Direct answer: circulate one Evidence Card before the review, spend the meeting verifying completeness and describing changes without causes, choose the single decision that matters most, and close with a One-Decision Record containing the action, owner, evidence basis, due date, stop condition and next review date.
This Analytics Review Agenda begins after metrics have already been defined and captured. It is not another list of KPIs or a dashboard setup guide. A creator can run it alone in twenty-five focused minutes or use it as a short meeting format with approved collaborators.
Separate workflows own the detailed evidence. Use the profile conversion test log when a completed profile test needs interpretation. Bring only its approved summary into this review rather than reopening every raw record.
Set the review contract
Choose a recurring time, time zone, period boundary, facilitator and decision owner. State which sources must be ready before the review and how late or incomplete data will be marked. A consistent appointment reduces the temptation to review only after a strong or weak result.
Define the meeting output: one Decision Record, optional evidence request and a dated note of items deliberately left unchanged. If a subject cannot lead to a decision within the review's scope, route it elsewhere instead of letting it consume the session.
Keep attendance narrow. A solo creator needs only the creator. A small team should include people who own the evidence or the action. Additional observers can read the record later.
Prepare the Evidence Card before discussion
The Evidence Card is a one-page pre-read. Include review ID, selected period, comparison basis, source timestamps, completeness state, metric definition versions, three neutral observations, known operational changes, open discrepancies and candidate decision questions.
Link to source records rather than copying fan-level information. Use aggregate or de-identified evidence appropriate to the question. A candidate question should be answerable through an owned action, a defined test or an evidence request.
Distribute or finish the card before the meeting begins. If the evidence owner cannot distinguish complete from missing fields, the first decision may simply be to repair capture. Weak evidence should not be dressed up with a longer discussion.
Open with data quality, not performance
Spend the first five minutes checking period coverage, source freshness, missing states, changed definitions, duplicated records and reporting delays. Confirm that the selected comparison uses the same scope and unit.
Classify the Evidence Card as usable, usable with caveats or not decision-ready. A caveat must name the affected observation. If the evidence is not ready, stop performance interpretation and assign the smallest repair needed.
This gate protects the rest of the agenda. A polished chart cannot compensate for a changed denominator or a partial period accidentally compared with a full one.
Describe observations without explanations
Read each observation in neutral terms: which measure, which period, which comparison and how large the recorded change was. “New subscribers were lower than the previous complete period” is an observation. “Subscribers were bored” is an unsupported explanation.
Limit the list to three observations that could affect a current decision. Stable results can be relevant when they show that an existing action does not need interruption. Do not force movement where the source precision cannot support it.
When churn evidence is involved, bring the bounded output from the churn reason tagging template. Preserve unknown reasons and the sample boundary instead of turning tagged records into a claim about every subscriber.
Separate known context from hypotheses
Known context is a dated event: posting paused, a price changed, a promotion link broke or content delivery moved. A hypothesis proposes a relationship: the pause may have contributed to fewer profile visits. Keep the two in separate fields.
List no more than three plausible hypotheses for the selected decision question. For each, record supporting evidence, contrary evidence and what observation would make it less plausible. This prevents the loudest first explanation from becoming fact.
Do not spend the review proving every cause. The agenda exists to choose the next useful action. A hypothesis can remain unresolved while the creator selects a low-cost evidence request.
Copy the Analytics Review Agenda
Use this timed sequence as the practical artifact:
- Minutes 0–5: verify period, sources, definitions and completeness.
- Minutes 5–10: read up to three neutral observations and known context.
- Minutes 10–15: choose one decision question and list bounded hypotheses.
- Minutes 15–20: compare action options, evidence needs and constraints.
- Minutes 20–25: complete the One-Decision Record and close every other item.
If the review runs longer, identify which input or decision right was unclear. Improve the next Evidence Card rather than permanently extending the meeting.
Choose one decision question
Frame the question around something the creator can control in the next period. “Which profile element should remain unchanged while we complete the current test?” is more actionable than “Why is growth slow?”
Score candidate questions on decision urgency, evidence readiness, reversibility, creator effort and interference with current work. Select one. Send additional promotion ideas to the promotion backlog prioritiser rather than launching them during the review.
A decision can be to continue unchanged. Deliberate continuity is useful when a test needs more complete exposure or when the evidence does not justify disruption.
Complete the One-Decision Record
Fields: review ID; decision question; chosen action; action owner; evidence card version; observations relied upon; known caveats; start date; completion condition; stop condition; expected evidence; next review date; and status.
Write the smallest observable action. “Keep the current bio through the next complete test window” is clearer than “optimise conversion.” If the action is a new test, link its separate pre-run brief and do not redesign it inside this record.
State what will remain unchanged. This protects the evidence from simultaneous edits and tells collaborators which ideas were intentionally deferred.
Close loops from the previous review
Before finalising the new action, inspect the last Decision Record. Mark it completed, continued, stopped, superseded or blocked. Add the evidence produced and whether its completion condition was met.
Do not mark an action successful merely because a headline metric increased. Confirm that the agreed action occurred and the promised evidence was captured. Outcome interpretation follows the original question and limitations.
Carry a blocked action forward only with a named blocker and owner. Otherwise, close it and choose a feasible next step. Repeatedly copying stale actions creates the appearance of work without learning.
Handle conflicting signals
A review may show rising profile visits, flat subscriptions and incomplete source attribution. Resist combining these into one story. Record each observation, the shared time boundary and the missing evidence.
Prioritise the decision closest to a controlled action. The creator might preserve the destination while repairing source capture. A later complete period can then support a profile or channel decision.
When metrics use different lags, record the expected observation window. Same-week comparison can be unsuitable when an action takes longer to appear in the selected outcome.
Run the agenda asynchronously when needed
A solo creator can complete the Evidence Card, leave it for a short cooling-off period and return to write the Decision Record. This reduces reactive changes immediately after seeing a result.
For asynchronous teams, assign comment deadlines and decision authority. Contributors may add evidence or challenge a hypothesis, but one named owner closes the record. Unresolved comments become evidence requests, not endless discussion.
Version the card after material corrections. The final record should point to the exact version used, allowing a later reviewer to distinguish the original decision basis from subsequent updates.
Audit the review habit monthly
Count how many reviews ended with a complete decision, how many actions closed, how many were blocked and how often missing evidence stopped interpretation. These are process signals, not creator performance scores.
Remove agenda sections that never affect decisions, and improve fields that repeatedly create ambiguity. Keep the twenty-five-minute constraint unless the creator's scope genuinely requires more.
Sample one old record and reconstruct the reasoning from source to observation to action. If that trail is not understandable, strengthen evidence locators and caveat fields before adding more metrics.
Keep an observation parking area
Some valid observations will not become this week's decision. Put them in a dated parking area with source, period and reason deferred. Do not convert them into actions after the meeting without reopening the decision process.
At the next review, remove items whose evidence is stale, merge duplicates and promote only those that now affect an owned question. This preserves useful signals without allowing the agenda to become a rolling list of anxious reactions.
Limitations and sources
Limitations: this agenda improves review discipline but cannot make incomplete or differently scoped data comparable. Observed movement does not establish why it happened, and a single weekly decision cannot resolve every creator question. Timing and review length should adapt to the creator's workload.
Google Analytics documentation supports explicit comparisons while describing reporting differences and limitations. GOV.UK performance guidance supports using measurements to improve a service. The Evidence Card, timed Analytics Review Agenda and One-Decision Record are original SirenCY editorial artifacts.