Keeping a Decision Journal

Why this matters

Six months after a hire, a truck purchase or a price change, you cannot remember what you expected it to do. Memory edits the expectation to match the result, so a call that half-worked feels like the plan all along, and a call that failed feels like bad luck. You get the outcome but never the lesson, which means the next big decision gets made with the same blind spots. A decision journal is the fix: a short written record made BEFORE you know the result, read back when you do. The value is entirely in the gap between the two, and that gap only exists if the first half was written down.

What an entry looks like

Every field below exists because skipping it breaks the review later. Here is a real-shaped entry from a four-person HVAC shop, then the steps that build it.

DECISION #14          logged 2026-03-09
Hire a full-time install helper, start by Apr 1
OPTIONS  (a) hire helper
         (b) sub out one install a week
         (c) keep declining installs past 3 a week
CHOSE    (a)
EXPECT   by Jul 1: installs 3 -> 5 a week
         owner field hours 46 -> 38 a week
SURE     70%
CHANGE MY MIND IF
         fewer than 4 installs a week booked in May
         helper cannot run a changeout solo by June
STATE    rested, good week. Office manager unsure.
REVIEW   2026-07-01

It took about six minutes. The steps below are the fields in the order you fill them.

Step 1: Decide what earns an entry

Log a decision when it is hard to undo or expensive to get wrong: a hire, a firing, a vehicle, a lease, a new service, a price change, dropping a big account. A useful working line is anything costing more than about two weeks of payroll to reverse, or anything you will still be living with in a year. See related: Reversible and Irreversible Decisions in a Small Shop.

Why: the journal dies when it becomes a chore. Two to four entries a month is sustainable for years; fifteen is abandoned by May. What breaks if you skip the filter: you log the easy, frequent calls (which supplier for filters) because they are easy to write, and the three decisions that actually shaped your year never get an entry.

Step 2: Write the options, including the ones you rejected

List at least two alternatives to the one you chose, and include "do nothing" if it was real. In entry #14, option (c) was the status quo: keep turning down a fourth install each week.

Why: the review later asks "was this the best available move," not just "did it work." You cannot answer that without knowing what else was on the table. What breaks: a decision that worked passably gets scored as good, when a cheaper option you dismissed in ten seconds would have done more. The rejected options are also where you find your habits: if subcontracting appears as option (b) in nine entries and is chosen in none, that is a pattern worth seeing.

Step 3: Write the prediction as a number with a date

"The helper will free me up" cannot be scored. "Installs go from 3 to 5 a week by July 1, and my field hours drop from 46 to 38 a week" can. Name the quantity, the unit, the level you expect, and the date by which it should show.

Why: a vague prediction can be read as correct whatever happens, which makes the review useless. A numeric one can be plainly wrong, which is the point. What breaks: you review in July, find 4 installs a week, and decide that was "basically what I meant." The number protects you from that.

Keep it to one or two predictions per entry. The first one is the headline that gets scored true or false later. If you cannot think of a number, the decision is probably not yet clear enough to make.

Step 4: Put a percentage on how sure you are

Write how likely you think the headline prediction is to come true: 50%, 70%, 90%. Not "pretty sure."

Why: over a year the percentages become your calibration record, which is the most useful thing the journal produces (see the year-end pass below). What breaks: without a number you can never learn whether your "I'm certain" means 90% or 60%. Most owners discover theirs means something closer to the second. See related: Overconfidence and Calibrating Yourself.

Step 5: Write what would change your mind

Name the early evidence that would tell you the decision is going wrong, specific enough to recognize when it arrives. In #14: fewer than 4 installs a week booked in May, or a helper who cannot run a straight changeout alone by June.

Why: once money is spent, everyone reads new evidence in the decision's favor. Writing the warning signs while you are still neutral is the only reliable way to notice them later. What breaks: the warning arrives, you explain it away, and the review months later shows you had the information in May and acted on it in October. This field also feeds the kill-criteria habit for larger commitments. See related: Setting Kill Criteria Before You Start Something New.

Step 6: Note your state and who disagreed

One line: tired, angry, rushed, flush after a good week, scared after a slow one. Then whoever argued against it.

Why: the journal is the only place where state and outcome ever get lined up against each other. After twenty entries you may find the decisions made in a cash squeeze or after a bad callback week went worse. What breaks: without the line, the pattern is invisible, and the dissent you overruled is forgotten exactly when it turns out to have been right. See related: Making Big Decisions When You Are Exhausted or Angry.

Step 7: Set the review date before you close the entry

Pick the date by which the prediction should be visible, and put it on the calendar as an appointment, not a someday. Thirty days for a price test, 90 to 120 for a hire, six to twelve months for a vehicle or a new service.

Why: an unreviewed journal is a diary. What breaks: the review never happens, and the entry you wrote so carefully teaches nothing.

The review: reading the gap

On July 1 the owner opened #14 and wrote three lines underneath it:

• Actual: installs averaged 4.2 a week through June (the headline 5 was missed). Owner field hours averaged 41, not 38.
• Change-my-mind check: May bookings averaged 4.0 a week, right on the line, not under it. The helper ran his first solo changeout June 18, inside the window.
• Why the gap: the helper was fine. Installs were capped by how many replacement quotes the owner could run, about 9 a week, at a close rate that turns 9 quotes into about 4 installs. The helper added crew capacity to a shop whose limit was sales calls.

That last line is the lesson, and it is a different lesson from "the hire worked" or "the hire failed." The decision was reasonable, the execution was fine, and the prediction was wrong because the owner located the bottleneck in the wrong place. The next entry, #19, was about getting a second person running quotes.

When you read a gap, sort it into one of these:

• The world surprised you. A heat wave, a supplier shortage, a competitor closing. Note it; do not adjust much.
• You had the information and misread it. Like #14: the quote numbers were in the job software the whole time.
• The decision was right and the execution failed. The plan needed a step nobody took.
• The prediction was right for the wrong reason. Installs hit 5 because a competitor closed, not because of the helper. Treat this one as a miss; you would not get lucky the same way twice.

Resist scoring the decision purely by how it turned out; that judgment belongs to a sibling. See related: Outcome Bias: Judging a Decision by How It Turned Out.

The year-end pass: calibration

Once a year, put every scored entry into a small table by the confidence you wrote. This shop's first full year had 26 entries:

Read it row by row. When this owner wrote 80% or higher, the prediction came true 63% of the time, and when he wrote 60 to 79% it came true 67% of the time. His most confident calls did no better than his moderate ones, so in practice his "80%" carried no extra information. That is the most common finding in a first year, and it has a direct use: the next time he feels 90% sure about a hire or a vehicle, he knows his track record at that feeling is closer to two in three, and he leans harder on the change-my-mind field.

Two cautions on reading the table. Eight entries in a row is a small count, so one more hit or miss moves it by about 12 points; treat a single year as a hint and two years as a pattern. And score only the headline prediction you wrote at the time; adding a second prediction afterwards because it came true is the journal lying to you.

Then read the misses together. In this shop, 4 of the 10 misses involved the owner's own capacity (his hours, his sales time), which said more about the business than any single review did.

What quietly kills the habit

These are not covered by the steps above, and each one has ended a journal that was otherwise working:

• Entries written after the fact. You decide on Tuesday and log on Friday, once the first week's bookings are in. The prediction is already contaminated. If you could not log it at the time, mark the entry "late" and leave it out of the calibration table.
• The journal becomes a place to vent. Long entries about the customer who wronged you, with no prediction and no date. Keep feelings to the one state line.
• Reviewing only the failures. You open the journal when something goes wrong, so the successes never get checked, and you never learn which wins were luck. Review every entry on its date.
• Keeping it somewhere you do not see. A notebook in the truck is lost in a year. A recurring calendar block on the first Monday of the month, with the journal file linked, is what keeps it alive.

References

• Daniel Kahneman, Thinking, Fast and Slow (2011), on hindsight and the reconstruction of past beliefs
• Philip Tetlock and Dan Gardner, Superforecasting (2015), on scoring numeric predictions and calibration
• See related: Overconfidence and Calibrating Yourself, Outcome Bias: Judging a Decision by How It Turned Out
• See related: Reversible and Irreversible Decisions in a Small Shop, Setting Kill Criteria Before You Start Something New