How to Analyze Changes in Website Traffic

Traffic went up. Traffic went down. The number itself is the least interesting part — what matters is whether the change is real, what caused it, and whether it will continue. This guide gives you a repeatable method for answering those questions instead of reacting to every wiggle in the chart.

Choose a valid comparison period

Every "traffic changed" claim is really a comparison between two windows of time, and bad window choices create phantom changes:

• Compare like with like. Full weeks against full weeks, complete months against complete months. Comparing a 7-day window that contains a weekend against one that doesn't will show a "change" that's just day-of-week mix.
• Avoid partial periods. "This month vs. last month" on the 20th compares 20 days against 30. Percentage deltas from partial periods are noise.
• Match the window to the question. Week-over-week answers "did last week's launch work?" Month-over-month shows trend direction. Year-over-year answers "are we growing?" while controlling for seasonality.

Seasonality

Most sites have rhythms: B2B traffic sinks on weekends and in late December; retail spikes in November; education follows semesters. A month-over-month drop that recurs every year isn't a problem to fix — it's a pattern to expect. Year-over-year comparison is the standard control: December 2026 against December 2025, not against November 2026. If you have less than a year of history, note suspected seasonal effects rather than concluding around them.

Comparing periods in Umami

Umami's Compare insight overlays two periods directly:

1. Select the date range to analyze (for example, Last 7 days).
2. Open the Compare insight and choose:
   • Previous period — the same-length window immediately before, or
   • Previous year — the same dates one year earlier.
3. Read the metric cards (percentage change per metric, with up/down indicators) and the overlaid graph of views and visitors across both periods.

Details are in the Compare documentation.

Rule out measurement changes first

Before explaining a change, verify it isn't an artifact of how you measure. The classic false alarms:

• Tracking script removed or broken by a redesign, a template change, or a tag-manager edit. Symptom: an abrupt, uniform collapse across every source and page.
• URL structure changes — a migration that moved /blog/post to /articles/post makes old pages "lose" all their traffic while new URLs start from zero.
• New subdomains or missing templates — a section of the site that never got the script simply doesn't exist in your data.
• Bot filtering or blocker changes — shifts in how much automated or blocked traffic is excluded move numbers without any human behavior changing.

A useful habit: keep a shared changelog of site deployments, tracking changes, and marketing launches. Half of traffic forensics is just lining the chart up against that log.

Localize the change with breakdowns

A site-wide percentage hides the actual story. If visitors are up 30%, the increase is almost never uniform — it's concentrated somewhere. Use the Breakdown insight to find where:

The pattern of concentration usually identifies the cause class:

• One referrer → a link, mention, or algorithm change on that platform.
• One page → content performance: a post ranking, being shared, or slipping.
• One country, often with odd engagement metrics → possible bot traffic.
• Everything at once → either a measurement problem or something brand-level (press, an outage, a major campaign).

Where relevant, also compare new-versus-returning behavior: a spike of first-time, single-page visitors means reach; growth in returning visitors means retention. A retention report tells you whether spike visitors ever came back.

Attach the change to an event

With the change localized, match it against what happened in the world:

• Product launches and releases — expect spikes on launch content and referral surges from wherever the announcement spread.
• Campaign starts and stops — check UTM-tagged traffic; a "mysterious" drop is often just a paused ad campaign. See campaign tracking.
• Outages — site downtime creates a gap, but so does downtime of a major referrer.
• Search-ranking movement — gradual rises or slides concentrated in organic-search referrers and specific pages.
• Tracking changes — from your changelog, per the section above.

Worked example: investigating a traffic spike

A SaaS marketing site sees visitors up 42% this week versus last week. The analysis:

1. Compare confirms the increase is concentrated on three days, Tuesday–Thursday.
2. Breakdown by referrer shows the growth comes almost entirely from one aggregator site; other sources are flat.
3. Breakdown by path shows the traffic landed on a single blog post, not the homepage or product pages.
4. Conversion check: signups (tracked as a signup goal) rose only 4%. The spike audience read one post and left — reach, not demand.
5. Conclusion for the team: the post resonated on that platform; syndicate similar content there, but don't revise the traffic forecast or credit the spike to product interest. Next week's "42% drop" is already explained.

Note what step 4 adds: pairing traffic analysis with conversion tracking is what separates "we got visitors" from "it mattered."

Common mistakes

1. Explaining noise. Constructing a narrative for every daily wiggle, especially on low-traffic sites where percentages swing wildly. Consequence: constant false alarms and strategy whiplash. Remedy: judge changes against your site's normal variation, and analyze weeks rather than days.
2. Comparing invalid windows. Partial months, mismatched weekday mixes, or holiday weeks against normal weeks. Consequence: phantom changes that vanish when the window closes. Remedy: complete, like-for-like periods; year-over-year when seasonality is plausible.
3. Stopping at the site-wide number. Reporting "traffic up 20%" without localizing it. Consequence: the wrong team takes credit, and a one-off referral spike gets baked into forecasts. Remedy: always break down by referrer and path before reporting.
4. Confusing measurement changes with audience changes. A redesign drops the tracker from half the templates and the ensuing "traffic crisis" consumes a quarter's roadmap. Consequence: real cause unfixed, fake cause "solved." Remedy: rule out tracking artifacts first, every time.