Bet-the-company: the ones that failed sized capacity to a signal that was not a purchase
Ford built for 150,000 electric pickups against 200,000 reservations and sold 24,165. Twenty-four bets on record, and the difference is not conviction.
Holding is not winning here either
Twenty-four bet-the-company decisions are on record. Fifteen held, three were reversed and five held in part. As everywhere in this corpus, held means the company never changed its mind, and some of the most expensive entries in this set are held.
Beyond Meat scaled into peak demand and stayed committed while revenue and the stock collapsed and losses topped a billion. Boeing's decision to re-engine the 737 rather than build a new aircraft held, and the record's verdict lists what followed from it: crashes, grounding, halted production. Meta's metaverse investment continued and losses grew every year through 2025 while the core business kept funding it.
None of those was reversed. All of them were bets that the company was unable to stop.
The reversals have a measurement problem, not a nerve problem
Ford decided in January 2022 how large a factory to build for the electric F-150. It built for 150,000 units, because nearly 200,000 reservations looked like multi-year demand at a $40,000 price it believed it could deliver. It sold 24,165 in 2023, while holding 72% of the electric-pickup segment in the fourth quarter. The segment was real and small; the reservations were not orders.
Apple priced Vision Pro in June 2023 as a developer-platform filter rather than a consumer device, on the reasoning that a new platform is worthless empty and a high price keeps out casual buyers while letting in committed developers. Within two weeks of launch some of Apple's most dedicated customers were returning the device, and fewer than 100,000 units had sold in the US by mid-2024.
Both sized an irreversible commitment against a signal that had not been paid for.
What the held ones committed to
Nvidia built CUDA ahead of demand from 2006 and the bet stood until it became the standard for AI. Tesla sold premium variants first and the $35,000 car arrived after the ramp, exactly as sequenced. Eli Lilly filed tirzepatide for obesity in 2022 and it became the world's best-selling drug.
The difference between these and Ford is not conviction, and it is not that they were right earlier. It is that Nvidia's spending scaled with adoption, Tesla's capacity was built behind revenue rather than ahead of it, and Lilly's commitment was a filing rather than a factory. Each kept the size of the irreversible part proportional to what had actually been demonstrated.
How to run this decision
Drawn from what separated the records that stuck from the ones that were undone. A step nobody in the record can be seen taking is not here.
- 01
Separate interest from purchase, in the model
Reservations, waitlists, pre-orders with refundable deposits and survey intent are interest. Ford had nearly 200,000 reservations and sold 24,165 units. Apply a conversion rate you can defend from something you have actually sold before, and if you have never sold anything like it, say that out loud.
- 02
Size the irreversible part to what has been demonstrated
The commitment can be large. The part that cannot be unwound in twelve months should be sized to proven demand, not to the plan. A factory is irreversible. A filing, a software investment and a production line that shares tooling are not.
- 03
Build capacity behind revenue where the physics allow it
Tesla sequenced premium variants first and the cheap car after the ramp. That ordering is available more often than it is used, and it converts a bet on demand into a bet on execution.
- 04
Decide who is allowed to say stop, and give them a number
The held-and-failed cases in this set are companies that could not stop: Beyond Meat, Meta, Boeing. Name the person, the metric and the date at the outset, while nobody's reputation is attached to the answer.
- 05
Ask what the bet costs if it is right but slow
Most of the failures here were not wrong about direction. Ford held 72% of the segment it bet on. Apple built a device people admired. Both were early by an amount their cost structure could not absorb, which is a different risk from being wrong and needs its own number.
- 06
Write the downside in the currency you would actually pay it in
Not a percentage of revenue. Plants, headcount, a dividend, a credit rating, the years the next strategy cannot be funded. AT&T's bet cost it half its dividend, and that is the unit the board would have understood at the start.
What it looks like when this is going wrong
- Capacity is sized against reservations, waitlists or expressions of interest.Ford F-150 Lightning, 2022.
- A high price is justified as a filter for the right kind of customer.Apple Vision Pro, 2023.
- Nobody has been named who is allowed to stop it.Beyond Meat, 2021. Meta, 2021.
- The irreversible commitment is larger than anything that has been demonstrated.Ford, 2022.
- The downside is expressed as a percentage rather than as the specific thing you would give up.The five partial verdicts on record.