Conversion Performance AI analysis
Analyse with AI produces a written assessment of one conversion path: where attempts progressed or stopped, what the evidence suggests about possible barriers, and which changes are worth making. It reads the dashboard's own evidence for the selected path and filters, and it changes nothing on the website.
The analysis works step by step and then across the whole journey, so the report contains a per-step assessment and a single prioritised action list drawn from them.
Before you start
Select one conversion path. The analysis assesses a specific configured path, so Analyse with AI stays unavailable while All conversion paths is selected.
Set the dates and any audience, device or campaign filters first. The analysis uses the dashboard's current selection and cannot be re-scoped from inside the panel.
Run an analysis
- Set the path and filters on the dashboard.
- Select Analyse with AI.
- Check the scope listed under New analysis.
- Select Analyse with AI in the panel.
- Follow progress, or close the panel and return later.
- Open the finished report from Recent, or from the panel when it completes.
Progress runs through four stages: preparing the evidence, assessing the required steps, connecting the whole journey and saving the report. The step stage reports how many steps are complete.
Work continues after the panel closes. Reopen the panel and check Recent rather than starting a second analysis for the same selection.
Read the report
The report opens on the Analysis tab, headed by the period, device and audience it covers and the time it was created.
| Part | What it contains |
|---|---|
| Headline and summary | The main conclusion for the whole path |
| Recommended actions | A single prioritised list, each with what we see, why it matters, the recommended action and a completion check |
| Supporting analysis | Expandable sections: one across the journey, then one per required step |
Each action carries a priority of immediate, near term or can wait. Where a suitable measure exists, the action states its current baseline and what to look for, using the same path configuration and filters.
An analysis can recommend nothing. Where the evidence does not justify a change, the report says so rather than manufacturing a task, and that is a finding in its own right.
The Evidence tab
The Evidence tab shows the measurements behind the report and the qualifications that apply to it.
Recorded step outcomes lists the progressed and confirmed stopped counts for the steps discussed. One journey can appear at more than one step, so these rows are not additive.
What to bear in mind collects the limitations. Shared limits appear first, followed by any note specific to a single step, such as a measurement covering only part of the attempts.
What the analysis will and will not conclude
Whole-journey conclusions use completed attempts and confirmed drop-offs only. Attempts still open or without settled outcomes are background counts and are never treated as abandonment.
Counts establish what was observed, not what caused it. The analysis distinguishes an observation from a hypothesis, and an absence of confirmed stopping supports leaving a page alone rather than proving the page is effective.
Several boundaries are held deliberately:
- The configured final interaction defines completion. Configured value is not verified revenue, and the analysis will not infer sales, lead quality or return on investment.
- Attempts are not distinct people or sessions, and dashboard session metrics are context rather than settled-attempt rates.
- No statistical significance or causal effect is claimed, and small samples are described as such.
- Page content is read from the current page and, where one exists, an existing screenshot. The current page can differ from what earlier visitors saw, and a capture does not confirm what was visible.
Add an action to the Board
Select Add to Board on an action, then choose a project where more than one is available. The task carries the action's wording, its acceptance criteria and the evidence behind it.
The task is a record for planning and review. It does not implement the change, publish anything, or monitor the outcome automatically. Adding the same action to the same project twice returns the existing task rather than creating a duplicate.
Recent, export and delete
Recent lists saved analyses for the selected path, newest first, with the date, status and the scope each was run with. Unread reports are badged as new. Reports run with different filters appear together, so check the scope line on each.
Completed reports are shared with everyone who can access the website. An analysis still running is listed only for the person who started it, so a colleague will see it once it finishes rather than while it works.
Export report in the report header saves a standalone copy of the analysis and its evidence.
The five most recent completed reports are retained for each path. Export anything that must outlive that. Delete removes a saved report for everyone with access to the website; tasks already added to the Board keep their own evidence and are unaffected.
Credits
Running an analysis uses credits, and available credits are checked before any work starts. Reopening a saved report does not run a new analysis and does not use credits again.
Where credits are insufficient, the panel reports it and no analysis begins.
Troubleshooting
Analyse with AI is unavailable
Select one conversion path. The button stays disabled while all paths are selected, and its tooltip says so.
The analysis could not be finished
The panel reports that completed work has been retained and offers Retry saved result. That retry re-checks the work already done without running a new analysis. Use it before starting again.
The selection is too large
The analysis supports up to twelve required steps across the selected path's versions. Where the selection contains too much evidence for one analysis, choose a shorter period or a smaller path.
I closed the panel and lost the analysis
Work continues in the background. Reopen the panel and check Recent before starting another analysis for the same selection.
The report recommends no changes
That is a valid outcome. Where the evidence shows no confirmed barrier, the analysis reports that a change is not justified by the sample rather than inventing one.
An older report is no longer listed
Five completed reports are retained per path. Export reports you need to keep beyond that.
The report and the dashboard disagree
The report analyses settled attempt outcomes for the path, while the dashboard's headline measures are session-based. The two answer different questions and are not reconstructed from one another.