Decision log for small businesses: what to record and when to review

Most small businesses make dozens of consequential decisions a year: a supplier is changed, a price goes up, a part-time role is created, a software subscription is cancelled. A few weeks later the reasoning behind them has usually faded. Why was that supplier chosen? What was expected to happen? Was the result better or worse than hoped? Without a written trace, these questions get answered from memory, and memory tends to rewrite the story to fit the outcome.
A decision log is a simple remedy. It is a running list of individual decisions, each with the reasoning at the time and a date on which it will be looked at again. It differs from a set of standing rules, which settle recurring situations in advance. The log deals with one-off choices, and its main value lies in the second look.
Why a written trace helps
Software teams have long kept records of this kind, known as architecture decision records. Microsoft's guidance for cloud architects, a vendor source and written for software work, makes a point that carries over to any small organisation: "A decision that's made but never recorded will likely be forgotten" [1]. The same guidance recommends recording the context and rationale, because without them people cannot judge whether a decision still applies once circumstances change [1].
There is also a human reason. People tend to be too optimistic about their own plans. The UK Treasury's Green Book, the guidance for appraising public spending, defines optimism bias as "the proven tendency for appraisals to be over-optimistic about key assumptions", and advises practitioners to adjust for it [2]. The setting is public investment rather than a corner shop, but the pattern is familiar to anyone who has watched a launch date slip. A written expectation makes the gap between plan and result visible later, rather than leaving it to be forgotten.
Finally, knowing about biases does not remove them. In their article on big decisions, Kahneman, Lovallo and Sibony wrote that "awareness of the effects of biases has done little to improve the quality of business decisions", and suggested examining how a recommendation was produced as well as what it says [3]. A log is one lightweight way to make the process visible.
What deserves an entry
Recording everything would turn the log into a chore that gets abandoned. A short test helps decide what goes in. A decision is a candidate when at least one of these applies:
- It is costly to undo: a contract longer than a few months, a hire, a change of premises or a large purchase.
- It changes how customers are treated: prices, opening hours, delivery terms, a new channel.
- It rests on a real uncertainty: a guess about demand, a supplier's reliability or how long a task will take.
- It was close: the team was split, or the owner changed their mind more than once.
- It sets a precedent that others may later point to.
Routine choices that are cheap and easy to reverse can stay out. For many small teams, three to six entries a month is a workable volume.
The fields of an entry
An entry should take about five minutes. Seven fields are usually enough:
Microsoft's template guidance also includes a confidence level and a status, noting that a low-confidence decision "could prove useful for future reconsideration" [1]. Both are worth borrowing. Status can be as simple as three words: open, confirmed, replaced.
The expectation field does most of the work. A sentence such as "this should improve things" cannot be checked; "payout within two working days" can. When a number feels forced, a plain observable statement is a reasonable substitute, for instance "no customer complaints about the new opening hours in the first six weeks".
Three entries, briefly
The examples below are invented for illustration.
A part-time bookkeeper. Decision: hire eight hours a week instead of continuing with the owner's Sunday evenings. Expectation: month-end close by the fifth working day, owner's weekend time freed. Confidence: medium. Review: after the third month-end.
A shorter Saturday. Decision: close at 14:00 rather than 17:00. Expectation: fewer than one in five weekly sales lost to the later hours, based on the last three months' till data. Confidence: low, because summer figures were used. Review: eight weeks, against the till report.
A price rise. Decision: raise the standard service fee by 6%. Expectation: no more than a handful of cancellations in the first quarter. Confidence: medium. Review: end of the quarter, comparing cancellations and new enquiries with the same period a year earlier.
When to review
The review date is set when the entry is written. Three types of trigger keep the log from becoming a graveyard:
- A fixed date. Chosen according to how quickly the result shows up: weeks for opening hours or a price change, a quarter or more for a hire or a supplier switch.
- A signal. A condition that brings the review forward, such as two missed deliveries in a month or a cost that passes a stated figure.
- A change in circumstances. A new competitor, a rise in a main input cost or a staff departure can undermine the reasoning even when the results look fine.
A monthly half-hour slot can cover everything: open the log, filter for entries whose review date has passed, and work through them one at a time.
Running a review
A review has four questions, taken in order:
- What happened? Compare the result with the expectation, using the signal named in the entry rather than impressions.
- What was known at the time? Re-read the original reasoning before judging it. A decision can be sound and still turn out badly, and the reverse.
- What would be done differently with today's information? This separates a lesson about the decision from a lesson about the forecast.
- What now? Mark the entry as confirmed, adjust it, or replace it.
The Green Book makes a similar point at a larger scale: evaluation "involves assessing an intervention after it has been approved and implemented", and should be considered from the outset rather than added at the end [2]. In a small business, deciding the review date on the day of the decision is the practical equivalent.
When a decision is replaced, Microsoft's guidance suggests leaving the original in place and writing a new record that supersedes it, linking the two [1]. This keeps the history honest. It also allows a later reader to see that a change of course was a response to evidence rather than an accident.
Common problems
- Writing the entry after the result is known. Entries made in hindsight tend to match the outcome. Writing before or on the day of the decision is what gives the log its value.
- Vague expectations. If nothing could show the decision was wrong, it cannot be reviewed.
- Reviewing outcomes only. Good processes sometimes produce poor results because of chance. Reading the original reasoning guards against drawing the wrong conclusion.
- Turning the log into a blame record. If the review is used to assign fault, entries become cautious and less candid. A log works better as a learning tool.
- Too much detail. A paragraph is usually enough. Long documents belong elsewhere, with a link from the entry.
- No owner. A named person, not "the team", is responsible for reading the log on the review day.
Where to keep it
A shared spreadsheet is enough: one row per decision, with the seven fields as columns and a filter on the review date. A page in the team's existing notes tool works equally well. What matters is that everyone involved can find it, that it sits where the team already works, and that the review slot is in the calendar. Anything more elaborate can wait until the habit has held for a few months.
What the log cannot do
A log does not make decisions better by itself, and it does not guarantee that the same mistakes will not recur. It records what was thought and what was seen, and it creates a regular occasion for comparing the two. Over several months, patterns tend to show up: forecasts that run optimistic in one area, a type of supplier that disappoints, a kind of decision that deserves more time. Those patterns are the material from which better judgement can be built, at the cost of a few minutes per decision and half an hour a month.
Sources
Verified on 7 November 2026.
- Microsoft Learn, Azure Well-Architected Framework, Maintain an architecture decision record (ADR) (page updated 13 April 2026). Vendor source, written for software architecture and used here by analogy. https://learn.microsoft.com/en-us/azure/well-architected/architect-role/architecture-decision-record
- HM Treasury, The Green Book (2026 edition; page updated 28 October 2026), GOV.UK. https://www.gov.uk/government/publications/the-green-book--2/the-green-book-2026
- Kahneman, D., Lovallo, D. and Sibony, O., The Big Idea: Before You Make That Big Decision..., Harvard Business Review, June 2011. https://hbr.org/2011/06/the-big-idea-before-you-make-that-big-decision
About the author: The Prodability Editorial Team writes practical guides on organisation, processes and management for small businesses. Website: Prodability
Disclosure: the authors have no financial relationship with any organisation mentioned in this article.


