Change Correlation / Daily Movers
Pro FeatureChange Correlation powers the in-app Daily Movers briefing: each day it detects which key metrics moved materially, identifies when the move happened, and produces a ranked shortlist of suspected contributors. The result is a concise written summary of what changed and what probably drove it — delivered every morning without you having to dig.
Change Correlation and the Daily Movers briefing are available on Pro, Scale, and Enterprise plans. Pro includes up to 90 days of history; Scale and Enterprise include up to one year. The feature is not available on Free. Upgrade your plan to enable this feature.
Correlation, not causation. Change Correlation surfaces statistical associations between metric movements and candidate events in the same time window. A high-confidence ranked contributor is evidence worth investigating — it is not proof that the contributor caused the change. Treat every output as a starting point for your own review, not as a definitive diagnosis.
What it does
Every day, Change Correlation runs against the previous day's data for each of your tracked websites. It checks five key metrics for material moves, collects everything that changed in the same window, ranks contributors by how closely they correlate with the move, and writes a short briefing that summarises the ranked evidence.
The briefing appears in the Daily Movers tab of your domain dashboard. It is generated fresh each day; history accrues over time up to the retention window for your plan.
Metrics covered
Change Correlation monitors five key metrics:
| Metric | What it measures |
|---|---|
| Revenue | Daily revenue from connected payment providers |
| Signup conversion | Visitor-to-signup rate for your defined signup goal |
| Checkout completion | Cart-to-purchase rate for your defined checkout goal |
| Bounce rate | Share of single-page sessions with no meaningful engagement |
| Exit rate | Share of pageviews that are the last in a session, per page |
How a mover is detected
A metric qualifies as a mover only when two conditions are both true:
- Material move: the day-over-day change exceeds a fixed threshold relative to the trailing 7-day baseline.
- Minimum volume floor: the metric had enough events that day to make the reading statistically meaningful. Sub-floor readings are flagged separately as possible noise rather than included in the main ranked list.
This two-condition gate prevents low-traffic days from generating false alarms. If your site saw very few signups on a given day, a 40% swing is not a mover — it is noise, and it is labelled as such.
Candidate collection
When a mover is detected, Change Correlation collects everything that changed in the same trailing window that could plausibly be related:
- Deploys and Git commits — via connected deploy markers
- Error clusters — new or spiking error groups from Error Tracking
- Web Vitals regressions — material drops in LCP, CLS, INP, or FID from Core Web Vitals
- Traffic-source shifts — changes in the mix of referral sources, direct, organic, and paid
- AI-source-mix shifts — changes in the share of traffic attributed to AI assistants and crawlers via AI Influence
- Reconciliation context — whether the move looks like a real change or a measurement artefact (loss vs delay), when reconciliation data is available
- Segment-mix shifts — changes in the composition of visitor segments (geography, device, plan tier if tracked)
Deterministic weighted ranking
Candidates are ranked by a deterministic weighted scoring function — not by a model guessing at importance. The ranker runs fully before any summary is written, and the summary only rephrases what the ranker already determined. Nothing appears in the briefing that is not already in the ranked list.
The scoring function weighs four factors for each candidate:
| Factor | What it captures |
|---|---|
| Temporal proximity | How closely the candidate's timing aligns with the onset of the metric move |
| Route / segment overlap | Whether the candidate's affected routes or segments match where the metric moved |
| Magnitude | How large the candidate's own change was relative to its own baseline |
| Novelty | Whether this candidate is new or returning (a brand-new error cluster scores higher than a recurring one) |
Each ranked candidate receives a confidence band:
| Band | Meaning |
|---|---|
| High | Strong statistical association across multiple factors; investigate first |
| Moderate | Meaningful association on at least one factor; worth reviewing |
| Low | Weak or indirect association; included for completeness |
| Tentative | Association is plausible but evidence is thin; treat as a lead, not a finding |
Confidence bands reflect the strength of statistical association between the candidate and the metric move. They are not a statement about cause and effect.
The written briefing
Once ranking is complete, a concise briefing is written that summarises the ranked evidence in plain language: which metric moved, by how much, compared to the baseline, and which candidates ranked highest and why. The briefing does not introduce any interpretation beyond what the ranker already determined.
Where a candidate has supporting detail available — a specific route, a specific error group, a specific traffic source — that detail is included. Where supporting detail is not available for a given contributor type, the briefing notes the association without fabricating specifics.
Reconciliation context
If server-vs-client reconciliation data is available for your site, Change Correlation uses it to distinguish two types of metric moves:
- Real change: the move appears in both client-side and server-side data, suggesting something genuinely changed in user behaviour or funnel performance.
- Measurement artefact: the move is present on one layer but not the other, suggesting a tracking loss, a webhook delay, or a data pipeline gap rather than a real behavioural change.
When a mover looks like a measurement artefact, the briefing says so plainly — and the reconciliation context is included as one of the ranked candidates.
Sub-floor movers
Metrics that moved materially in percentage terms but did not meet the minimum-volume floor are not promoted to the main ranked list. Instead, they are grouped in a separate Possible noise section of the briefing. This keeps the main list focused on moves that are statistically meaningful, while still surfacing low-volume swings in case they are relevant to you.
Drill-down links
Each ranked contributor in the Daily Movers briefing deep-links into the relevant part of your dashboard:
- Error clusters → Error Tracking
- Web Vitals regressions → Core Web Vitals
- Traffic-source shifts → Sources tab
- Deploy markers → deploy timeline in the Performance tab
Plan availability and history window
| Plan | Available | History |
|---|---|---|
| Free | No | — |
| Pro | Yes | Up to 90 days |
| Scale | Yes | Up to 1 year |
| Enterprise | Yes | Up to 1 year |
History accrues from the day the feature is first active for your website. There is no back-fill for days before activation.
Related
- Server-vs-client reconciliation — distinguish real metric changes from measurement artefacts
- Core Web Vitals — Web Vitals regressions surfaced as Daily Movers candidates
- Error Tracking — error clusters surfaced as Daily Movers candidates
- AI Influence — AI-source-mix shifts surfaced as Daily Movers candidates
- Golden-Path Monitors — dual-signal guardrail for your most important conversion path
- Conversion Incidents — auto-detected drops in defined goals and funnels