Experiments and attribution
Compare message or campaign variants, keep each customer's variant stable, and measure the result. Goals, revenue, and report limits.
이 페이지에서
An experiment helps you compare variants, such as two popup messages or two campaign branches. You set the share of the audience for each variant and choose a goal, such as a payment. The report shows how each variant performed.
A variant stays assigned to a customer or to one campaign run. If the same event is processed again, the assignment does not change. This is deterministic allocation: the same participant receives the same variant.
Configure a version
Add two to five variants and allocate exactly 100% of the audience. In the API, shares use basis points: 100 points equal 1%, so the total must be exactly 10,000. You may set one control group, for example a group that sees no message. Choose who keeps the assignment:
- Subscriber: the same person remains in the same variant.
- Journey: one campaign run stays in its branch, but a later run for that customer may receive another variant.
The assignment is saved with the experiment version and assignment ID. A repeated button event (callback) cannot move a participant to another variant.
Before starting, set the audience, main goal, and optional additional goals. Choose where the report gets a monetary value and set the attribution window, the time after an interaction during which a result counts. Depending on your setup, the value can come from a constant, an event property, a payment, or a CRM deal. The experiment version can reference compatible content.
Versions and lifecycle
An experiment can be draft, running, paused, completed, cancelled, or archived. Starting uses the current unchangeable version; pausing does not change the assigned variants. You can archive a completed or cancelled experiment.
After the first participant sees a variant (an exposure event), changing the content or allocation creates a new version. The old version remains available in the report. Select the right version number when comparing results. You choose a winner manually, once; the choice remains visible after archiving. It does not by itself prove that a difference between variants is statistically reliable.
Facts and attribution
The report shows variant assignments, views (exposure), deliveries, opens, clicks, replies, goal completions (conversions), unsubscribes, and complaints. Repeating the same event does not count it twice: each fact has a stable source event ID. Revenue is stored separately by currency in its smallest units: 1250 with EUR means €12.50. Integrations send the amount as a signed 64-bit number and the currency as an ISO 4217 code.
Revenue is grouped by currency. EUR, RSD, and USD are never silently added together. To compare amounts across currencies, use an exchange rate for a specific point in time.
Choose one attribution model for a report:
- First touch credits an eligible conversion to the first touch.
- Last touch credits it to the latest eligible touch.
- Assignment credits it to the assigned experiment variant.
The selected model and the unchangeable version's attribution window are shown in the UI and API response. Models are not mixed in one total.
Warnings and cohorts
Warnings flag possible data problems: the experiment is still running, a variant has fewer than 30 views, a variant was assigned but never shown, audience shares differ substantially from the plan, or clicks and conversions outnumber views. Check delivery and event recording before choosing a winner. These warnings do not measure statistical significance.
A cohort here is the list of people assigned to a variant. You can page through it; the API uses a cursor to request the next page. A report filtered by segment uses the segment's currently published membership, not its membership on the experiment date.
Compatibility with popups and campaigns
When an older multi-variant popup moves to the shared experiment system, its past events must be transferred first. A returning visitor keeps the variant already shown, so the content does not change mid-journey. Popup statistics keep their existing response format and use the old counters until the old and new counts match.
For a campaign split (split), the system saves the time it switched to the new accounting (cutover). Runs started before that time keep the previous allocation. The shared report becomes available after those runs finish; missing historical assignments cannot be reliably recreated.
API and MCP examples
The REST report needs a PAT with experiments:read and the Analytics bot
right. The range is bounded RFC3339 UTC and the JSON does not contain a real
token or customer payload.
curl -X POST "$BASE_URL/api/bots/$BOT_ID/experiments/$EXPERIMENT_ID/report" \
-H "Authorization: Bearer $PAT" \
-H "Content-Type: application/json" \
-d '{
"version": 2,
"attribution_model": "assignment",
"from": "2026-07-01T00:00:00Z",
"to": "2026-08-01T00:00:00Z",
"segment_id": "00000000-0000-4000-8000-000000000000"
}'
400 means an invalid report request. 404 means the experiment or explicitly
requested version is unavailable in this bot. 409 applies to an optimistic
update conflict or a winner-selection conflict, not to a missing report
version. Read an assignment cohort with version, optional variant, limit,
and the returned cursor as query parameters.
MCP exposes read-only shared experiment tools with the same experiments:read
scope:
{"name":"get_experiment_report","arguments":{"bot_id":"<bot-id>","experiment_id":"<experiment-id>","version":2}}
list_experiments and get_experiment are also available. MCP does not start,
pause, edit, choose a winner, create assignments, or write facts.