Skip to content
All case studies

Home Services / Agencies (GoHighLevel) · Reference build (self-built, deployable per account)

An AI Brain That Runs Your Whole GoHighLevel Lifecycle, From the First Lead to the Repeat Job

A complete AI revenue lifecycle engine running on GoHighLevel for a home-services brand. It takes a lead from capture through triage, first touch, conversation, booking, onboarding, and retention, with an AI making the judgment at every stage. The whole engine runs webhook-only, so it needs no GoHighLevel API token and drops onto the base plan, and it sits on a five-layer reliability spine that makes the AI safe to trust with live customers.

Results

  1. 8 stagesFull lifecycle, capture to retention, on one AI brain
  2. 5-layer spineAudit, failover, human-gate, observability, multi-tenant config
  3. No API tokenWebhook-only, runs on the base GoHighLevel plan

Role

Solo build: architecture, AI decision design, n8n workflows, reliability spine, observability dashboard

Tools

  • GoHighLevel
  • n8n
  • Claude (Haiku + Sonnet)
  • Gemini (failover)
  • Supabase

The problem.

Triggers are easy, judgment is hard

Native GoHighLevel automation reacts to events. It does not triage a lead by how hot it is, read a reply in context to decide between answering and booking, or route an ambiguous case to a human. Judgment at every lifecycle stage is the actual deliverable, and it lives outside the CRM.

An AI that writes to customers has to be trusted

A wrong AI send is a burned lead. A real lifecycle engine has to know when it is unsure and hand off to a person instead of guessing, and it has to keep running when the model behind it has a bad minute.

The API path locks you to a plan tier

Reading and writing GoHighLevel through its API sits behind higher plan tiers and per-account tokens. An engine built on that dependency cannot drop onto most agency client accounts without a tier upgrade and a credential project.

The solution.

  1. Step 1

    A webhook-only brain

    GoHighLevel fires a webhook on every lifecycle event to one of seven n8n workflows. Each calls Claude for the judgment and writes the result back as a native GoHighLevel action. No API token, runs on the base plan, GoHighLevel stays the system of record.

  2. Step 2

    Eight lifecycle stages with AI judgment

    Capture and triage (three-pass consensus), first touch (brand-voice opener), conversation (intent read: answer, book, stop, or hand off), booking (link on the base plan, fixed-time appointment with API mode), onboarding (plan plus welcome sequence on a won job), and retention (review-and-maintenance pass on a completed job).

  3. Step 3

    A five-layer reliability spine

    An audit trail of every AI decision, automatic Claude-to-Gemini failover, a human-in-the-loop gate for low-confidence or ambiguous calls, a live observability dashboard over the audit trail, and a one-file config that points the whole engine at a new tenant.

Architecture diagram, click to zoom

Screenshots

The engine's own observability dashboard: KPI strip, lifecycle-stage index from capture to retention, and a stage-filterable log of every AI decision with the model and outcome.
The engine's own observability dashboard: KPI strip, lifecycle-stage index from capture to retention, and a stage-filterable log of every AI decision with the model and outcome.
A single contact's conversation trail: each inbound message, the intent the engine read, the action it took, and the raw decision JSON behind it, in order.
A single contact's conversation trail: each inbound message, the intent the engine read, the action it took, and the raw decision JSON behind it, in order.
A reply-and-opt-out trail: the engine reads the intent, honors the opt-out immediately, and writes the whole decision to the audit log.
A reply-and-opt-out trail: the engine reads the intent, honors the opt-out immediately, and writes the whole decision to the audit log.

The impact.

Demonstrated capability (build-facts)

  • Eight lifecycle stages live, capture through retention
  • Seven n8n workflows, all running
  • Three-pass AI triage consensus that only routes on agreement
  • Five-layer reliability spine: audit, failover, human-gate, observability, multi-tenant config
  • Webhook-only architecture, runs on the base GoHighLevel plan with no API token

Why it is built this way

  • The reliability spine is the product: it is what makes AI judgment safe to put in front of live customers
  • Refusing the API dependency makes the engine portable across GoHighLevel accounts on the plan they already have
  • One config file turns the build into a multi-tenant template, one account per config

Have a similar problem?

Tell me what is going on and I will tell you what I would do about it. No obligation.

Get in touch →