Meridian Churn Signals
Sign in
PIE runtime service · Meridian Bank

Ingest customer events.
Get churn risk out.

The concrete implementation behind the Collect churn signals capability in PIE. Send interaction events over MCP, and read computed churn risk scores plus top contributing factors per customer.

Register with PIESee example event

Real-time ingestion

Accept events from core banking, support, mobile, and web channels via MCP.

Risk scoring

Weighted model derives a 0–100 churn score plus top contributing factors.

OAuth 2.1 protected

Every MCP call acts as the signed-in user, with per-tool authorization.

Step 1

Register with PIE

Add this service as an external MCP server in the PIE platform. Both the MCP client and its authorization server live on the same origin.

Base URL
…
MCP endpoint
…
Service token (PIE): send Authorization: Bearer <PIE_SERVICE_TOKEN>. The same random value is stored here and pasted into PIE — calls run as a fixed service principal. Best fit for workflow orchestrators.
OAuth 2.1 (fallback): MCP clients that don't use a service token complete the standard flow at /.well-known/oauth-protected-resource and each call acts as the signed-in user.

Step 2

Exposed MCP tools

Five tools cover the full ingest → score → read loop.

Step 3

Example payload for ingest_event

Supported event types: transaction_declined support_ticket_created login_failure nps_submitted website_visit mobile_app_crash

ingest_event · payload
{
  "customer_id": "CUST-001",
  "event_type": "transaction_declined",
  "timestamp": "2026-08-03T08:06:06.881Z",
  "payload": { "amount": 249.90, "currency": "EUR", "reason": "insufficient_funds" },
  "source": "core-banking"
}

How it works

01

Ingest

PIE sends interaction events via ingest_event or ingest_events_batch.

02

Score

Events land in the raw log and are scored on demand or nightly.

03

Read

PIE reads the current churn_risk_score and top factors per customer.