Choose complete windows
Do not compare a finished month with the first few days of the next one. For weekday-sensitive activity, use complete weeks or equivalent weekdays. Document the reporting timezone and the timezone used for transaction records.
Keep the metric consistent
Users, sessions, orders and revenue describe different things. Write the definition before calculating a percentage. More sessions can coexist with fewer orders; showing only the first measure can conceal the commercial question that needs investigation.
Annotate relevant events
Record promotions, closures, migrations and measurement incidents. Simultaneous changes alone do not establish causation. When collection methods change, show a break in the series instead of suggesting that old and new observations are directly continuous.
Preserve absolute volumes
Percentages can disguise a small starting point. Show starting and ending values alongside the absolute difference. Use a longer window for small segments where appropriate, and explain collection limitations.
- Choose window
- Define metric
- Annotate
- Compare
Worked example
Illustrative example: moving from four to eight orders is +100%, but only four additional orders. Also show visits, opening days and measurement incidents so readers can interpret the movement.
Check before moving on
These checkboxes are not saved.
Sources & method
Sources reviewed on 1 October 2026. Worked calculations use explicitly hypothetical inputs.
Topic provenance: Enfin des statistiques par semaines ou mois sur Google Analytics (2008-04-04). Search Engine Feng Shui. Byline retained in the publication feed: François Houste.
This guide is newly written in 2026. It does not reproduce the earlier text and is not attributed to its historical author.
Complete a two-period comparison before calculating a percentage
Fill the scope record before interpreting a change. The complete fictional example below compares sessions within one invented Analytics property whose reporting timezone is UTC. It is not an export from this website, a count of unique people or a measured improvement.
- Choose the same metric definition, property, hostname and landing-page scope. Preserve the exact channel filter; a session-based source/medium filter is not interchangeable with a first-user acquisition filter.
- Choose two complete windows and record their timezone. The example uses 5–11 October and 12–18 October 2026, both Monday to Sunday, in UTC. Check completeness when exporting; a date label alone does not establish that all observations have arrived.
- Record changes to consent, collection rules and site behaviour. If the definition or collection changed, show a break or restrict the comparison; do not hide it inside a percentage.
- Show the first value, second value, absolute difference and percentage denominator. Treat a zero baseline as a separate boundary case.
Completed fictional record: metric: sessions; timezone: UTC; source/medium scope: google / organic; country, device and landing-page filters: unchanged and explicitly recorded; window A: seven complete days, 100 recorded sessions; window B: seven complete days, 120. The descriptive difference is 120 − 100 = 20; relative to A it is 20 ÷ 100 × 100 = 20%. No collection change is assumed for the teaching calculation; a real comparison must verify that condition.
Two checks that stop a misleading chart: if B contains only Monday to Wednesday, do not compare it with the complete A as a weekly change. If A = 0 and B = 5, the absolute difference is 5 but division by the baseline is undefined; do not label it “+500%”. Equal dates and filters also do not make Search Console clicks equal to Analytics sessions: Google documents their different definitions and collection systems.
Comparison sheet: export/version/time ___; property and hostname ___; metric and definition ___; session or first-user scope ___; date boundaries/timezone/completeness ___; filters ___; collection or consent changes ___; A ___; B ___; absolute difference ___; percentage denominator or undefined ___; remaining limitation ___. This record supports a reproducible description, not attribution of the change to a campaign or site edit.