The decision you have to unmake

The revisit condition you wrote months ago has fired, and you are the person who argued for the thing. Why that makes you the least likely to act, and how to reverse without it reading as flip-flopping.

It is months later. The SSO work shipped, the deal closed, and two more followed it. You are now at forty-four connections on the hosted provider, and the monthly bill has stopped being a rounding error.

Somewhere in the decision record you wrote when you closed the review thread is a line you never expected anyone to read: revisit if we pass roughly forty connections.

The arithmetic is easy. That is not the hard part. The hard part is that you are the person who wrote the recommendation, argued for it in the thread, and told a VP it was the right call. Every lesson in this course so far has produced a commitment. This one asks you to break one, in public, having been its loudest advocate.

Why you are the wrong person to notice

The intuition is that the person closest to a decision is best placed to spot it going wrong. They have the most context, they see the bill, they know the tradeoffs. The research says the opposite, and it is worth being precise about how strongly.

In 1976 Barry Staw ran an experiment where people allocated funds to a business unit that had performed badly, varying whether the participant had been responsible for the original investment. The result: people committed the greatest amount of resources to a failing course of action when they were personally responsible for the negative consequences (Staw, “Knee-deep in the Big Muddy”). Not despite being responsible. Because of it. The usual explanation is self-justification: reversing course means conceding the first decision was wrong, and doubling down postpones that.

Where people get burned

Read the implication carefully, because it is uncomfortable and it is about you. The person with the most information about whether this decision has gone bad is structurally the least likely to act on it. Your expertise does not protect you here; your authorship is the risk factor.

Naming the trap is most of the defence. “I should push through, we are nearly there” feels like determination. It is also exactly what the effect predicts you would feel, which is reason enough to stop and check the numbers rather than trust the feeling.

The line you wrote is what frees you

This is why the ADR ends with a revisit condition, and why the review-thread lesson insisted on writing one even though it takes two minutes and feels like paperwork.

You set that threshold before you were invested. At the time you had no sunk cost, no public position to defend and no bill with your name on it. It is a message from a version of you with nothing to protect, and it converts the question from do I admit I was wrong? — which self-justification will fight — into has the condition I specified been met?, which is arithmetic.

You have now done this three times, and it is worth seeing as one technique rather than three tricks:

The assumption with an expiry date, so an unanswered question stops blocking you.
The trigger on a monitored risk, so you know when to escalate without relitigating it in the moment.
The revisit condition on a decision record, so you know when to reopen it.

Not every decision deserves reopening

A revisit condition firing means look again. It does not mean reverse. The question that decides it is how expensive the door is to walk back through.

A reversible decision — a vendor you can migrate off, a library you can swap — should be reversed quickly once the evidence turns, because the cost of being wrong for another quarter exceeds the cost of the switch. An irreversible one — a data model every customer now depends on, a public API, a migration you cannot undo — deserves a much higher bar, and if you find yourself facing one you did not think was irreversible, that is worth noting for next time: the bar for making it should have been higher too.

For the SSO decision, honestly: buying was a two-way door. That was one of its virtues, and it is a reason the original call was defensible even if you now switch.

Reversing without it reading as flip-flopping

The fear is reputational, and it is not irrational — people who change position constantly are hard to plan around. The difference between judgment and vacillation is entirely in how you present it.

Lead with the trigger, not the feeling. “We said we would revisit above forty connections; we are at forty-four” is a different opening from “I have been thinking, and I am not sure we made the right call.” The first is a plan executing as written. The second invites everyone to relitigate the original decision.

Separate what changed from what you got wrong. These are different admissions and they carry different weight. “We now have four times the connections we modelled” is the world moving. “I underestimated how fast enterprise deals would land” is an error. Both are fine to say. Conflating them makes a changed circumstance sound like a personal failure, or worse, dresses a real misjudgement up as bad luck.

Show both costs. What switching costs, and what staying costs. A reversal argued only on the cost of staying looks like enthusiasm for a new idea.

And the one people get wrong: do not apologise for a decision that was correct on the information available.

Judging a decision by how it turned out rather than by how it was made is a specific error, and it has a name. Annie Duke, who played poker professionally before writing about decisions, calls it resulting: the habit of reading a single outcome backwards into a verdict on the choice. A good decision can have a bad outcome, because the world contains variance. Buying at the time bought a quarter, unblocked a $400K deal, and left a two-way door open. That was right, and it remains right even though the connection count went somewhere you did not expect.

Apologising for it anyway is not modesty; it is expensive. It teaches everyone watching that decisions are judged on outcomes they could not control, and the rational response to that lesson is to stop making decisions that could look bad — which is exactly the conservatism you do not want from your team.

Retrieval check

Without scrolling up: according to Staw, which people commit the most resources to a course of action that is already failing — and why does that make the revisit condition, rather than your own judgment, the thing you should trust?

Check your answer

The people who were personally responsible for the original decision and its negative consequences. Responsibility increases escalation rather than reducing it, through self-justification — so being the expert and the author makes you more likely to double down, not less.

Which is why the revisit condition matters: you wrote it before you had anything invested, so it is the judgment of the only version of you that was disinterested. Trusting it turns do I admit I was wrong into has the condition been met, and the second question is one your self-justification has no grip on.

Now apply it: write the reversal

Hands on

Unmake the buy decision

Done when: You have a reversal memo that leads with the trigger, separates changed circumstances from misjudgements, shows both costs, and does not apologise for the original call.

Open with the trigger, quoted from the decision record: what you said you would revisit, and the number that says it has fired. Two sentences.
Split the reasons into two explicit lists: what changed in the world and what you got wrong. Be honest about the second list — if it is empty, say so and mean it; if it is not, put the item in it rather than smuggling it into the first list.
Give both costs. What building it now costs in weeks and risk, and what another year of per-connection fees costs. Use the estimate habits — a range with its top branch named, not a single number.
State the recommendation and — since you are about to make a new decision — give it its own revisit condition. You are not exempt from the thing that just saved you.
Now delete every apology for the original decision. Then read it again and delete the ones you did not notice were apologies: the “admittedly”, the “in hindsight I should have”, the “my mistake was” attached to something that was not a mistake.
Bring it into the chat. I’ll read it as the VP who approved the original plan, which is the reader whose reaction you are actually worried about — and I’ll flag anything that reads as flip-flopping rather than as a plan doing what it said it would do.

That is the whole loop

A forwarded Slack message became a framed problem, then a load-bearing question, then a decision with its alternatives shown, then a review that stress-tested it, a range you could defend, a pre-mortem that retired a risk, an update that told someone the truth early — and finally a document you wrote at the start that gave you permission to change your mind at the end.

That last part is the argument for all of it. The artifacts are not ceremony. Each one is a message to a later, more pressured, more invested version of you, written while you could still think clearly.

That is the SSO project, start to finish. One lesson remains, and it is a different scenario rather than a continuation: what to do when you arrive on a project that already has a plan and it is not yours — the same sequence, run backwards.

Either way, the next thing worth doing is running this on something real. Bring an ask from your own work — badly formed, forwarded, with no spec — and we will start again at the one-pager. It goes faster the second time and it is considerably less comfortable, which is how you will know it is working.

The decision you have to unmake

Somewhere in the decision record you wrote when you closed the review thread is a line you never expected anyone to read: revisit if we pass roughly forty connections.

Why you are the wrong person to notice

Where people get burned

The line you wrote is what frees you

This is why the ADR ends with a revisit condition, and why the review-thread lesson insisted on writing one even though it takes two minutes and feels like paperwork.

You have now done this three times, and it is worth seeing as one technique rather than three tricks:

The assumption with an expiry date, so an unanswered question stops blocking you.
The trigger on a monitored risk, so you know when to escalate without relitigating it in the moment.
The revisit condition on a decision record, so you know when to reopen it.

Not every decision deserves reopening

A revisit condition firing means look again. It does not mean reverse. The question that decides it is how expensive the door is to walk back through.

For the SSO decision, honestly: buying was a two-way door. That was one of its virtues, and it is a reason the original call was defensible even if you now switch.

Reversing without it reading as flip-flopping

Show both costs. What switching costs, and what staying costs. A reversal argued only on the cost of staying looks like enthusiasm for a new idea.

And the one people get wrong: do not apologise for a decision that was correct on the information available.

Retrieval check

Check your answer

Now apply it: write the reversal

Hands on

Unmake the buy decision

Done when: You have a reversal memo that leads with the trigger, separates changed circumstances from misjudgements, shows both costs, and does not apologise for the original call.

Open with the trigger, quoted from the decision record: what you said you would revisit, and the number that says it has fired. Two sentences.
Split the reasons into two explicit lists: what changed in the world and what you got wrong. Be honest about the second list — if it is empty, say so and mean it; if it is not, put the item in it rather than smuggling it into the first list.
Give both costs. What building it now costs in weeks and risk, and what another year of per-connection fees costs. Use the estimate habits — a range with its top branch named, not a single number.
State the recommendation and — since you are about to make a new decision — give it its own revisit condition. You are not exempt from the thing that just saved you.
Now delete every apology for the original decision. Then read it again and delete the ones you did not notice were apologies: the “admittedly”, the “in hindsight I should have”, the “my mistake was” attached to something that was not a mistake.
Bring it into the chat. I’ll read it as the VP who approved the original plan, which is the reader whose reaction you are actually worried about — and I’ll flag anything that reads as flip-flopping rather than as a plan doing what it said it would do.