--- name: lead-qualification-call description: Qualify a warm, consenting lead by phone with CALL-E, capture needs, budget, timing, decision authority, and interest as structured JSON, and route the disposition to a human or CRM without cold outreach. license: MIT --- # Lead Qualification Call Use this skill when a sales or marketing team has **warm** inbound leads that have explicitly consented to a call and wants one disclosed CALL-E phone call to qualify them. The call collects needs, budget range, timeline, decision authority, and interest level, and returns a fail-closed structured JSON result with a disposition a human or CRM can act on. This skill qualifies a lead. It does **not** sell, book, close, or write to a CRM. A human reviews the result and owns every next step, including any CRM update. ## When To Use - Qualify leads from an opt-in form, callback request, or consent-gated campaign - Collect needs, budget range, timeline, decision authority, and interest in one disclosed call - Separate high-intent leads from low-intent ones with transcript-evidenced structured results - Feed a disposition (qualified / needs_human / not_interested) to a human-run sales queue or CRM routing ## When Not To Use - Cold outreach, scraped lists, unsolicited marketing, or SPAM-style dialing - Medical, legal, financial, emergency, collections, or political calls - Selling, closing, negotiating, or collecting payment on the call - Calling a number the user did not provide and authorize - Hidden retries, recurring schedules, or provider-side campaigns - Writing to a CRM or booking a meeting before a human reviews the result ## Required Inputs - `lead_id`: stable local identifier (3-64 chars of letters, digits, dot, dash, underscore) - `campaign_id`: campaign or source-batch identifier for idempotency and routing - `lead_source`: Google Form, website callback, ticket, partner, or other consent-gated source - `to_phone_e164`: E.164, authorized - `consent`: must be `true` - `authorized_reason`: why this specific call is allowed - `company_name`: business name disclosed on the call - `product_offer`: what is being qualified (offer name or one-line description) - `timezone`: IANA timezone for quiet-hours decisions - `qualification_fields`: the subset of BANT fields the team wants, chosen from the schema contract Optional: `locale`, `language`, `region`, `lead_first_name`, `notes` passed as call context. ## Preflight 1. Confirm the user authorized this one qualification call and that `consent` is `true`. 2. Confirm the phone is E.164 and came from the consent-gated lead record. 3. Confirm the lead is warm: the record has an explicit outreach basis or callback request. If the lead is cold or `do_not_call` is true, refuse. 4. Build the compiled task from the goal template below and review a masked preview before any CALL-E plan. ## Dry-Run Preview Preview is the default. It performs **no network access and no call**. From the skill directory: ```bash python3 -m json.tool assets/sample-lead.json ``` Review the compiled task, the result schema, the masked phone, and the idempotency key before touching CALL-E. All example numbers are reserved fictional numbers (see `references/examples.md`). ## CALL-E Goal Template ```text You are an AI phone assistant calling on behalf of {company_name}. Disclose immediately that you are AI and that this is one lead-qualification call. Speak with {lead_first_name}. If you have the wrong person, apologise and end the call. Purpose: understand whether {product_offer} fits this person's needs. Do not sell, pitch, close, negotiate, collect payment, or give medical, legal, or financial advice. Ask only the approved qualification questions: - What problem or need is this person trying to solve? - What budget range applies: under 1k, 1k to 5k, 5k to 20k, or above 20k? Mark not disclosed if they decline. - What timeline applies: immediate, 1 to 3 months, 3 to 6 months, 6 months or more, or unknown? - Are they the decision maker, or do they need to include someone else? - How interested are they: high, medium, or low? Base this only on what they say, not on guessing. - Would they like a human from the team to call back, and when? Do not infer answers from silence. If the person declines to answer a question, record the value as unknown, not_disclosed, or low as appropriate. Never pressure, repeat an unanswered question aggressively, or invent an interest level. If the call is a voicemail, leave a short message and record disposition voicemail. ``` ## Structured Result ```json { "needs": "short text, empty if not disclosed", "budget_range": "under_1k | 1k_5k | 5k_20k | 20k_plus | not_disclosed", "timeline": "immediate | 1_3_months | 3_6_months | 6_months_plus | unknown", "is_decision_maker": true, "interest_level": "high | medium | low", "callback_requested": true, "disposition": "qualified | needs_human | not_interested | blocked_compliance | voicemail | no_answer | wrong_number" } ``` A `qualified` result is an **interest signal, not a closed sale**. `needs_human`, `voicemail`, `no_answer`, and low-confidence answers all route to a human and are never auto-qualified. See `references/qualification-schema.md` for the full contract. ## Live Planning Only after explicit user authorization and CALL-E authentication: ```bash export CALLE_API_KEY="" calle call plan --to-phone --goal "" --timezone --language English --region ``` `calle call plan` is planning, not execution. Do not start the call unless the user separately confirms. Read `references/calle-workflow.md` for the exact plan/run/status sequence, preflight, and idempotency handling. ## Cancellation And Idempotency Idempotency key: `lead_qualification:{lead_id}:{campaign_id}`. If the user cancels before dial, do not execute. If a result is ambiguous, route to a human rather than calling again. Never automatically retry an error or unknown outcome; that would create a duplicate qualification call. ## Safety Notes Read `references/safety.md` and `references/examples.md` before any live planning. Read `references/crm-routing.md` before routing a disposition, and `references/calle-workflow.md` before the live sequence. No CRM write and no meeting booking happens without a human review.