The cost of a reversal is decided by how long you defend it, not by whether you were wrong
Coca-Cola reversed New Coke in seventy-nine days and the brand was fine. AT&T took six years and $40 billion. Twenty-one reversals on record, and what the ones who sat through a trough had decided in advance.
Two reversals, the same admission, very different bills
Twenty-one reversal decisions are on record and fourteen were never undone a second time. This is the decision type about changing your mind, and the record prices it precisely.
Coca-Cola reformulated its ninety-nine-year-old flagship in April 1985 on blind taste tests where the sweeter formula beat both Pepsi and the original, after twenty years of share erosion. Hotline calls reached some 8,000 a day by June alongside roughly 40,000 letters, and the original came back as Coca-Cola Classic seventy-nine days after launch.
AT&T decided in October 2016 to buy its way into media rather than stay focused on connectivity, seeking to own the pipes and the content and escape the wireless price war. Net debt stood around $156 billion at the end of 2021. In April 2022 it took $40.4 billion in cash to deconsolidate WarnerMedia, and it exited DirecTV entirely.
Both companies concluded they had been wrong. One took seventy-nine days and kept its brand. The other took six years, a balance sheet and a dividend.
The ones that held sat through a trough they had priced
Adobe went subscription-only in 2013. The choice stood and subscription revenue overtook product revenue after a one-year trough. That trough was not a surprise, and the decision to sit through it was made before it arrived.
SAP set a hard end-of-support date in 2020; the deadline stood, cloud revenue rose and license revenue declined structurally, with many customers still on legacy. Salesforce cut 10% of staff in 2023 and margins vaulted past target years early. Microsoft embraced and hosted open source in 2018, and Linux now runs the majority of Azure customer cores.
The distinction that matters is not conviction. It is that each of these companies had decided in advance what a trough would look like, which meant the bad quarter when it came was evidence the plan was working rather than evidence it was not.
The middle case
Airbnb froze new Experiences submissions in April 2023 rather than abandon the product, to protect the core and cut mediocre supply that erodes trust. It removed roughly 5,000 listings for not meeting standards in May 2024, canceling forward reservations, and reopened host submissions that September after a pause of about seventeen months. Experiences relaunched larger in 2025 with a new Services layer.
That is a reversal that was designed as one. The freeze was reversible by construction, the thing being protected was named, and the reopening was a decision rather than a retreat.
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
Decide what a trough looks like before you launch, in numbers
Adobe knew subscription revenue would dip before it crossed. Write down the shape and depth of the expected bad period and the date it should end. Without that, every bad quarter is an argument about whether to continue, and the argument itself is expensive.
- 02
Separate the trough from the signal that you were wrong
They look identical for the first two quarters. The only way to tell them apart afterwards is to have described them differently beforehand, naming at least one measure that behaves one way in a trough and the other way in a failure.
- 03
Make the first version reversible on purpose
Airbnb froze intake rather than killing the product, which cost seventeen months and nothing else. Where the strategy allows a pause instead of an exit, take the pause, and say at the outset what would reopen it.
- 04
Put the reversal decision with someone who did not make the original
The people who committed cannot cheaply conclude they were wrong, and the record does not reward waiting for them to. Name the person and the date at the start, while nobody's reputation is attached to the answer.
- 05
Count the cost in what you will have to sell
AT&T's reversal was paid in WarnerMedia, DirecTV, a dividend and a balance sheet. When you commit, write down what you would have to give up to undo it, because that is the actual size of the bet and it is rarely the number in the paper.
- 06
Reverse in public and completely
Coca-Cola brought the original back under its own name and stopped defending the decision. Partial reversals in this record stay expensive, because the company keeps paying for both positions while the market keeps asking which one is real.
What it looks like when this is going wrong
- No one wrote down what a trough would look like before launch.The seven reversed and partial records on record.
- The people who made the decision are the only ones who can stop it.AT&T, 2016.
- The commitment cannot be undone without selling an asset.AT&T, 2016.
- A pause was available and an exit was chosen instead.The reversed records on record.
- Both positions are being funded while the market is told one is real.Southwest, 2025.