Forecasts Are Lies (But Useful Lies)

I've built a lot of forecasts in my life, but the first time I truly felt "forecast hubris" is seared into memory. I built what I considered at the time a "bulletproof" forecast at my kitchen table. In the middle of the pandemic. The variables: revenue, invoice-level billing models, expense, time, entropy, universal decay. It had it all, and I could zoom up and down, in and out. In a world flipped upside down, this felt right side up.

At that same kitchen table, the world's timelines started to stretch and the model was out of date and pretty much wrong within days, if it was ever "right." The certainty faded, then was punctured. I was just in the eye of the Covid hurricane, mistaking a bit of temporary calm for something much more permanent.

Years have since gone by, and I guess I'm not really interested anymore in: was the modeling wrong, or was the world wrong? The assumptions underneath that model were expiring faster than I could update them, and reality was setting in fast back then.

There is no real crystal ball that tells us exactly what might happen next. These things are structured stories that can tell us a few things about the future, or possible futures, some built on considerably better evidence than others. And the stories are told, and we agree to believe them for a while.

In the absence of any real certainty, a shared story remains one of the fastest ways to move a group of people forward. Inevitably, the future will arrive looking somewhere between "spot on" and "nothing like the plan." But precision in the reported number matters...somewhat less than we all think it does.

Precision feels good and is needed in a lot of things. I want my airplane to be precise; I hope and pray it is once we're packed away like sardines in the sky. At 30,000 feet I want as many decimals as I can get.

On the ground, decimals bring something different compared to a round number. At sea level or thereabouts, I'm doubtful of decimals because I remember that kitchen-table forecast, full of decimals, and now recognize how much gravitas and fake precision they add to anything. You can glance at a churn probability of 14.327% and easily walk away with confidence you shouldn't have that the future looks pretty well understood.^[1] Or you can suggest a range of 14-25% based on history and two "hey, need a sanity check here" questions to a single manager. In my experience, the decimals receive fewer questions and do the work of persuasion that a methodology couldn't. Round numbers or ranges invite an argument, whereas decimals sort of imply that argument already happened somewhere else, and it was resolved.

The trouble with ranges is that eventually somebody still has to decide. Some headline number might land anywhere between $80 million and $100 million. It might even be the most intellectually honest answer anyone can truly give. Maybe you plan for eighty and add hires if the year starts looking more like a hundred. Fine. Somebody still has to decide when, exactly, "looking more like a hundred" is good enough to spend real money. Decisions, decisions, decisions. They have to leave the room.

Having done this for a while now, I understand that part of it a lot better, and I'm wary of false precision. I net out here: I just don't feel the same way when I see 14.327%. The crystal ball is really headlights. Useful and necessary, but you're still driving in the dark.