A technical view of financial data migration to Salesforce
At Pervasive, we encounter on a daily basis the heaps of data, disparate formats, and other challenges involved in moving to a new, multifaceted SaaS system, such as Salesforce CRM. A few years back, SunTrust Bank realized there were over a million records worth of contact, call history, etc. from a wide variety of random systems that they needed to migrate into Salesforce.
Best Practices for Data Migration
- Know what order objects should be inserted and what IDs must be preserved in the migration process – While provisioning the target system, Pervasive worked with Salesforce, and subject matter experts from all stakeholders on logical object planning to determine which objects were necessary for migration and how they were related.
- Identify quality rules and discriminators – Determining rules which the data should follow is essential. Using Data Profiler, we initially profiled the data and retrieved score cards to analyze what problems had to be dealt with. Then, we could catch and clean those records as they were migrated.
- Set up a proper mirrored test environment to determine how long it will run, and to catch any volume or time related ‘gotchas’ before show-time – Migrations can require a lot of time to move data, have a lot of bandwidth and API limitations to consider, and it is also difficult to undo the migration if a problem is found. With SunTrust, Salesforce.com’s Sandbox proved perfect for this.
- Prepare your users, whether co-workers or customers, for the switch. – Do some upfront impact analysis of network and related systems. Provide “scheduled maintenance” reminders for anyone still using either the old or new systems.
- Perform the migration at an off-peak time when network traffic and risk is lowest.
- Immediately begin validation of the data once the migration is complete. Don’t delay. If there are any problems, finding them fast is vital.
By following these best practices, Pervasive was able to get SunTrust’s first fully verified data loads completed in less than 30 days.