Understanding Blu Chat

Overview

Blu Chat is Intempt's conversational agent surface, not a page of its own: it's invoked from, and operates over, every other object in the console: a journey, a brand kit, an account, an analytics report, a meeting. Instead of clicking through forms, you describe what you want and Blu works out the steps.

Blu is optional by design. Every core feature of Intempt, CRM, journeys, experiments, dashboards, segments, and non-AI workflow steps, works fully without it.

Blu's work falls into five kinds of task:

Where you can use it

Blu Chat has a few entry points into the same conversation:

• The docked panel, opened from the toggle button in the console header. This is where it lives on most pages.
• A floating pill, a compact way back into an open conversation while you work elsewhere on the page.
• Directly from an object, a list, detail, or canvas view (journeys, brand kit, accounts, analytics boards, and more) opens Blu already scoped to what you're looking at.

Asking Blu to build something "from" an existing item mostly starts a new conversation rather than continuing the one already open on that item.

What it can do

A few confirmed examples of what you can ask, grouped by task type:

Analyze

• "How is revenue trending this month?"
• "Show me weekly revenue for the last 3 months"
• "What's our conversion rate?" This is ambiguous on its own, so Blu asks a clarifying question with a set of options before it answers.

Large results don't dump into the chat: asking for something like "show me top 150 users by lifetime value" paginates the answer, and a request that would return thousands of rows ("show all transactions this quarter") exports to CSV instead of rendering inline.

A question that matches an existing Insights, Funnel, or Retention report answers using that same engine; see Analytics for what each report type covers. A question that doesn't match any of those gets a new analysis composed for you instead: see Building a Custom Report with Blu.

Create and Author work the same way, asked and answered in the same conversation, with their own object types: see Generating Brand Assets, Using the Asset Library, and Drafting Emails and Replies with Blu for what each one covers in detail.

Unrecognized or newer content types inside a reply can fail to render rather than showing an error, so an unusually-shaped answer may simply look incomplete.

Where its answers come from

For Analyze, Blu's numbers come only from Intempt's own metric and attribution engine, the same engine that powers Insights, Funnel, and Retention reports. Blu narrates the result; it doesn't compute its own numbers. Asking Blu a question returns the same figure you'd get by building the equivalent report yourself in Boards.

Blu proposes, you approve

Outside of a direct answer (Converse) or a read-only question (Analyze), anything Blu is about to do goes through a plan you see and approve before it runs:

1. Approve, or something stricter. Most items need a click to approve. A destructive or high-impact action (for example, one that would affect a lot of records) instead asks you to type a confirmation phrase that has to match exactly, so it can't be approved by accident.
2. Edit before approving. If a plan item has editable fields, you can change them and have Blu regenerate that item, re-entering approval rather than skipping it.
3. Review or discard. You can expand any item to inspect it without running anything, and discard a pending item at any time; anything that depended on it is cancelled with it.
4. Skip, when it's offered, only appears on an action that doesn't change anything and doesn't affect much, never on a destructive or high-impact one.

Analyze answers are read-only by default; saving a result as a report or dashboard is the one Analyze action that still goes through this approve step.

Where to go next

• How Intempt works
• Generating Brand Assets
• Using the Asset Library
• Drafting Emails and Replies with Blu
• Building a Custom Report with Blu
• Analytics
• The Glossary