> ## Documentation Index > Fetch the complete documentation index at: https://docs.orbit.devotel.io/llms.txt > Use this file to discover all available pages before exploring further. # Outbound AI Studio: brief-to-draft, variant generator, and winner insights > One dashboard page that groups every AI-native outbound capability — draft a campaign from a plain-language brief, generate A/B variants for an existing campaign, and read A/B winners across finished campaigns. # Outbound AI Studio The AI Studio tab under **Outbound → AI Studio** groups every AI-native outbound capability onto one page, so you do not have to learn three separate entry points. It has three tabs: **Brief to draft**, **Variant generator**, and **Winner insights**. Any teammate who can open **Outbound → Campaigns** can open AI Studio; each capability inside it consumes your organization's AI credits the same way the other AI surfaces do. **The AI proposes, a person approves.** Nothing in AI Studio commits to a live campaign automatically. Generated copy only ships after you review it on the campaign's A/B editor, and a winner promotion only decides *which* message went out, never *whether it should have* — frequency caps and quiet hours still gate every send. ## Brief to draft Type or dictate what the campaign should do in plain language — "inviting past buyers to a 20%-off weekend sale" is enough — and the platform drafts the message copy, chooses an audience, and recommends a send time. Review the draft in the composer before anything becomes a campaign. This is the same brief composer you may have seen as **From brief** on the Campaigns toolbar; the AI Studio simply mounts it as a first-class tile. Below the composer sits a **Saved drafts** area that collects every brief you kept with "Save draft and close", so you can resume, rename, duplicate, or delete drafts without re-opening the composer. Three quick prompts pre-fill the brief with real account data, triggered from suggestion cards elsewhere in the dashboard (for example, from the Campaigns list): * **Resend to failed deliveries** — targets recipients a recent campaign failed to reach and drafts resend copy. * **Scale the winning variant** — seeds the brief with the declared A/B winner from a completed campaign. * **Re-engage inactive contacts** — targets the segment with no inbound activity in the last 30 days. Each quick prompt displays its evidence before the composer opens — for example, "Targeting 4,120 recipients who failed delivery on 'Weekend blast'" — so you can see what the AI will work with. Where no matching data exists, the brief falls back to a generic prompt rather than fabricating context. ## Variant generator Pick a launched campaign (completed, running, or paused) and optionally add a tone hint like "playful" or "more formal". The generator rewrites the campaign's base message into two to four alternatives, each returned with a `name`, a `message_body`, and a short `rationale` explaining the angle — a different hook, value framing, or call to action. Saving works one of two ways: * Open the campaign's A/B editor from the link below the results and post the generated set onto the campaign — either verbatim or after editing. * Regenerate with a different tone hint until one of the outputs reads the way you want. The generator stays grounded in your own copy: it rephrases claims the base message already makes and keeps the same personalization tokens and opt-out language. A suggestion never duplicates the opt-out line, and it never invents discounts, offers, or urgency your original did not carry. ```text Screen sketch — Variant generator theme={null} ┌───────────────────────────────────────────────────────────────┐ │ Variant generator │ │ ───────────────────────────────────────────────────────────── │ │ Campaign: [ Weekend blast · SMS · Completed ▼ ] │ │ Tone hint (optional): [ playful ] │ │ │ │ Generated variants — review before saving │ │ • Summer splash "Weekend's here! {{first_name}}, 20% off…" │ │ Short urgency hook, upbeat tone. │ │ • Quiet savings "{{first_name}}, save 20% this weekend…" │ │ Understated elegance — softens urgency to inclusiveness. │ │ │ │ [ Open campaign A/B editor ] │ └───────────────────────────────────────────────────────────────┘ ``` ## Winner insights The third tab lists your last completed campaigns. Expand any row to read its A/B results: the declared winner with the winning copy inline, the per-variant delivery, open, click, and reply rates, and the metric the winner was decided by — delivered, opened, clicked, or replied. While a verdict is still accruing, the row shows live per-variant rates and names the metric the verdict will use, so the numbers are never unlabeled. Two quick actions hang off each row: * **Use winner as brief** switches you to the Brief tab with the winner's copy pre-loaded and instructions to scale it to a broader audience. * **View campaign** opens the campaign detail page for the full experiment setup. ```text Screen sketch — Winner insights theme={null} ┌───────────────────────────────────────────────────────────────┐ │ Winner insights │ │ ───────────────────────────────────────────────────────────── │ │ ▼ Weekend blast SMS completed 2026-08-26 ✓ │ │ 🏆 Winner: Summer splash — decided by Click rate │ │ "Weekend's here! {{first_name}}, 20% off this weekend…" │ │ Delivered 98.2% · Opened — · Clicked 6.4% │ │ ─ Quiet savings … Clicked 4.1% │ │ [ Use winner as brief ] [ View campaign ] │ └───────────────────────────────────────────────────────────────┘ ``` ## Worked example: brief → draft → experiment → winner 1. On the **Brief** tab, describe the campaign and accept the AI's draft copy, audience, and send time. 2. Launch the campaign. As recipients respond, the experiment compares variants. 3. Back in AI Studio, open **Winner insights**, find the campaign, and expand it. When the engine has enough deliveries, the row shows "Winner decided by Click rate" (or whichever metric you chose) with the winning body. 4. Choose **Use winner as brief** — the Brief tab opens with the winning copy pre-loaded and asks you to broaden the audience. Review and send again. The loop reached "which message to scale" with a human reviewing every hop, which is the default trust policy. ## Failure modes * **Variant picker shows "No eligible campaigns found."** Only launched campaigns qualify. Create and run a campaign first, then return — this reads as an empty state, not a generation error. * **Nothing comes back from a generation.** The model occasionally returns no usable variants, usually when the campaign's base message is empty or invalid. Set a `message_template` on the campaign, or retry with a clearer tone hint. * **A campaign lists no A/B test on expansion.** Campaigns that ran A/B-disabled report cleanly as "did not run an A/B test" rather than erroring. Enable A/B at creation for the winner rollup to be meaningful. * **Generation fails with a friendly error message (not a blank result).** Network issues, per-tenant rate limits on the LLM route, or an unconfigured AI key all surface the same way — check the toast, then retry. * **A winner existed, but another verdict replaced it.** Manual select-winner calls update the verdict; the rollup reads the latest state on expand. ## Screen sketches The two code-fenced blocks above are labeled screen sketches of the Variant-generator and Winner-insights tabs — they mirror the layout you see on the page, not literal screenshots, so minor wording differences are expected. ## docs.json nav snippet The nav entry is additive; paste it after the campaigns section in the Guides group: ```json theme={null} "guides/campaign-ab-testing", "guides/campaign-end-to-end", "guides/outbound-ai-studio" ``` ## See also * [A/B test a campaign](/guides/campaign-ab-testing) — variants, traffic splits, and how winners are chosen. * [Launch a campaign end to end](/guides/campaign-end-to-end) — the full operator loop from brief to results.