How to Keep a Decision Journal with AI

Sudsy Van is a one-person mobile car-detailing business run by Priya Okonkwo, turning over about 62,000 dollars a year from 340 recurring bookings. A second van and a second technician would cost roughly 41,000 dollars up front and 3,900 dollars a month to run, and Priya has been circling the decision for eleven weeks without writing a single number down.

A decision journal does not make the choice for you. It forces you to state the call, the assumptions behind it, and what you expect to happen, then revisit the entry months later when the outcome is known. This workflow uses ChatGPT to run that loop in four steps.

Why the Journal Has to Start Before the Decision

Most people journal after the fact, which is just storytelling. The value comes from writing your reasoning down while you are still uncertain, because that is the only version you cannot retrofit to match whatever happened.

  • The decision, in one sentence, dated.
  • The assumptions you are betting on.
  • The outcome you expect, with a number and a date.
  • A review date fixed in advance.

Four fields, every time. The review date is the one people skip, and it is the one that turns a journal into a feedback loop.

The field worth guarding most is the expected outcome. Owners tend to record only the upside, then feel cheated when the downside arrives. Write the success number and the failure number on the same day, in the same entry, and the eventual review will have something honest to grade your thinking against.

Priya’s eleven weeks of circling is normal. The reason she has not decided is that the choice has never been written down where she can see all of it at once. The van cost, the monthly runway, the booking growth she is hoping for, and the risk of splitting her attention between two crews all live in her head, where they change shape every time she thinks about them.

Step 1: Log the Decision and Its Assumptions

State the call plainly, then list what must be true for it to be the right one.

Act as a decision coach. I run a mobile car-detailing business. My decision is:
whether to buy a second van and hire a second technician this quarter.
Write a decision journal entry with these fields:
- Decision: one sentence, dated today.
- Type: reversible or hard to reverse.
- Options considered: at least three, including doing nothing.
- Assumptions: list every belief the decision rests on, each as a testable claim,
  for example "recurring bookings grow 15 percent once we can serve two suburbs
  a day".
- Information gaps: what I do not know yet and how I could find out cheaply.
Ask me for any missing number instead of inventing one.

What you get: a dated entry that separates the decision from the beliefs under it. Priya’s largest assumption turned out to be that a second van would not cannibalize her existing bookings, which she had never checked.

Push for testable claims rather than feelings. “I can handle the extra admin” is a feeling; “hiring a second technician keeps my own hours under 45 a week” is a claim the review can actually check. If an assumption cannot be checked, it is not an assumption, it is a hope.

Step 2: Turn the Expected Outcome into a Prediction

An expectation you cannot measure is a wish. Write it as a number you could be wrong about.

Convert each assumption from the entry above into a prediction.
For every prediction give:
- The metric, the baseline value, the target value, and the date I should check it.
- A confidence level from 0 to 100 percent.
- The cheapest signal that would tell me early that I am wrong.
Then, separately, write what a bad outcome would look like at 90 days and at
12 months.

What you get: a set of dated, checkable predictions. Instead of “the van will pay for itself”, you get “monthly revenue reaches 9,400 dollars by 1 March, confidence 60 percent”. That is a claim the review can actually grade.

Confidence levels are what make the entry useful months later. A prediction you were 90 percent sure about and got wrong teaches far more than one you were 55 percent sure about, because the surprise points straight at a blind spot you did not know you had.

Step 3: Set the Review Date and the Kill Criteria

Decide now, while you are calm, what would make you reverse course, so the exit does not depend on how you feel in month four.

For this decision, define:
- The review date and a mid-point check-in date.
- Three kill criteria: specific numbers that, if hit, mean I shut the second van
  down regardless of how I feel.
- The sunk-cost trap to watch for, stated plainly.
- A single question to ask myself at the review, worded so it can be answered
  yes or no.

What you get: a pre-committed exit. Priya’s kill criterion was blunt: if the second van has fewer than 40 recurring bookings by month three, it goes. Writing that down in advance is what stops a slow bleed from becoming a two-year one.

Set the criteria in numbers you would not argue with later. The entire point of deciding the exit while calm is that the version of you in month four will be tired, invested, and busy inventing reasons to keep going.

Step 4: Run the Review and Score Yourself

On the review date, the entry becomes a scorecard. Read your old reasoning, grade each prediction, and separate a bad decision from a bad outcome.

It is the review date. Here is my original entry and here is what actually happened:
<paste the entry and the real numbers>
Score each prediction as correct, partly correct, or wrong, and give an accuracy
percentage.
Then answer:
- Was the decision sound given what I knew, even if the outcome was poor?
- Which assumption was most wrong, and what should have warned me?
- What single habit would improve my next decision?
Then write the next entry's opening line for the follow-on decision.

What you get: an honest scorecard and a written lesson. The distinction between a poor decision and a poor outcome is the part most owners miss, and it is exactly what the journal is built to surface.

Score the entry honestly, including the parts that flatter nobody. The journal compounds only if you record the misses as carefully as the wins, because the misses are where the next decision gets its edge.

Decision Journal Deliverables

  • A dated entry with the decision, options, and testable assumptions.
  • A set of predictions with metrics, targets, dates, and confidence levels.
  • Pre-committed kill criteria and a named sunk-cost trap.
  • A fixed review date and a mid-point check-in.
  • A scored review separating decision quality from outcome quality.
  • The opening line of the next entry.

Pitfalls That Turn a Journal Into a Diary

Writing it after the decision. A retrospective entry is a story, not a prediction. Log the call while you are still unsure.

Vague expectations. “It should work out” cannot be graded. Force a number, a target, and a date.

Never setting a review date. Without it you never learn, and the same mistake repeats next year with a bigger number.

Judging the decision by the outcome. A sound call can fail on luck and a reckless one can win. Score the reasoning and the result separately.

One quieter pitfall is making the entry so long you stop keeping it. Two hundred words is plenty; the journal survives on being fast enough to write on a phone between jobs. Keep the four fields fixed, and the habit lasts. For a planning companion that turns a stated goal into dated checkpoints, see our goal-to-plan prompt, and for a weekly habit that keeps the review cadence alive, our weekly review workflow fits the same loop.

Sources & Notes

The four-field decision journal and its review loop are our own framing of standard decision-log practice, reviewed on 2026-10-12. Sudsy Van, Priya Okonkwo, and the revenue, cost, and booking figures are illustrative examples rather than a real business. Financial figures should be checked against your own accounts before acting. Language models change between versions, so re-check any prompt that references your specific numbers or review cadence.

声明:1.本站所有文章,如无特殊说明或标注,均为本站原创发布。本网站中【文本生成/文生图】使用了DeeSseek人工智能生成内容技术, 算法来源说明:本网站未自行开发、训练或部署深度合成算法,所使用的人工智能算法来自:DeepSeek、Chatgpt、Google Gemini。 2.为了网站加载速度优化,本站预览图片进行压缩处理,所以有些不是太清晰,并不影响AI提示词生图、AI设计等内容。3.任何个人或组织,在未征得本站同意时,禁止复制、盗用、采集、发布本站内容到任何网站、书籍等各类媒体平台。如若本站内容侵犯了原著者的合法权益,可联系我们进行处理。--GUUNN.COM