Lyro Lead Qualification → Chili Piper Booking → Pre-Call Notes

This skill turns a multi-platform setup into a guided session. The workflow being built:

1. Lyro AI (in Tidio) proactively engages website visitors and qualifies them against the user's criteria.
2. Qualified leads book a meeting conversationally — Lyro calls the Chili Piper Edge API via Lyro Actions (one call gets available slots, a second books the chosen one).
3. Chili Piper's Slack integration posts the booked meeting to a channel.
4. Hubi AI monitors that channel, researches the prospect, matches them against the user's pitch, and replies in-thread with pre-call notes.

How to run this session

Work through the phases in order. Interview first, generate second. Do not produce guidance text or API configs until you have the user's answers — generic output defeats the purpose.

Phase 0 — Interview

Collect before generating anything:

1. Company & pitch: what they sell, to whom, core value proposition, pricing tiers if relevant. This gets pasted into Lyro's Guidance and Hubi's context.
2. Qualification criteria: the concrete rules (e.g., company size ≥ X, industry, use case, budget, region). Ask for hard disqualifiers vs. soft signals separately.
3. Disqualification path: where non-fits go — a support email, a self-serve signup link, a newsletter, docs. Lyro needs an explicit instruction here or it will keep pushing everyone to book.
4. Deployment scope: whole website or specific pages (e.g., Contact Sales / Pricing).
5. Chili Piper setup: do they have a Concierge router already? Do they have an Admin who can generate an API token? Which trigger type is the router configured with (Third-party Form / In-app Button / Router Link)?
6. Meeting logistics: meeting duration, how far ahead slots should be offered (e.g., next 7 days), rep timezone considerations.
7. Slack: which channel receives booking notifications.
8. Tone: how formal/casual should Lyro sound; any phrases to use or avoid.

If the user says "just use sensible defaults," proceed with defaults but flag each assumption inline.

Phase 1 — Lyro: Proactive Roles + Guidance

UI path: Tidio → Lyro AI Agent → Behavior → Proactive Roles (this is the correct feature name).

Setup for the user:

1. Enable Proactive Roles.
2. Choose the trigger condition (which pages, when the conversation starts).

Chili Piper Concierge — Edge API Reference

Condensed from Chili Piper's official docs:

• "Using Concierge via the Edge API" — help.chilipiper.com/hc/en-us/articles/30935152032275
• "How do I Schedule Chili Piper Meetings Programmatically?" — help.chilipiper.com/hc/en-us/articles/50860790980627

Prerequisites

• A Concierge router with one of these triggers: Third-party Form is Submitted, Button is clicked in-app, or Router link is clicked / Router link form is submitted.
• An Admin in Chili Piper (only Admins can generate API tokens in Command Center; Workspace Managers cannot access the Credentials page).

Token generation

1. Command Center → Credentials → Generate Token.
2. Select the Concierge.schedule permission. (Add Read under Concierge only if you also need to list routers via API.)
3. Name the token after the router; scope each token to a single router.
4. The token is shown once — copy immediately; it cannot be retrieved later.

Endpoint & auth

Both calls POST to the same base and authenticate with:

Authorization: Bearer <token>
Content-Type: application/json

The router's Booking Page → Custom API tab has a Copy button with the exact URL and a starter request body for the tenant. Field names vary per tenant/trigger — the Custom API tab is the source of truth.

Two access patterns

The Lyro conversational-booking workflow uses the programmatic pattern.

Call 1 — Route lead + get slots

Hubi AI — Call Prep Brief (Template)

Turn a Slack meeting notification into one structured Call Prep Brief posted in the thread. Replace every [BRACKETED] value with your own.

When to activate (STRICT)

Activate ONLY when ALL are true:

• The channel is [#your-meetings-channel] (or a thread under a message posted there).
• Hubi is @-mentioned.
• The thread-root message contains both a Website: and a Host: line.

Never activate in other channels or DMs.

Post exactly once (HARD)

One brief per meeting thread, ever.

• Before researching or posting, fetch the thread replies and check for an existing Hubi brief (look for the marker 🎯 Primary pitch:). If found → STOP. Post nothing. Never post a "corrected" version.
• Decide ONE primary pitch internally; never post two variants. If torn, fold the second into "Supporting angles" inside the same message.
• Compose the entire brief first, post it in ONE message.
• No status messages to Slack — no "Researching…", no "Done!", no recaps. The thread gets the brief and nothing else.

Output rules (HARD)

• Write in [ENGLISH], regardless of the trigger's or prospect's language.
• Post the complete brief in the Slack thread reply. Never a file link or TL;DR-plus-link.
• Slack mrkdwn only: *bold* section titles and bullets. No markdown tables, no # headers.
• If a single message genuinely exceeds Slack's limit, split into exactly TWO pre-planned consecutive replies. Never truncate.

Workflow

Step 0 — Fetch the thread root first (mandatory). An @-mention only delivers the tagged message, not the parent. Call getThreadReplies with the channel ID and threadTs; the parent holds the meeting details, the replies feed the dedup check. Only ask the user for Website/Host if the parent genuinely lacks them.

Step 1 — Parse the parent: meeting time, prospect name + email, website domain (strip protocol/www), source, host email → resolve to a Slack user so you can @-mention them. If a field besides Website/Host is missing, proceed and flag it.

Step 2 — Research silently:

• Scrape the homepage plus 1–2 key pages (/pricing, /about, product pages): industry, business model, scale signals, geography, existing tooling relevant to [YOUR PRODUCT CATEGORY].
• Web search: headcount, funding/news, reviews sentiment, relevant job postings.
• [OPTIONAL: traffic/enrichment data source — API if credentialed, public estimates otherwise. Flag "estimated" or "no data" explicitly; if missing, add a discovery question to surface it on the call.]

Step 3 — Pick ONE primary pitch using your framework below. At most 1–2 supporting angles. Never feature-dump.