AI Search (AI Assist)
AI Search lets you describe what you want in plain English and have JourneyLayer fill in the builder for you — using your project's real events and properties. It's available in an increasing number of modules, all powered by one shared, module-aware framework.
AI Search never performs risky actions. It does not publish campaigns, start journeys, send messages, or delete anything. It only prefills a builder, applies a filter, or creates a draft that you review and save yourself through the normal flow. Every result is validated against JourneyLayer's existing schemas before it reaches the builder.
How it works
- You type a request (e.g. "cart abandoners in the last 7 days").
- The request is sent with your project's real event + property catalog as grounding.
- The model returns a structured, validated result — never free text and never raw SQL.
- The builder is prefilled. You review it (segments show a live count), adjust if needed, and save.
Unknown names are never guessed
If you mention an event or property that isn't in your project's catalog, AI Search will ask you to clarify and suggest the closest real names — it will not silently substitute a different event. This keeps your queries honest.
Where it's available
| Module | What it does | Example |
|---|---|---|
| Events | Fills the Events explorer with a behaviour query (did / did not, counts, property values). | "users who viewed pricing 3+ times but didn't purchase" |
| People (Find People) | Fills the Find-People audience with did / did-not events + a date range. | "users who did Add to Cart but not Purchase in the last 7 days" |
| Segments | Builds a draft segment (event behaviour + geography + user properties) for you to review and save. | "cart abandoners", "users from Mumbai who purchased more than 2 times", "has a push token but no email" |
| Funnels | Builds a draft funnel (an ordered list of steps) to review and save. | "funnel from App Opened to Product Viewed to Purchase" |
| Trends | Prefills the Trends builder (events over time, optional split-by). | "trend Add to Cart by city for the last 30 days" |
| Cohorts | Prefills the retention Cohort builder (start event → return event). | "weekly retention from App Opened to Purchase" |
| Pivots | Prefills the Pivot builder (an event broken down by 1–2 properties). | "pivot purchases by device and campaign source" |
| Campaigns | Drafts a campaign — audience + conversion goal + content suggestions. Never sends. | "draft an email campaign for cart abandoners" |
| Journeys | Drafts a journey — entry trigger, waits, messages, goal check. Never starts. | "when users add to cart, wait 30 min, email them, then check Purchase" |
Every module runs through the same shared framework, so the behaviour and safety rules above stay consistent everywhere.
Module notes
Events
Maps your request onto the same structured query the Events builder uses: a
primary event, did / did not event blocks (with optional count thresholds
like "3+ times" or "spent over 5000"), interest/affinity rules, and property
filters. Always review the populated chips before running.
People (Find People)
The Find-People builder supports did / did-not event lists + a date range. If you ask for a property or geography filter (e.g. "from Mumbai"), AI Search builds the events it can and adds a note pointing you to Segments, which supports the full filter set — it won't quietly drop your constraint.
Segments
Produces a complete draft segment validated against the Segment DSL, using the same rule types the builder renders:
- Event behaviour — did / did not, with optional count ("more than 2 times" → at least 3) and per-event property filters.
- Geography — country / region / city ("from Mumbai").
- User & app properties — e.g. "has a push token but no email" →
$push_token is setANDemail is not set; device type, plan, source, etc.
The draft is shown in the builder with a live audience count. Nothing is saved until you click Save — AI Search never creates the segment for you.
Analytics — Funnels, Trends, Cohorts, Pivots
The four analytics builders accept the same plain-English assist, each validated against the exact DSL its engine consumes:
- Funnels — "from A to B to C" becomes an ordered, multi-step funnel draft (2–10 steps). Review the live conversion preview, then Save.
- Trends — chart 1–5 events over time, with an optional split-by property ("by city"), a measure (events vs unique users), and a granularity.
- Cohorts — retention from a start event to a return event, bucketed by day/week/month.
- Pivots — break one event down by 1–2 of its properties, counted by events or unique users.
For all four, AI prefills the builder — it never runs analysis or saves anything on its own, and unknown event/property names come back as a clarify.
Campaigns & Journeys — draft/config only
AI Assist for Campaigns and Journeys is strictly draft and configuration help. It will:
- Campaigns — propose the audience (validated as a real campaign qualification — saved segment or inline did/did-not behaviour), an optional conversion goal, and plain-text subject/body suggestions.
- Journeys — propose a draft flow: an entry trigger (e.g. enrol when a user does Add to Cart), waits, message steps, and a goal check. The plan is expanded into a valid journey graph in the builder.
What it will never do:
- ❌ publish a campaign · ❌ send a message · ❌ start a journey · ❌ change any live automation.
Even if you ask it to "send now", "publish", or "start this journey", it only prepares the draft and reminds you to review and click the Save / Publish button yourself. Unknown events or segments come back as a clarify with suggestions — never guessed.
Availability
AI Search requires the platform to be configured with an AI provider. If it isn't enabled on your deployment, the assist box returns a clear "not configured" message and the builders work exactly as before.