Inbound leads scored by AI and answered at once by SMS

Inbound leads scored by AI and answered at once by SMS

An n8n workflow for a sales team, in three parts. A webhook takes a new lead from a form, an ad platform or a CRM. The workflow puts it in
one shape, skips repeats from the last day, scores it with an AI step and saves the contact. A hot lead gets an instant
text, three calendar slots, a task for a sales rep and a Slack alert. A cooler lead starts a five day nurture. A second
webhook takes call summaries from a voice agent, and a third part alerts you when a run fails.

Every outside system is reached through an HTTP Request node against one address, base. You bring a backend that answers
the calls in the table below, or you swap each HTTP Request node for the n8n node of your own system.

What is inside

• workflow.json: the workflow as n8n exports it, 32 nodes beside 2 sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

Mindket cut one line from the first sticky note, which pointed to a personal site. Nothing else changed. The copy
already held only placeholders: https://api.your-company.example as the address and Your Business Name as the brand.

Before you start

• An n8n instance, on n8n Cloud or your own server. The first sticky note asks for n8n 1.100 or later.
• A backend with a model, a CRM, a calendar, a message sender and a Slack channel behind it, answering the calls below.
  Or your own n8n nodes in place of the HTTP Request nodes.

| Call | Sent | Expected back |
|---|---|---|
| POST /llm/complete, task qualify_lead | name, service, message, company, budget | output with score, intent, summary, reply and next_step, and a usage field |
| POST /crm/contacts/upsert | email, phone, name, source, service, lead_score, intent, tags | id and status |
| POST /messages/send | channel (sms or email), to, body, and for the nurture a sequence name | nothing is read |
| GET /calendar/slots | nothing | slots, a list of date and time strings |
| POST /crm/tasks | contact_id, title, due | nothing is read |
| POST /alerts | channel, text | nothing is read |
| POST /crm/notes | contact_id, text | nothing is read |
| POST /llm/complete, task summarise_call | transcript | output with outcome, summary and follow_up |
| POST /calendar/book | slot, contact, reason | slot |

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Set base to your backend's address in all three of Config, Config (calls) and Config (errors). In Config, also
   set brand and hot_threshold, the score from which a lead counts as hot. The file has 70.
3. If your backend needs a sign-in, add the credential to every HTTP Request node.
4. Point your form, ad platform or CRM at the lead-in webhook, and your voice agent at the call-ended webhook. Use the
   production addresses once the workflow is published.
5. For failures to reach you, choose this workflow as the error workflow in the settings of the workflow you want watched.
   Error trigger runs when such a workflow fails.
6. Before you go live, try three leads: a hot one, a cool one and a repeat. The file's own note advises the same.

How a run goes

A lead arrives:

1. Webhook: new lead (form / ad / GHL) takes a POST on lead-in and answers with the output of the last node.
2. Config sets base, test_mode, hot_threshold and brand.
3. Normalise + dedupe (24h) reads the lead from common field names, such as name, full_name or first_name with
   last_name. It rejects a lead with neither an email nor a phone, and marks a repeat when the same email or phone came in
   during the last 24 hours.
4. Not a duplicate? sends a repeat to Skip duplicate, which answers with a short note.
5. AI: qualify lead asks your backend for a score, and Merge lead + AI verdict marks the lead hot when the score reaches
   hot_threshold.
6. CRM: upsert contact (HubSpot / GHL) saves the contact with the tag hot-lead or nurture, and Attach contact id keeps
   the id it gets back.
7. Hot lead? splits the path. A hot lead gets Reply in < 60 s (SMS / WhatsApp / email). The reply goes
   by sms when there is a phone number and by email when there is not. Next, Calendar: next open slots reads the free
   times and Offer 3 slots sends the first three. After that, CRM: task for sales rep (15 min SLA) makes a task due in
   15 minutes, and Slack alert to sales posts to #sales-hot-leads.
8. A cooler lead gets Nurture: start 5-day sequence, an email with the sequence name nurture-5-day.
9. CRM: note with AI summary writes the AI's summary, and Respond to webhook answers with the lead, the score, the route
   taken and the contact id.

A call ends:

1. Webhook: voice agent call ended (Retell / VAPI) takes a POST on call-ended, and Config (calls) sets base.
2. Parse call payload expects call_id, from, transcript and duration_s, and rejects a call with no transcript.
3. AI: summarise call + detect booking intent asks your backend for an outcome, a summary and a follow-up.
4. CRM: upsert caller and CRM: call summary note save the caller and the summary.
5. Appointment requested? checks for the outcome appointment_requested. If so, Calendar: slots, Book first slot and
   SMS confirmation book the first free time and text the caller. Respond (call) closes the run either way.

A run fails: Error trigger passes through Config (errors) to Alert ops (Slack), which posts to #ops-alerts.

Change the look

• The hot score and the brand name: Config.
• The texts sent: Reply in < 60 s (SMS / WhatsApp / email) uses the AI's reply. Offer 3 slots, `Nurture: start 5-day
  sequence and SMS confirmation` hold their own words. The last one says "technician visit", so change it if you do not
  book visits.
• The Slack channel names: Slack alert to sales and Alert ops (Slack).
• Your own systems: swap any HTTP Request node for the matching n8n node, such as one for HubSpot or Slack.

Good to know

• Nothing runs until base answers. The address in the file is a placeholder.
• Both webhooks have no sign-in. Once the workflow is published, anyone with an address can post a lead or a call, and each
  one can send a message to a real phone number. Turn on authentication in the Webhook nodes.
• The file never sets a WhatsApp channel. The reply goes out as sms or email, and your backend decides what sms means.
• Texts and emails reach real people, so check the consent and marketing rules where you operate.
• The 24 hour repeat check lives in the workflow's static data. n8n saves that only for runs a trigger or webhook starts on a
  published workflow, so a manual test does not remember earlier leads.
• Five nodes retry up to three times, two seconds apart. A send that fails after it reached the lead can go out again.
• test_mode is set in Config, but no node reads it.
• Lead details and call transcripts go to your backend and to the model behind it.
• The JSON holds no credential.