Quarterly Review: What to Continue and What to Stop

On the first Monday in November, a DTC power-station brand holds its quarterly review. Six dashboards, one per action: mentions up, cited pages up, AI-referred sessions up. The CEO asks which one gets cut for Q1. Silence, then the familiar answer — keep all six, and spread the budget until every action gets half of what it needs.

A quarterly review should end with a stop list, not a report card: run each action through incrementality, cost, and certainty in that order, sort it into scale, hold, narrow, or stop, and default to stop when all three come back blank.

Why the review turns into “keep everything”

First, the numbers genuinely rise, mostly not because of you. Entry points keep moving and answer formats keep widening, so mentions climb for an action nobody touched. Credit that drift to six actions and every dashboard proves itself.

Second, there is no comparison. Last quarter the team rewrote pages, patched the feed, and ran outreach in the same twelve weeks; all six actions claim the improvement. Without a control you cannot say what would be missing had the March rewrite never shipped, so nothing can be cut. That is incrementality testing: a blank is “unknown,” never “effective.”

Third, nobody owns the stop. Adding budget passes an approval; removing it passes through blame.

Fourth, the reporting unit is wrong. By channel, each one brings the metric that flatters it. By action, they fit one table: questions covered, pages changed, person-days eaten, whether anything sold.

Three filters, and the order is not optional

Filter	Answers only	Source	If it fails
Incrementality	What disappears from answers and orders if we stop	Fixed prompt set, holdouts	Mark “unknown,” not “effective”
Cost	Person-days bought one unit of increment	Cash, hours, queue time	Narrow scope or cheaper form
Certainty	Survives a model version and a new owner	One-sentence mechanism, no single point	Demote to pilot, hold budget

The order exists because of the denominator. Incrementality first, or cost has nothing to divide by: “we saved money” may mean we saved on worthless work. Cost second, or “effective” becomes “at any price.” Certainty last, because it sizes the bet, not the decision. Reversed, cheap and useless survives while effective work resting on one writer gets over-funded.

Sourcing is an org problem. Citation records, crawl logs, and order margin must join on URL and SKU, or the filters run on memory; see the data pipeline.

From filters to an agenda in four steps

Reading dashboards aloud is an agenda failure. Fixed steps, two hours, decisions out.

Step	Do	Owner	Done when
1 List actions	Only what consumed real budget or person-days, ten rows max	Growth or analytics	Each has an owner and start and end month
2 Run incrementality	One sentence: what would be missing if we stopped	Growth plus action owner	No blanks, only certain or unknown
3 Run cost	Cash, hours, and queue time into person-days per increment	Finance plus action owner	Two actions compare on one scale
4 Verdict and signature	One of four verdicts; scale and stop need a signer	Business owner	A stop list and next quarter’s schedule exist

Step 1 deletes rows on the spot. An action with no owner and no start and end month never really ran; file it as “not started.” An exhaustive list times out the meeting.

Four verdicts: scale, hold, narrow, stop

Verdict	Condition	Move	Signer
Scale	Measurable increment, acceptable unit cost, repeatable mechanism	Add people, markets, or page types on the same criteria	Business owner
Hold	Increment exists, cost is high, or it has not run a full quarter	Keep funding, write the next review’s condition	Project owner
Narrow	Mechanism holds in one market, language, or category	Keep that slice, pause the rest	Project owner
Stop	Two quarters without measurable increment, or cost never buys increment	Stop funding, keep the assets	Business owner and finance

Stop does not mean delete. Pages and feed fields are paid-for sunk cost, and pulling one can remove the sentence that cites you. What stops is capacity: outsourced word count, tool subscriptions, reserved hours. Which pages to merge or retire belongs to a content portfolio audit, run beforehand.

Narrow is the verdict teams skip: they choose between all and nothing, then kill a mechanism that worked in German because it failed in English.

Example: six actions, three stops

Example — brands, page counts, and hours are invented.

Action	Incrementality	Cost	Certainty	Verdict
Localize 12 core pages for Germany	Cited in German questions at local prices	Medium, one-time	High, repeats English mechanism	Scale to the next language
Complete agent feed fields	Not measurable, agent traffic still small	High, touches master data	Medium, platform-dependent	Narrow to top 3 SKUs
20 generic blog posts a month	No control, scattered citations	Medium, ongoing	Low, undifferentiated	Stop, use 4 data-led pieces a quarter
Third-party review outreach	Two new independent citations	Low	High	Scale
Full structured-data rebuild	Shipped, nothing left to test	Low, paid	High	Stop (complete), keep in maintenance
Second monitoring tool	Overlaps the existing one	Annual fee	Low	Cancel subscription

The decision is never “which action worked” but “which slice worked.” Two rows scale, one narrows, three stop; two of those free real spend, and it funds the scale items. A review hands out a schedule, not a grade.

Where this breaks

First, quarter one has no baseline. Without a fixed prompt set, the incrementality column is empty; the only honest output is “build a baseline.”

Second, the answers say something false. Fix the fact first; “no measurable increment” is not a reason to stop a correction.

Third, platform-dependent actions. Read listings and feed work with platform rules; one flat quarter is not failure. Watch field completeness, not orders.

Fourth, the review as accountability hearing. Once stop means admitting failure, the team polishes metrics and the next table is useless. Reward the voluntary cut.

Fifth, “no time to do it” is not “can’t measure it.” Both look blank; one needs capacity, the other a stop.

This quarter, do one thing: list last quarter’s actions in ten rows or fewer, three cells each — what would be missing if we stopped, how many person-days it cost, whether it survives a model change. Unfilled rows get marked “unknown” and one line of budget next quarter. For the rhythm, browse the measurement and growth hub; for where the first quarter starts, return to the 90-day roadmap.