Back to case studies

AI Automation / 2026

Automotive Sales Intake

An automated intake workflow for automotive enquiries: capture, classify, price, record and route. Built in n8n and tested across eight scenarios, with human review for sensitive cases.

Discuss a project

The challenge

Bookings, sales questions and complaints arrived as unstructured emails in one inbox. Each needed a consistent process and a traceable record.

The workflow also had to avoid inventing missing details or pricing work it could not match, and it had to keep refunds, legal threats and other sensitive cases away from an automatic reply.

Objective

Capture, classify, price and route enquiries, with human review for sensitive cases and an audit trail from receipt to response.

  • Extract customer, vehicle and service details without guessing.
  • Price from an approved matrix and mark insufficient data instead of estimating.
  • Record the inquiry in the CRM and notify the team.
  • Send an automatic reply only when the case is safe to automate.

Who uses it

A sales and operations team that needs every enquiry to become a consistent record, a priced response and a clear next action.

The solution

A 33-node n8n workflow connects Gmail, Gemini, Google Sheets, HubSpot and Slack. It skips duplicates, assigns tracking references and extracts customer, vehicle and service details. Price estimates come from an approved Google Sheets matrix; missing data is marked insufficient.

Refund, legal and fraud-related enquiries route to human review without an automatic reply. HubSpot records, Slack notifications and a Sheets activity log keep the process traceable. There is no separate custom admin dashboard: n8n is the runtime, and the CRM, inbox and Slack are the operating surfaces.

  • Gmail trigger with one-minute polling.
  • Inquiry IDs and tracking references across every record.
  • Sensitivity classification with a human-review branch.
  • Master activity log in Google Sheets.
  • Deduplication and idempotency through a data store.

Key features

  • Inquiry identificationEvery message is assigned an inquiry ID and tracking number, so a case can be referenced end to end in Slack, the CRM and the activity log.
  • AI extraction with a never-guess ruleStructured extraction of customer, vehicle and service data. Missing data is marked insufficient, never invented.
  • Pricing without hallucinationA vehicle-tier and service lookup against an approved pricing matrix returns real estimates and refuses to price what it cannot match.
  • Sensitivity-aware routingRefund, legal and escalation cases are flagged automatically and routed to human review with no automatic reply; standard inquiries get an immediate, professional answer.
  • Full traceabilityHubSpot deals embed tracking references, and a master activity log records status at every stage, from the original message to the final reply.
  • Idempotent processingA data-store check prevents the same message from being processed or replied to twice, even with a crash or retry.

How the workflow flows

  1. 01 / Incoming emailGmail trigger, one-minute poll
  2. 02 / Normalise and dedupeDuplicate messages skipped
  3. 03 / IDs and trackingINQ and AUTO references
  4. 04 / ExtractionAI pulls vehicle, service and intent
  5. 05 / Tier and pricingMatrix lookup in Google Sheets
  6. 06 / Sensitivity checkRefund, legal and fraud routes
  7. 07 / Record and notifyHubSpot contact and deal, Slack alert
  8. 08 / Review or replyHuman review, or automatic Gmail reply

How it was built

The workflow was mapped from the exceptions first, because the ordinary enquiry is the easy part.

  • Sensitive cases never receive an automatic reply; they are routed to a person with the context attached.
  • Missing details are explicit. The extraction step is instructed not to guess, and the pricing step refuses an unmatched service.
  • Every stage writes to a shared activity log, so a case can be followed from the original email to the final action.
  • Duplicates are stopped before they create a second CRM record or a second reply.

Notable decisions

The workflow is designed to be honest about what it does not know.

  • No custom admin dashboard was built. n8n is the runtime, and the CRM, Gmail, Slack and Sheets are the operating surfaces.
  • Pricing is authoritative only when the lookup matches. An unknown service or vehicle tier is not assigned a guessed price.
  • Sensitive and uncertain cases are kept with a person, even when that means a slower response.

Current status

The workflow and integrations are built. Eight end-to-end scenarios passed: booking, sales inquiry, complaint, refund, legal threat, missing vehicle data, unknown service and duplicate email.

No production adoption, customer volume or business result is claimed. The project is presented as a built and tested workflow, not a live revenue system.

Scenarios tested

  • Booking inquiry
  • Sales inquiry
  • Complaint
  • Refund request
  • Legal threat
  • Missing vehicle data
  • Unknown service
  • Duplicate email

Delivery status

  • Completen8n workflow with 33 nodes
  • CompleteGmail trigger and normalisation
  • CompleteDeduplication and idempotency
  • CompleteInquiry IDs and tracking numbers
  • CompleteHubSpot contact and deal creation
  • CompleteAI extraction with Gemini
  • CompletePricing matrix lookup
  • CompleteSensitivity classification
  • CompleteSlack team notifications
  • CompleteAutomatic reply versus human-review routing
  • CompleteMaster activity log in Google Sheets
  • CompleteEnd-to-end test execution across eight scenarios