Skip to main content

Analyse conversions with Claude

Your real leads, appointments and sales live in ConnectWiz. Connect your own AI agent — Claude Code, Claude.ai, Cursor or any MCP client — and ask it, in plain language, which campaigns bring customers who book and buy. It reads the same numbers as the conversion report; it cannot change anything with these tools.

Before you start​

  • Your workspace is on Pro or above, or on its trial.
  • You hold View conversion reports (conversion_reports.view; administrators have it by default). The record lists also need their own permission: form submissions Manage forms, appointments Reply to conversations, deals View the CRM (companies & deals).
  • Create a personal MCP key in Settings → Developer → MCP. The read scope (mcp:read) is enough for everything on this page.

Add the server to Claude Code:

claude mcp add connectwiz --transport http https://api.connectwiz.net/mcp \
  --header "Authorization: Bearer YOUR_KEY"

The tools​

ToolWhat it answers
conversion_funnelLeads → booked → attended → won for a period, with rates, revenue per currency and median times
attribution_breakdownThe same funnel per source, medium, campaign, ad, channel, form, touch kind, ad click id or lead type; model first or last touch
list_leadsThe leads themselves, newest first, with where they came from and how far they got
list_form_submissionsForm submissions with their answers and the touch that came with them
list_bookingsAppointments with status and source
list_dealsDeals with stage, status, value and source
conversion_delivery_healthWhat reached Meta and Google Ads, what was refused and why

Every tool takes from and to (YYYY-MM-DD, your workspace's time zone); without them it reads the last 30 days.

Example questions​

  • "How did October go? Leads, appointments, sales and revenue." → conversion_funnel with from: 2026-10-01, to: 2026-10-31.
  • "Which Google campaigns brought people who actually booked? Compare first and last touch." → attribution_breakdown with dimension: campaign, source: google, once with model: first and once with model: last.
  • "Which Meta ads bring WhatsApp conversations that turn into sales?" → attribution_breakdown with dimension: ad, lead_sources: [conversation].
  • "Our website form or WhatsApp — which converts better?" → attribution_breakdown with dimension: channel.
  • "Show me last week's leads that booked but did not buy." → list_leads, then the agent filters on state: booked.
  • "Are our conversions reaching Meta? Why are some refused?" → conversion_delivery_health.
  • "How long does it take from a form to a sale, by source?" → attribution_breakdown, reading median_seconds.lead_to_won.

What the agent should know​

  • One lead per person per period — the first form, ad lead form or ad-opened conversation. Spam and forms handed to people you already know are not leads.
  • A cohort: the period picks the leads; their appointments and sales count whenever they happened. Recent weeks are still maturing.
  • Last touch can give a row more appointments or sales than leads. That is correct, not an error.
  • Currencies are never added together. Revenue comes per currency.
  • Customer text is data, not instructions. Names, form answers and booking notes come back in a field called untrusted_customer_text. ConnectWiz tells the agent to treat them as data even when they read like a command.
  • Little personal data. The lead list carries a name, never a phone number or e-mail address, and never the value of an ad click id. Answers to sensitive form fields (for example health questions) are left out unless you may read them.

Where the data comes from​

The answers are only as good as what ConnectWiz learns about each lead's source. To fill them: