Somewhere Between Six and Seven

Reality does what it will.

Man plans, God laughs. Everybody's got a plan until they get hit. Life is what happens to you while you're busy making other plans.

We all know reality doesn't play nice, refuses to be easily pondered or neatly presented. We want it to play along, so so badly, but it doesn't and never will. Things happen in intervals of their own choosing. Sometimes random things happen three times in a row. Our human brains take in thing after experience after event, guess which ones mean something, decide "hmmm that one probably." Then, we backtrack. Then un-backtrack. Our brains are laughably incapable of seeing three things in a row without kicking off at least a theory.

So, we smooth.

Everywhere, we smooth, mostly without ever calling it smoothing. I barely ever think to call it that, it's kind of a weird phrase but I don't know, it's the one. Somebody you know acts strangely, just once, and you don't change your feelings in their direction because, well you can't say it, but they looked pretty tired yesterday. Really tired. Chalk it up to bad sleep. One-off. Or, a terrible week is demoted quickly to just “a rough patch.” We take all these constant little observations and fit them into memory as something smaller, so that we can think about them and compare to all the other weeks, all the other days that person hadn't been up all night taking care of their sick child.

We have to do this; it's how we simply get by day to day. There's stuff to do, and we need to get it done. The alternative, perfect memory of every single thing, well I don't know what that would be like. Maybe horrifying? Exhausting at least.

The trouble is that occasionally your weird Tuesday really shouldn't have gone through that smoothing process because it's something or the beginning of something. And does that Tuesday come with a little tag attached telling you what you have on your hands, the smooth-it-out thing or the pay attention thing? Unfortunately, no it does not.

Forecasting pulls the issue to the forefront. We ask economists what xyz thing will do, what it means, etc. We ask the weather people to tell us what the sky will choose that day. What's the hit rate on these asks? You are asked in your day job, what is going to happen? Will this be done? Will that come in?

We have to make decisions, make calls, and someone has to commit. The dice are in your hands. Time to roll them bones.

A number moves. Something changed, but maybe nothing changed. Uh oh, did it change but not for the reason everybody currently would like. Meanwhile, unlike your opinion (since: smoothed) about somebody's weird Tuesday, the number jumping around has cash decisions sitting on top of it, so eventually somebody has to decide whether or not it's part of life now.

Which is itself a type of smoothing: acceptance that stuff is now jumping around in a new way. And once the evidence or events or whatever forces us to admit the world has now changed, we immediately smooth the changed world into a new baseline. That bad week is now just an upcoming set of slightly worse weeks overall. A few bad Tuesdays, and now you've a cranky coworker.

Reality doesn't really recognize the calendar. March ends because we say it does because Caesar said it does. It's over. Yet midnight arrives and somewhere a truck is still driving. Your project is only halfway finished, and you're also needing a particular signature to come in and what if it comes in in April but is effective Mar 31st? Your machine has five useful years left according to one estimate and seven according to another. None of these things really care when the day ticks over and the month closes. It's not exactly a spiritual event.

But there's a Balance Sheet and Income Statement to make and tea leaves to analyze. For finance teams or capital-F Finance the department, it's the starting gun. Gotta close.

Someone has to draw the line somewhere. This thing happened before the line, and that one happened after it. This expense belongs here, truly, it's happy here. Don't move it from where it's happy. That one? Has to go in pretty much all the months. This weird month probably means very little, it's actually this other month that might require everybody rethink everything.

A surprising amount of the work is taking things that happen continuously and forcing them into boxes that can be compared with other boxes. Months. Quarters. Years. Actual. Plan. GaaP. Non-GaaP. One version of the future against another, set against a past set of prior boxes.

Because reality is coming up. Reality gets another turn.

Neat little boxes create other problems, though. Once you've somehow captured bits of reality like little Pokemon, you have to decide what the differences between the boxes mean.

March was worse than February. Okay. Is that information? Maybe. Should it be...smoothed?

Who knows?

Finance has an odd relationship with certainty. It's a function, team, position, whatever you want to call it, that spends all day producing and thinking about numbers that look (extremely) specific about things that, if you gave everyone truth serum, probably (?) no one could possibly know with specificity.

How long will that laptop remain useful? Five years. Why five? Well, actually three. Some of this inventory is wet (oops), how much does that change what it's worth? And while jiu jitsuing questions like these, somewhere along the way uncertainty has to become something closer to certainty like: $4,783,219.

Really? That $219 at the end there? You sure about that?

By the way, this is all normal. I'm not sure what else anyone would expect a Finance person to do. You can't put “somewhere between probably and the Lord knows” into things you need to use to get any work done around here. Maybe round a bit more? $4,783,000 then. And just use GAAP.

So Finance spends a remarkable amount of time taking things still happening, or which haven't yet happened or might not even happen, and giving them fleeting numerical identities. Estimates. Reserves. Forecasts. Little fireflies that flash alive in one sheet or another for a brief time.

Then reality gets its next turn, rolls them bones.

Now, you get to compare your little fireflies with whatever reality just did. And while you're looking at a six and somehow a seven (do dice come with seven?), reality throws the next set of dice right in your face.

So, seven. The dice you're used to go one through six. What to do with seven?

You can "adjust your priors" or use whatever dressed up language you want, but do you now accept a world of seven-sided dice? Will some dice have six sides, but one of those sides will have seven dots? Has something genuinely changed? Or is somebody (reality) using trick dice. It's hard to decide whether you could see seven again, next and oh god what if there's an eight? Is six already how things used to work?

New information brings a kind of charisma to the game. Something that just happened has an unfair advantage over all those locked and caged fireflies that came before. Bad information comes with charts, explanations and concerned facial expressions. Good information comes with all that too, only everyone's skin looks a little healthier. Either case can make the previous prediction feel stale or even embarrassing: "we should have known about sevens."

So, someone (often Finance) has to decide how much authority to give this new, charismatic info. It's authority dial time: if every new observation turns that dial to the "full authority" side, suddenly your view of the world starts to seem like a loosely hitched trailer, stuck behind reality, and now it's whipping around. Set that dial to "no authority" and you'll probably end up spending a long time explaining why reality hasn't actually changed while reality is standing right over there, very obviously changed.

Somewhere in between, you tune that dial. You smooth.

Which sounds pleasant. Smooth it out with some math, and math is reassuring if you never learn about Russell's Paradox. Even still, sometimes that math gets the job done. But the judgment underneath is much messier: the story needs a dial adjustment, but how much?

Right about here is where it can all turn nasty. Noise being filtered and signal being damped can look almost identical. So which are you doing? Filtering noise, or dampening signal?

That weak March may be random, or the first month where something happened you'll later wish you'd noticed a lot earlier. A strong March? Hey, things improved around here. Or maybe not, after all it seems that future good news slated for April arrived ahead of schedule. March just borrowed from next month.

You'll know which one it was eventually.

Eventually does you no favors today, though.

A while ago, a few of us used to run a monthly forecasting competition at the company we were at. The word "competition" is overstating what it was. The "prize" was the right to be annoying until the next month. But it was fun, and everyone submitted the same handful of numbers for the upcoming month. Revenue. Expenses. Cash. A couple of others depending on what was going on in our corner of the world.

Five big numbers or so, nothing exotic. Just: what do you think it will all net out to? The collective spent hours and hours of life thinking on and around those numbers and guided the machinery that produced them, which should have made predicting them easier. Some one should have been cleaning up every month.

Nobody did.

One entry: nailed revenue, wrong on cash. Somebody else: expenses almost exactly right, off on revenue. You'd stare at your entry, then the actuals, and get that same thousand-yard stare when you clean that little silver stuff off a scratcher ticket. Sorry, try again. Next time, kiddo. Why do I do this?

You'd feel extremely clever about one number, the next one was revealed, then the next, and you'd discover you knew far less than you thought.

Of course we missed. That's not even that interesting; everyone misses all across the world all the time. But thinking back on that competition, what always got me was that everyone had... pretty good reasons. Their misses made some sense.

Same company. New month heading right at us. Same five questions. But vastly different ideas on how to tune those authority dials.

You could even hear the logic when people explained their choices afterward, held up their failed scratchers. I thought this or that mattered more. Or, I figured that was a one-off. Or, yeah I knew that was supposed to happen, I just didn't think it would.

In miniature, that competition is the whole problem. Knowing more and more and more doesn't make the dial disappear, it only adds to the pile of things that need a decision: how much of this do I believe?

The secret superpower of the competition, now obvious to me in retrospect: losing didn't matter at all. There were zero promotions on the line. No one was getting panicked phone calls or hauled into a tense, unhelpful meeting. At worst, you'd get some light teasing.

Which made everyone willing to play the game, truthfully. "Here is exactly what I think, written down."^[1]

But add consequences and formality to "forecasting." The fate of the economy rides on your interest rate dot plot. The fate of the army relies on your munitions estimates. Your scratcher ticket now determines the fate of the world.

Which is kind of the world we live in. Sometime. Some of what we're doing matters, even if we're not always sure which of it does. Every single dial tuner behaves differently in this world. Once the competition becomes a practice that determines real world consequence, smoothing has another job to do beyond being only a question of belief. Now, it's also a question of what beliefs can the organization reasonably act on, quick, before reality gets another turn to throw again.

And the throws keep coming day by day: a six, a seven (weird), a five, back to six (phew), an...eight (huh). An eight on a six-sided die. Meanwhile, your numbers have tenants now. Somebody has moved into six and is building stuff.

That tenant can't really work in days and honestly doesn't care right now because they're unpacking their boxes and setting up shop. You smooth for that tenant. Six is fine, you say, stay in there.

See, smoothing isn't just a statistical technique. Some people can't operate on the same clock as what you're measuring, so a lot of the time smoothing is for the people downstream from it at least as much as the people doing the smoothing. You don't want your tenant in unit six asking if they have to move all the time.

If the forecast changes faster than those changes can be usefully absorbed, it can't coordinate anything. So you hold it in place, smooth it out, adjust a few dials, when you need to. Of course, you have to do all this and make sure you don't miss that now the six-sided dice seem to have a seven on them a lot more often.

Eventually someone notices all these sevens. That's more than usual, right, and where are they coming from? Hurry up, I have a meeting today about this and what's going on here?

In moments like this, you can find people at all levels of doing forecasting (central banks, banks, finance teams on earnings calls, weather people) reaching for words just perfect for the job: timing, seasonality, one-time, mix, temporary. Transitory, we all remember that one.

There are practically infinite spices in the spice rack. But “timing” in particular is great because time is the fourth dimension, and we can all easily grasp what's meant by it: [the thing] happened, it just happened at an unforeseen time. No big deal.

Timing. Issue: explained, resolved. Excellent.

Words are the biggest smoothers around. And again, smoothing might be completely right. One number jumping doesn't immediately signal anything. Sometimes the deal is one day late. Who. Cares. The danger is that a useful explanation can become a little storage box.

Seven goes in. Close lid. Label: Timing.

Then another seven shows up. Huh. Hmmm. Timing, right?

Well, there is room for it in the box...

Right about here is where smoothing gets hard to distinguish from explaining away. The first feels like good, practiced judgment. The second? Also feels like judgment, but it's one of those entities in disguise: a bad bargain.

Eventually the Timing box fills up with sevens. Someone peeks inside and they are just falling out of there, spilling all over the floor. The label "Timing" is no longer an explanation but a threat.

You can imagine a lawyer for reality walking in and dropping exhibit after exhibit on the table (Seven, seven. Another seven) screaming "ladies and gentlemen of the forecast, the days of Six are over." Nope. Nothing like that. It will go like this: someone will look at the last few periods you care about, poke around in that timing box, look back at the periods, back at the box, then say something like "okay, definitely think we shouldn't call this timing anymore, team."

The sevens had been there for a while, but what changed was the Royal We and our willingness to alter the picture.

Does that family in Six give up their apartment, move into Seven? Not usually, they just knock down a wall without a permit (illegal!) and now they're in Six-Point-Three or even somehow a range (Six-to-Seven) or just Six*. The existing model is more likely to bend to what reality has been saying, which is a universal learning, like a law of physics. It's very rare in life to discover The Truth. More likely to happen? You lose confidence in whatever explained the Old Truth, so the Old Truth gets a little update.

So how big is the update, then? Once you've admitted the world changed, how much do you change with it?

It wasn't Timing! Seven is everywhere; it's all anyone has ever heard of. Seven will destroy the economy. And now we are jiggling the authority dial endlessly, and now we are spinning it like the wheel of a pirate ship, trying to make sure we handle that Seven isn't Timing.

It's plain to see why. The previous reaction was, perhaps, an underreaction. Now that everybody can look backward and see all those sevens sitting there, a great tightening commences. All forecasts from here on out: more conservative. Spending anything on anything: needs another approval. Secret buffers are added to the buffer of the buffer. We are at war with or on behalf of Seven, no one really knows that part yet. Targets must adjust and standards must harden. Every project now gets asked to denounce Seven but also perhaps, would you like to work for Seven?

Six really did move. You learned something and now you all agree the system should respond to it. It's just that the right order of operations, and how hard to operate those operations, well there is no public playbook for any of that.

And usually, in my experience, people remember with an initial enthusiasm that wanes but persists. Memory is an interesting thing, and it often spoils a bit in old age, gets edited. Which brings out the probable problem when you over-learn: what should be a temporary response to Seven will probably have staying power. We like to put things into systems, like buffers and controls and approvals, but don't often incentivize taking them out again.^[2] In groups of people, it's like you put your hands on a hot stove once. Now you're all wearing oven mitts outside, and you're not in a kitchen.

Learning can make you react too much. Which is why we smooth. And we can smooth too much, so the learning comes a bit too late. And round and round we have to go.

Eventually you have some plan that needs a number. Those dice and dials and learnings are converging on what happens now, which is that a number is needed. You need a number.

At best, what you choose will be accurate. You will hit the mark, arrow through the apple. Underneath that at best, there's another, which is what you all think you know will forge a temporary agreement with your best guess at how long you can afford to wait before you have to make another decision.

When people talk about accuracy or super forecasting, that underneath usually gets lost. We obsess on the apple strike. How close was it? Accuracy is cool, and it's very much a real skill in the world; some people have it for real, like those secretive hedge funds you only read about. Hitting the apple is hard, but the other part of that is knowing when to try to hit the apple a second time, and a third time. It's exceedingly difficult to know how firmly to hold any view while reality is still throwing dice, and each throw produces evidence against and in favor of your decision to release the arrow.

It's about the art of holding. Hold too loosely and nothing can coordinate around you. Hold too tightly and the world changes while you're busy defending the old picture.

Most of the practice is in learning how to peacefully live in that moment between releasing the arrow and the bullseye you can just see. It's there, that bullseye, somewhere between Six and Seven.