Cross-source analytics without a warehouse migration

6 min readUpdated August 11, 2026

Cross-source analytics answers a single question using data that lives in several systems. For example comparing pipeline in a CRM against invoiced revenue in an ERP. Doing it safely requires a certified join key between the sources, agreement on currency and fiscal calendar, and a grain check so that joining across a one-to-many relationship does not double-count. Where those are absent, the correct behaviour is to decline rather than return a plausible number.

The question that has no home

"Which of our largest open opportunities have overdue invoices?" is an ordinary commercial question. It is also unanswerable inside any single system: the opportunities are in the CRM, the invoices are in the ERP, and nothing in either system knows that account 4471 and customer NEXATRON-CUSTOMER-259 are the same company.

The conventional answer is a pipeline: move both into a warehouse, model them, and ask there. That works, and it costs months. The question is whether it is required before anyone can get an answer.

What has to be agreed before a cross-source answer is trustworthy

A certified keyAn explicit, approved statement that this field in system A identifies the same entity as that field in system B. Matching on company name is a guess, and a guess that is right most of the time is worse than one that is obviously wrong.
GrainWhether the relationship is one-to-one. Aggregating across a one-to-many link without accounting for it inflates every total, quietly and consistently.
Currency and calendarTwo systems can both be right and still disagree, because one reports in local currency on a fiscal calendar and the other does not.
What to do when they are missingDecline, and say which of the four is absent. A cross-source total produced without them is not an estimate. It is a number with no defined meaning.

How Nexatron approaches it

Nexatron federates across connected sources using an embedded DuckDB engine, so a question spanning a CRM and an ERP can be answered without first consolidating them into a warehouse.

The governance is the point rather than the speed. A cross-source question runs only where a certified entity link exists between the sources; where the link is absent, or where the aggregation grain cannot be shown to be safe, the platform declines and names what is missing instead of returning a number that would look reasonable and be wrong.

The question that has no home

The conventional answer is a pipeline: move both into a warehouse, model them, and ask there. That works, and it costs months. The question is whether it is required before anyone can get an answer.

What has to be agreed before a cross-source answer is trustworthy

A certified keyAn explicit, approved statement that this field in system A identifies the same entity as that field in system B. Matching on company name is a guess, and a guess that is right most of the time is worse than one that is obviously wrong.

GrainWhether the relationship is one-to-one. Aggregating across a one-to-many link without accounting for it inflates every total, quietly and consistently.

Currency and calendarTwo systems can both be right and still disagree, because one reports in local currency on a fiscal calendar and the other does not.

What to do when they are missingDecline, and say which of the four is absent. A cross-source total produced without them is not an estimate. It is a number with no defined meaning.

How Nexatron approaches it

Nexatron federates across connected sources using an embedded DuckDB engine, so a question spanning a CRM and an ERP can be answered without first consolidating them into a warehouse.