What this template is for
Use the log for every material marketing decision across every channel: paid search, paid social, programmatic, email, the website and pricing changes that marketing proposes. One log for the whole programme is better than one per channel, because effects cross channels and the review that reads them is usually a single meeting. The performance reporting guide describes the log in a paragraph: date, decision, evidence, expected effect and, at a later review, the actual effect. It sits at the end of the monthly report, and its latest entries are the first thing the next review reads. The paid search account management playbook describes the same record for one account and asks that every change goes in, including changes made by agencies, colleagues covering holidays and automated recommendations someone accepted.
The log sits between other documents rather than replacing them. A large move of money gets its own argument in a budget reallocation memo, and the memo's decision becomes one entry here. A controlled test gets its readout in an experiment results report, and the decision the report produces becomes an entry here. The monthly paid media review checklist opens with last month's decision log and ends with new logged decisions, so the log is both the meeting's input and its output.
It is not three other things. It is not the platform's change history: the change history audit explains, from Google Ads Help, that the account's record shows what changed, who changed it and when, but never why or what was expected. It is not a log of tracking and definition changes: those belong in the measurement change log, which this log reads but does not repeat. And it is not a test write-up. When a research study informs a decision, the guide to reporting research findings keeps the study's own record; the decision it leads to is logged here.
The template
Each field below becomes a column in a spreadsheet or a shared document, with one row per decision. Keep the field names, so that anyone reading an old entry knows what each column means.
| Field | What to write |
|---|---|
| Date, area and owner | When the decision was taken, the account, campaign or channel it touches and the person who will answer for it |
| Decision | What changed, written specifically enough that someone else could undo it, including changes made by agencies or accepted recommendations |
| Evidence | The report, test or observation behind the decision, its date range and where it can be found again |
| Reversibility | Whether the decision can be reversed cheaply or not, which sets how much evidence it needed |
| Expected effect | The metric expected to move, in which direction, and the trade-off accepted, written before the result is known |
| Guardrail | The level of a second metric that would reverse the decision early, set from that metric's own recent range |
| Check date | When the effect will be read, allowing for conversion lag and for the period the comparison needs |
| Outside events | Releases, tracking changes, promotions and other changes in the same period that could explain the result instead |
| Actual effect and verdict | What happened by the check date, read the same way the expectation was written, and whether the decision stands |
Write the first seven columns on the day of the decision. Leave the last two empty until the check date. An entry with an empty verdict past its check date is the clearest signal in the log: something was changed and nobody looked.
How to fill each section
Date, area and owner
Write the date the change went live, not the date it was discussed, because that is the date anyone comparing periods will need. Name the area precisely: the campaign, ad set, audience, page or email programme. The owner is the person who will read the result and answer for it, which is not always the person who clicked the button. The performance reporting guide sets the cadence of daily, weekly and monthly reviews; note which review the decision came from, so the entry can be traced back to the discussion.
Decision
Write the change so that a colleague could undo it without asking you. "Adjusted bids" fails that test; the paid search account management playbook gives the contrast between a vague entry and a useful one, which names the campaign, the change, the reason in a sentence, the metric expected to move and the review date. Give the old and new values where there are any. Log changes made by an agency and recommendations accepted from a platform in the same way, and say that is what they were.
Evidence
Point to the evidence rather than pasting it. Name the report or export, its date range and where it is saved, so the decision can be checked later against what was known at the time. The playbook asks that inputs are kept reproducible for exactly this reason. If the evidence was an investigation of a past shift, the change history audit ends with a finding that names the cause, the evidence for and against, the action and the date to check, and asks that it be added to the decision log and linked from the edit's entry.
Reversibility
Classify the decision as a two-way door or a one-way door using the one-way and two-way doors framework. Cheap, quickly reversed changes, such as most copy, creative, scheduling and bid decisions, need a clear expectation and a date to check. Costly or slow changes, such as leaving a channel, a large commitment or a rebrand, need research or a test first. Writing the classification down lets others challenge the call before the work starts, and a one-way door logged with thin evidence is easy to spot at review.
Expected effect
Write the metric you expect to move, the direction and the trade-off you accept, before the result is known. Most changes trade one thing for another: a higher return target usually means less volume, and more budget eventually meets Diminishing ReturnsEach extra unit of spend or effort buys less extra result than the one before it.. The testing and scaling guide explains why scaling decisions should be judged on the marginal figure rather than the average, so say which one the entry will be read on. Tie the expectation to the KPIA metric tied directly to an objective. the plan already uses rather than inventing a new measure for the entry.
Guardrail
A guardrailA metric an experiment must not worsen beyond an agreed bound, watched beside the primary metric that decides the test. is a second metric whose movement would reverse the decision before the check date. The testing and scaling guide asks for the stopping point, the signals that trigger a pause and the review point to be agreed before a scaling step. Set the guardrail's level from the metric's own recent range rather than a round number: the control charts framework shows how to draw that range from the metric's history and how to adjust for SeasonalityPredictable, repeating changes in demand tied to the calendar, such as months, holidays, weekdays or hours. first. A guardrail inside the metric's ordinary variation will fire on noise.
Check date
Choose the date the effect will be read, and allow for two things. Recent days are incomplete, because conversions arrive after the click and leads are qualified later; the guide to comparing periods explains marking recent days as provisional and comparing periods at equal age. The comparison also needs enough of a period to cover the business's own weekly or monthly pattern. Put the date in the review calendar as well as in the log, so the entry is read when it falls due.
Outside events
List the other changes in the same window that could explain the result instead: site releases, price changes, promotions, feed changes, stock problems and anything in the measurement change log, which records tag, consent and definition changes. The change history audit points out that the account's own record never sees changes outside the account, so this column is where they meet the decision. Fill it at the check date, not only on the day of the decision.
Actual effect and verdict
Read the result the same way the expectation was written: the same metric, the same comparison and the same BaselineAn estimate of what a metric would have done without a given change, campaign or promotion, against which the effect is measured.. The guide to explaining a change in a marketing metric gives the method when the result is not what was expected: doubt the change until data, definition and calendar are cleared, break the metric into parts and match the shift to an event. Then write a verdict in a few words: stands, reversed, or changed, with the follow-up as a new entry. If the effect could not be separated from an outside event, say so; that is a finding too.
Filled example
Below, the log is filled for an invented business. The figures, dates and results are illustrative.
Plessikar Lettings: the first quarter's decision log
Plessikar Lettings is an invented lettings agency that advertises to landlords on paid search and paid social and sends a monthly landlord email. Its marketing manager started a single decision log in January. Three of the quarter's entries are shown, one per column.
| Field | Entry A: paid search target | Entry B: paid social audience | Entry C: landlord email landing page |
|---|---|---|---|
| Date, area and owner | 8 January; paid search, landlord valuation campaign; marketing manager | 15 January; paid social, broad landlord interest audience; marketing manager | 3 February; landlord email, valuation link; marketing manager with the web agency |
| Decision | Target cost per valuation request lowered from £60 to £50 | Interest audience paused; budget of £600 a month moved to the existing lookalike audience | Email link moved from the generic contact page to a new valuation page with a short form |
| Evidence | December search report: the campaign met its target in each of the last six weeks with lost impression share low; export saved in the reporting folder | December paid social report: 41 leads from the interest audience, 2 instructions, against 9 instructions from the lookalike at similar spend; CRM export saved | November and December email reports: clicks steady, valuation requests from email low; form recordings show visitors leaving the contact form |
| Reversibility | Two-way door: target can be restored in a minute | Two-way door: audience can be resumed, though learning restarts | Two-way door: link can be pointed back; the page stays live for search |
| Expected effect | Cost per valuation request falls; accept around 10% fewer requests | Instructions from paid social rise; accept fewer leads | Valuation requests from email rise; clicks unchanged |
| Guardrail | Valuation requests below 70 a month, the bottom of the last six months' range | Paid social instructions below the lowest month of the last six | Email unsubscribes above the highest of the last six sends |
| Check date | 19 February, allowing two weeks for requests to be qualified | 26 February, allowing for the agency's sales cycle from lead to instruction | 10 March, after two sends |
| Outside events | A competitor's office opened nearby in late January; no tracking changes | CRM lead source field renamed on 1 February, logged in the measurement change log | Spring landlord promotion ran from 1 March |
| Actual effect and verdict | Cost per request fell to £52; requests 84 against 92 the month before. Stands | Instructions from paid social rose from 11 to 14 a month; leads fell by a third. Stands; next step logged as a new entry to test a second lookalike | Requests from email rose from 6 to 15 over two sends, but the March send overlapped the promotion. Stands provisionally; read again on the April send |
At the March review the manager read the three entries first, before the channel reports. Entry C showed why the outside events column matters: the promotion could explain part of the rise, so the verdict was held for one more send rather than celebrated.
Common mistakes
- Logging the change but not the reason. An entry that says what changed duplicates the platform's change history. The value is in the evidence, the expected effect and the check date.
- Writing the expectation afterwards. An expectation written once the result is known always matches it. Fill the expected effect and the guardrail on the day of the decision.
- No check date, or one nobody reads. Entries without a date are never closed. Put the date in the review calendar and start every review with the entries that have fallen due.
- Separate logs per channel. Effects cross channels; a paused audience on one platform can move results on another. Keep one log the whole team writes to.
- Agency and automated changes left out. If an agency or an accepted recommendation changed the account, the log needs the entry just as much as for a change you made yourself.
- Guardrails set at round numbers. A guardrail inside the metric's normal range fires on noise and teaches the team to ignore it. Set it from the metric's own history.
- Several changes on the same day in the same place. Two edits to one campaign on one day cannot be told apart later. Space them, or log them as one decision and accept that their effects will be read together.
- Tracking changes logged here instead of in the measurement change log. Keep measurement changes in their own log, and list them in the outside events column when they overlap a decision.
The memo, results report and other documents that feed decisions into the log are in the templates library.
