survives compression

The last step of my month, for a couple of years, was cutting the month down.

Everything that had happened across several business units went into reconciliations, and the reconciliations went into schedules, and the schedules went into a pack, and the pack went into a handful of slides, and in the room the handful of slides became three sentences because the meeting was behind. Then somebody who had been in that meeting said one sentence to somebody who had not.

Call that a thousand to one. I was the one doing the compressing at every stage except the last, and the last one is the only stage that mattered.

Conclusions compress and confidence does not

Here is the thing I did not see for a long time. Compression is not lossy in a random way. It loses the same class of content every time, and it is a class I was producing deliberately.

A conclusion is short. It survives every stage, because it is the thing each stage exists to carry. What sits around the conclusion is the material that tells you how much to trust it, and almost none of that material is a sentence. It is carried in hedges, in qualifiers, in the fact that one item got four paragraphs and another got a line, in the tone of a caveat, in how much detail I was willing to go into when asked. All of that is exactly what a person cuts first, because in every draft it looks like padding, and at the sentence stage it is simply gone.

So the figure arrives at the far end intact and stripped. Not distorted. Cleaner. That distinction took me years, because I kept looking for a misunderstanding and there was never one. The number was right. The certainty around it was manufactured by the act of transmission, out of nothing, and every person in the chain was doing careful work.

I blamed the wrong party for a while

My reading at the time was that executives over-read numbers. That they wanted certainty and rounded toward it, and that the caveats were being ignored.

That is unfair and it is also lazy, because it locates the problem in a person rather than in a mechanism. Nobody in the chain ignored the caveat. The caveat was not in what reached them. A listener cannot discount a claim by an amount they were never told, and there is no honest sense in which they had access to it. I had put my uncertainty into a format that cannot travel and then been disappointed that it did not.

The fix follows from stating it that way. If the conclusion survives because it is short and repeatable, then the uncertainty has to be made short and repeatable too. Not hedged into a sentence. Built into the shape of the claim so that a person repeating it has to carry it.

The version that works is roughly: here is the number, the one thing that would change it is this, and we will know by the fourteenth. That survives, because it is compact and because it has a date in it, and a date is the single most repeatable object in any organisation. Somebody will carry that sentence to somebody else largely intact. A paragraph of qualifiers will not make it out of the room.

Where I overcorrected

I went through a stretch of leading with the uncertainty, on the theory that whatever comes first survives, and that was worse.

A caveat that gets compressed without its conclusion is not caution. It is a rumour that the number is wrong. I watched a perfectly good figure become untrusted for a quarter because the sentence that travelled out of the room was the risk to it rather than the figure, and the risk did not even materialise. Having no confidence attached is a defective transmission. Having negative confidence attached is an active one.

So the rule is not that uncertainty goes first. It is that uncertainty has to be as compressible as the conclusion, which is a genuinely hard writing problem and which I do not solve every month. Some uncertainty simply is not compressible. Some of it is four interacting assumptions and there is no honest short version, and in those cases what I actually do is pick which of the four to mention and suppress three, which is a decision I make on my own with no method, in the last ten minutes before a meeting.

The smaller thing

What I came to is not really about executives or about packs.

Every account of your own work that reaches anybody else has been through at least one compression you did not perform. The last one is always done by somebody else, in your absence, from memory, in a corridor, and you have no influence over it except in one respect. You get to choose which sentence is the easiest one to carry.

I did not know that was a choice for most of the years I was making it.