STACK · FIVE LAYERS
I — V

The stack.

HubSpot, n8n, Claude, BigQuery and Measurement — each layer with its own page, setup and example workflow.

  1. I
    01

    HubSpot

    HubSpot runs as the pipeline engine here. Lifecycle changes, lead routing and task creation happen in workflows that I set up and document so your team can maintain them later without me.

    WorkflowsListsPropertiesSequences
    Layer detail
  2. II
    02

    n8n

    n8n self-hosted on a Hetzner VM. Multi-step sequences, webhook triggers and whatever HTTP call a tool happens to need. The instance is yours — data, backups and update timing included.

    Webhook TriggersHubSpot NodeHTTP RequestSlack Node
    Layer detail
  3. III
    03

    Claude

    Claude sits right on the HubSpot record via MCP. The prompts are built per use case and run in batch, and every token lands as a log in BigQuery — you can watch along.

    Anthropic APIMCP ServerVercel AI SDKPrompts-as-Code
    Layer detail
  4. IV
    04

    BigQuery

    Sink for HubSpot, GA4 and LinkedIn Ads, with a joined fact table behind it. Looker Studio or Metabase as the frontend — C-level gets one URL and clicks around on their own.

    BigQueryAirbyteLooker StudioMetabase
    Layer detail
  5. V
    05

    Measurement

    Since the iOS updates and cookie banners, client-side tracking only shows you a slice of your conversions. Statistical MMM works on aggregate data instead, with no PII or cookies involved, and still makes channel contribution and saturation curves measurable.

    PyMC-MarketingBigQueryAirbyteCloud Run
    Layer detail

Each layer works on its own. It gets interesting once all five hang together.

Discuss the stack