Four kinds of surprise

The one I couldn't predict, and didn't need to

A port strike, a couple of years back. We had material sitting in containers we couldn't touch. Several lines were exposed. For about seventy-two hours it looked like it was going to be genuinely bad.

It wasn't. The dispute resolved in days, the boxes moved, and we never went line-down because we were carrying safety stock on the constrained items. Unglamorous, expensive to hold, and it did exactly the job it existed to do.

I've since heard people describe that strike as unpredictable. That's half right, and the wrong half. The labor contract expiration was public and had been known for years. Anyone could have circled the date. What nobody could know was the outcome: three days or three months, and the difference between those two is the difference between a nuisance and a quarter.

So prediction was never the tool. The date was already knowable. The question was how much buffer to carry against an event with a known date and an unknowable length, and that's an inventory decision, not a forecasting one.

My honest opinion, and people disagree with me on this: the industry has over-rotated on prediction and under-invested in buffer. Buffer doesn't demo well. Nobody builds a dashboard for it. It just quietly works on the days you needed it to.

The one where I was watching the wrong list

Forced-labor enforcement is the risk I've changed my thinking on most, and I'd rather explain the mechanism than any particular incident.

Most teams screen their suppliers against the UFLPA Entity List. It feels like diligence. It isn't, or not the useful kind, because by the time a company appears on that list the decision has already been made and your shipment is already sitting at the border. You're reading a record of the past.

The leading indicator sits one step earlier. Before entities get named, sectors get designated. Lithium was designated a high-priority enforcement sector in August 2025. Battery-materials entities were added to the Entity List in July 2026. That's the better part of a year of public warning, sitting in the same government publications everyone already receives and nobody reads until it's too late.

Which means the screen most people run is aimed at the wrong field. Don't ask "is this supplier listed." Ask whether this supplier's material and region sit inside a designated sector, and how far down your tiers that exposure runs. Same public data. Ten months earlier.

The one we did to ourselves

This is the one I find genuinely useful, and it's the least discussed.

Spring 2020. Automotive demand looked like it was falling off a cliff, and nobody knew how far down the bottom was. So automakers cut semiconductor orders. Given what anyone knew that month, it was a defensible call. I'd probably have made it.

Foundries don't hold empty capacity waiting for you to come back. Those slots went to consumer electronics, which was booming while everyone sat at home buying laptops. Then automotive demand returned far faster than any of us forecast, and the capacity was simply gone. Lead times on mature-node parts stretched past forty weeks. Assembly plants idled over components worth a couple of dollars each.

Here's the part I keep coming back to.

The earliest indicator of that shortage wasn't a market signal or a supplier warning. It was our own cancellation notices, sitting in our own systems, months before anything hurt.

Nobody joined the two thoughts. We released a capacity slot, and someone else took it, and we were not getting it back on the schedule our plan assumed. Each cancellation looked like a local decision made for sound local reasons. In aggregate, in a capacity-constrained market, it was a transfer of our position to somebody else.

I don't think this is rare at all. Demand plans get trimmed. Orders get pushed a quarter to help the cash position. A supplier gets dropped over price. Every one of those is reasonable in isolation and every one of them is invisible as a risk signal, because we don't classify our own decisions as market events. They are.

If I were standing up a watch list tomorrow, the first item on it wouldn't be anything external. It would be our own recent order reductions and supplier changes on anything sole-sourced or capacity-constrained, reviewed with one question attached: who is holding that slot now, and what does it take to get it back?

The one where the material arrived and the order didn't

The three above are all supply-side. Something I needed showed up late, or not at all, or turned out to be impossible to import. This one runs the other way, and it cost more than any of them.

We were supporting a major carrier on data center equipment. The forecast was large and it was credible. The relationship was good, the program had been in discussion for months, and everything about it looked solid. We committed several million dollars of material against it and built for months.

The order never came. The carrier made a marketing decision further upstream that changed what their data centers actually needed, and the additional storage capacity wasn't required any more. Nobody did anything wrong. Their business changed. The news reached us after the material was already on the floor.

No line went down. What I had instead was several million dollars of inventory looking for a home, which in this industry has a name and an accounting treatment. It's a quieter failure than a stopped line and it lands in a different part of the P&L, but a write-off spends the same as a delay.

Here's what I've concluded since. There was a signal, and it wasn't in any supply chain system.

A carrier that is about to change its network strategy does not tell its equipment suppliers first. It tells investors. It shows up in capex guidance, at an analyst day, in the careful language on an earnings call, often months before it reaches the people issuing purchase orders. We were reading their forecast. We were not reading their investor communications, and those two documents were telling different stories.

The other half of it isn't a prediction problem at all. If you are committing your own capital against someone else's forecast, that exposure is contractual, and the fix is a liability agreement that puts the customer on the hook for material bought to their number. I've watched a lot of teams treat that as an awkward conversation to have with a good customer. It is considerably less awkward than the write-off conversation.

And one internal number nobody reviews often enough: what share of your committed material sits behind firm purchase orders versus behind a forecast. That ratio is knowable today, and it's the best single measure of how exposed you are to somebody else changing their mind.

Four categories, four responses

What connects the three I could have caught is not the type of event. It's where the signal was living.

None of them were in supply chain data. The forced-labor exposure was in a government sector designation. The chip shortage was in our own cancellation notices. The E&O was in a customer's investor communications. Meanwhile we were watching lead times, supplier scorecards, and on-time delivery, and every one of those said the situation was fine right up until it wasn't. That isn't a tooling problem. It's a problem of where the tools are pointed.

Notice too what happens to inventory across these four. In the port strike, safety stock is the only reason we didn't go down. In the carrier program, inventory was the loss. Same lever, opposite sign. Anyone selling you a single posture on buffer isn't paying attention.

So, four situations and four different responses. Some risks are visible in public data nobody in your function reads, and the move is to watch the designation rather than the list. Some have a knowable date and an unknowable outcome, where you size a buffer and stop pretending forecasting helps. Some begin with a decision you made yourself, which means reading your own change log as a market signal. And some begin with a decision your customer made, where you read their investor deck and you get the liability agreement signed.

The skill isn't forecasting. It's classification, and then knowing which of the four you're actually looking at. Get that wrong and you'll carry inventory against something you could have seen coming, and build dashboards for something you never could.