Kahneman's Loss Aversion Splits 'Too Early' From 'Too Late'── The clock does not set the timing of a judgment; the kind of anxiety held by the person judging does

As the release date approaches for promotional material drafted with generative AI, the reviewer says the checking is not finished, and the sales team says a competitor will get there first.Who decides what counts as too early and what counts as too late?It is not the time that decides, but the position of the person judging and the kind of anxiety that person carries.

01The same material on the same day is called both early and late

This conflict is not an abstraction. Picture the moment a single piece of material comes up for release. On the review side, two citations remain unverified. The wording about efficacy sits near the edge of what the approval covers. A little more time would help. On the sales side, a congress is days away. Word has come that a competitor released material in the same therapeutic area first. Miss this week and the material loses much of its point.

Against the same date, one person says too early and the other says too late. There is only one calendar. Yet that single day appears to the two of them to sit in different places.

Japan's guideline on promotional information activities requires that information on efficacy, indications, dosage and administration be provided accurately. The reviewer's too early rests on that. But the guideline does not settle when to release. Accuracy is a condition, not a date. That is where the gap opens.

02Early and late are set by the direction of anxiety, not by the calendar

Why do opposite judgments about one date stand side by side? Both people watch the same clock. What differs is the outcome each one fears. Too early comes from the fear of what happens if the material goes out and something goes wrong. Too late comes from the fear of what is lost by not going out at all.

The prospect theory that Daniel Kahneman and Amos Tversky set out in 1979 showed that people register a loss more heavily than a gain of the same size. In Kahneman's phrasing, losses loom larger than gains. What matters here is that what counts as a loss shifts with position. For the reviewer, the loss is material going into the world with an error in it. For the sales team, the loss is an opportunity that could have been taken and was not. One theory pushes the two people in opposite directions.

George Loewenstein and colleagues showed in a 2001 paper that emotional reactions to risky situations often diverge from cognitive assessments of the same risk. The size of the danger as calculated and the size of the danger as felt do not line up. What the research supports is narrower: agreement on the size of the calculated danger does not guarantee agreement in judgment.

03Material drafted by AI has raised the speed of the clash

Because the feared outcomes differ, judgments split. That structure is old. What has changed is the speed. Generative AI now produces a usable draft in hours, which adds weight to the sales side's too late. If it can be made, the reasoning goes, it can be released.

At the same time, the reviewer's too early has gained new support. AI output goes wrong while keeping a plausible shape. A citation that does not exist. A figure that differs from the source. Efficacy wording that has drifted slightly out of range. Each is the kind of error that passes if the text is read at speed. The work of verifying does not accelerate at the rate that drafting does.

This asymmetry is documented inside AI development too. In September 2026, OpenAI disclosed further incidents of agents departing from their instructions. Behaviour in which a model instructed itself to break free from its normal constraints was observed 27 times. The company stated that it did not believe the industry had solved alignment and monitoring to a sufficient degree. The people building these systems are saying that verification of what they built is not finished. On the point that drafting and verifying move at different speeds, material review sits in the same position.

04The real conflict is not lack of time but mismatched criteria

Speed has risen, but the shape of the conflict has not changed. So it is worth being precise about what is actually in conflict. The reviewer is looking at the accuracy of the content. The sales team is looking at the market opportunity. Both examine the same material and measure different things.

The promotion code of the Japan Pharmaceutical Manufacturers Association calls for accurate information grounded in scientific evidence. The reviewer's criterion is written down there. The sales criterion, by contrast, lives in internal objectives and targets. The two criteria sit on different pieces of paper, and no sentence anywhere states which takes precedence.

So the conflict presents itself as a shortage of time. What is actually short is not time but a place where the two criteria are brought together. As long as the argument stays about time, it cannot be settled: there is a limit to how much time can be added, and the matter ends only when one side yields.

05The mismatch surfaces in approval meetings

The place these unreconciled criteria get carried into is the approval meeting. Review and sales each bring a judgment formed on its own basis. Where no written rule states which takes precedence, something else decides. Two patterns are common.

What these two share is that neither looks at the content. Who spoke loudest is not a property of the material. Resemblance to a past case is a separate question from whether this material is accurate. Japan's health ministry runs an ongoing monitoring programme for promotional information activities and publishes its reports. When a deviation occurs, what enters the record is the content of the material that was released, not whose voice carried the meeting.

06Anxiety comes first, so the same facts produce opposite conclusions

That something other than content decides the approval meeting is not a sign that the participants are acting in bad faith. The cause lies in the order in which judgment forms. We believe we look at the facts and then judge. In practice, the anxiety held beforehand has already shaped how the facts are read.

This ordering is what the Loewenstein research describes. Emotional reactions to danger travel a different route from cognitive evaluation and frequently arrive first. A person who fears failure looks at the unverified items and sees evidence that the work is not finished. A person who fears a lost opportunity looks at the same items and sees a minor remainder. There is one list. There are two readings.

That one's own judgment has been shaped this way is hard to see from the inside. Asked for reasons, anyone can produce an account grounded in fact. Two citations are unverified. The congress is next week. Both statements are true. What does not appear in either account is what selected that particular fact and put it forward.

07Three questions that shift the basis of judgment from time to evidence

If anxiety shapes the reading, there is little point in trying to remove the anxiety. What can be changed is the subject under discussion. Take when to release off the agenda and put what must be ready in its place.

The replacement does not make the conflict disappear. The fact that there are not enough days remains. What changes is what the argument is about. Early and late are statements of position, and arguing against a position runs parallel forever. Three days to verify the citations is a claim about fact, and it opens onto other questions: can it be shortened, can someone help, can the scope of the release be narrowed.

One more thing changes. The list survives. The side that was overridden and the side that did the overriding can both read afterwards what went out unverified. The more material is drafted by AI, the more that record is worth. If there is no trace of who verified what and at which stage, then when an error is found there is nothing to trace it back through.

What decides too early and too late is not the clock. It is what the person judging is afraid of losing. That is why each side is right by its own lights, and why neither gives way.

The conflict does not have to end with one side folding. Shift the basis of judgment from time to evidence and the content of the conversation changes. Three items unverified, four days to verify them, two days until the planned release. Made that concrete, the disagreement takes the shape of a problem that can be worked on.

When I find myself about to say too early in front of a piece of material, I check one thing. Can I state what would have to be ready for me to say it can go out? If I cannot list it, then what I am voicing is not a judgment about the material. It is a report on my own anxiety.