---
title: "Behavioral Source Migration"
slug: "behavioral-source-segmentation"
updated: 2025-08-05T10:33:07Z
published: 2025-08-05T10:33:07Z
canonical: "help.madkudu.com/behavioral-source-segmentation"
---

> ## Documentation Index
> Fetch the complete documentation index at: https://help.madkudu.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Behavioral Source Migration

This comprehensive guide provides a systematic approach to migrating behavioral data sources while maintaining model performance and data integrity. The process ensures seamless transition from legacy integrations to new data sources without disrupting ongoing sales operations or model accuracy.

The methodology outlined here emphasizes careful validation, performance comparison, and rollback capabilities to ensure business continuity throughout the migration process.

## Prerequisites - Week 1

**Critical requirement: MadKudu technical team involvement is mandatory for batch processing steps**

- Plan accordingly to MadKudu's availability for this project
- Identify the current behavioral data source and the target migration source
- Confirm which behavioral integrations are currently connected
- Assess the scope of migration (lead-level, account-level, or both)
- Verify technical requirements and access permissions for the new data source

## Data Source Setup - Week 1

- Establish the new connection
  - [Configure the new behavioral data source integration](/v1/docs/integrations-overview)
  - [Verify connectivity and data flow from the new source](/v1/docs/video-data-discovery-page)

## Analysis - Week 2

- Compare the volume of records per event per person and per month during the past 9 months
  - Extract historical data from both old and new sources
  - Analyze data volume consistency between sources
  - Identify any discrepancies in event tracking or data collection
  - Continue the migration process only if volume differences are acceptable

## Migration Preparation - Week 3

- Duplicate the event mapping, but put it on 'off'
  - [Recreate the current event mapping for the new integration](/v1/docs/how-to-create-or-edit-the-event-mapping)
  - Set the duplicated mapping to 'off' status to prevent immediate activation
  - Verify all event definitions and parameters are correctly copied
- Duplicate the behavioral models
  - Create copies of all active behavioral models (Lead level and/or account level)
  - Label duplicated models clearly for migration tracking

## Critical Migration Steps - Week 3

- Stop the batch - **⚠️ Important**: This step **can be performed by MadKudu only**
  - Ensure all current data processing is completed before stopping
  - Halt batch processing
  - Document the exact time of batch cessation for rollback purposes
- Switch event mappings
  - Turn on the new event mapping configuration
  - Turn off the old event mapping
- [Load new datasets for draft behavioral models](/v1/docs/likelihood-to-buy-training-and-validation-datasets)
  - Validate that all expected events from the new event mapping were properly loaded
- [Assign weights](/v1/docs/likelihood-to-buy-event-weights)
  - Assign the same weights and lifespan for each event that matches the same event from the old integration
- Compare the performance of the old model vs the new
  - Pay close attention to the Sample page, review with a pair

- **If performance is NOT similar:**
  - Turn the original event mapping back on
  - Turn off the new event mapping
  - Re-enable batch processing

- **If performance IS similar:**
  - Publish the new behavioral model(s)
  - Re-enable batch processing
  - Monitor the performance closely during the initial production period

## Post-Migration Monitoring - Week 3&4

- [Monitor model performance for at least one week post-migration](/v1/docs/how-to-monitor-your-madkudu-customer-fit-model-with-the-predictions-page)
- Gather feedback from sales teams on any observable changes
- Maintain ability to quickly revert to original configuration for 30 days
- Keep original event mappings and models in archive state

**Estimated time: 20 hours**