Question 1

What can I push into Customer.io from a Databricks model?

Accepted Answer

People and their attributes, custom objects, segments, person events and people devices. Between them that covers what a messaging tool needs: who somebody is, what they did, and which group they belong to. Devices are append only, so they are added rather than matched and updated. Everything else follows normal matching, where you map a column from the model to the identifier Customer.io uses.

Question 2

Can I add a new attribute without touching Customer.io?

Accepted Answer

Yes. New attributes and properties can be declared directly in the mapping for people, events and custom objects, so a new column from your model does not need setting up on the Customer.io side first. That keeps a change like adding a churn score or plan tier to one place rather than a coordination exercise across two tools and two owners.

Question 3

What happens if the SQL warehouse is asleep when a sync runs?

Accepted Answer

The query wakes it, and the sync waits rather than failing. Connection retry allows roughly ten minutes specifically so compute can start, which is usually plenty. The cost is latency on that run plus the compute charge for starting up. If your messaging syncs are spread through the day and the warehouse auto stops between them, you are paying that warm up repeatedly, so it is worth either grouping schedules or lengthening the auto stop interval and weighing that against idle cost.

Question 4

Who should own creating people, the lakehouse or Customer.io?

Accepted Answer

Whichever system genuinely knows first, and the sync mode enforces your answer. Update or create lets the model introduce people, update only enriches existing ones and skips the rest, and create only adds new ones while leaving matches alone. If Customer.io is already fed by your product, update only avoids surprise records. Separately, if a related record referenced in a mapping cannot be found, Customer.io can be told to ignore that relationship and still write the base record.

Question 5

Where should churn scoring live?

Accepted Answer

Wherever the model is trained, which for most teams using Databricks means the lakehouse rather than the messaging tool. A score built from product events, billing history and support volume cannot be recreated inside Customer.io, and approximating it there produces a second definition that slowly disagrees with the first. Reverse ETL moves the output rather than the logic, so the campaign fires on exactly what the model said.

Sync Databricks to Customer.io

Sync Databricks to Customer.io with Polytomic

People, events and objects

Declare attributes in the map

Bad rows do not stop a run

Trusted by businesses worldwide

Step-by-step integration guide

Connect Databricks in Polytomic

Connect Customer.io in Polytomic

Create a sync between Databricks and Customer.io

Map fields and enable the sync

Frequently asked questions

The sync and ETL platform that does it all

Powerful data sync

Advanced analytics

Enterprise security

Easy setup

Ready to connect Databricks with Customer.io?