Free first step
Mini Validation
- 10 pipeline outputs you choose
- Old and new job output compared, plus a re-check after fixes
- Findings within 10 business days after we are connected
- A fixed price for the parallel run, or a plan to run it yourself
The tables your old jobs produce, matched by the new ones, run after run.
Rebuilding SSIS packages, Informatica PowerCenter mappings, DataStage or Talend jobs, or years of stored procedures and SQL scripts as ELT in dbt, Coalesce, Databricks, or Snowflake? We run the old and new logic side by side on the same inputs and compare what each one produces, table by table and run by run, until your owners are ready to retire the old jobs.
Free first step
Rewritten logic is close, not identical. These are the places it drifts.
Which row a lookup returns when there are two matches. What a NULL does inside a concatenation. Whether trailing spaces get trimmed, and which rows go to a reject file. Rewritten SQL makes its own choices unless someone captured the old ones.
An SSIS data flow or a stored procedure converts strings to dates and decimals to integers without being asked. dbt models and Snowflake procedures cast differently, so rounding, truncation, and padding shift values that a row count never notices.
Deduplication that keeps the first row depends on whatever sort order the old engine happened to use. Joins, window functions, and row-by-row updates in stored procedures can pick different survivors once the same logic runs set-based.
A full load usually matches. The differences show up later: an incremental window that skips late-arriving records, a job that starts before its upstream load finishes, a history row that closes a day early, a delete that never propagates.
Mappings and procedures written fifteen years ago, by people who have left, hold the business rules. Comparing outputs is often the only practical way to prove the rewrite kept them, and to catch the rules someone changed on purpose.
We test what each job produces, not the code. The old logic might live in SSIS, Informatica, DataStage, Talend, Oracle Data Integrator, or stored procedures, and the new logic in dbt models, Coalesce nodes, Databricks jobs and notebooks, or Snowflake tasks and procedures. Either way the results land as tables, files, or extracts, and that is where we compare them. No connector to the ETL tool is needed.
| Layer | What we check |
|---|---|
| Full-load output | The first full run of the new jobs produces the same rows and values as the old ones |
| Every scheduled run | Each incremental or scheduled run matches the old job's run on the same inputs and load window |
| Transformations and business rules | Derived columns, lookups, aggregations, and rule-driven flags match, including NULLs, blanks, padding, and rounding |
| History behavior | Slowly changing dimensions open and close records on the same dates, with the same versions |
| Late, corrected, and deleted records | Late-arriving and corrected records, deletes, and rejected rows are handled the same way on both sides |
| Row level | Every column, matched by key, on the outputs your owners name as critical |
Planned changes go into the acceptance rules up front, so the results show only what needs a decision.
Ab Initio, SAP Data Services, scheduled Python, or something home-grown? The method is the same: we compare outputs. Tell us in your Mini Validation request →
Testing ETL jobs that are not moving? See ETL Testing & Monitoring →
Five steps, from a free Mini Validation to the signed evidence pack.
Every Migration Validation Program has a fixed price, set after the Mini Validation. The example shows where a common scope lands. The estimator gives you a figure for your own numbers in about a minute.
Typical starting estimate
From $10,000
Up to 25 pipeline outputs, row-by-row matching up to 100 million rows, 3 validation cycles, and 3 months of platform access. The Data Agent is included.
Example: SSIS to dbt
About $20,000
60 pipeline outputs, a 4-week parallel run, 4 months of access.
About 6 to 9 weeks
Estimate your price → Above $150,000, a sales executive scopes the work with you.
Short answers to what migration leads ask first.
Run the old and new jobs side by side on the same inputs and compare the tables they produce, not the code: a full-load comparison first, then every scheduled run through the parallel run, with transformations, history, deletes, and rejected rows checked. Logic you changed on purpose goes into the acceptance rules up front.
Any tool whose results land somewhere Validatar can read. Common moves are SSIS, Informatica PowerCenter, DataStage, Talend, Oracle Data Integrator, and stored procedures rebuilt in dbt, Coalesce, Databricks, or Snowflake.
No. We test what the code produces. dbt models, Coalesce nodes, Databricks jobs, and Snowflake procedures all write tables, and that is where we compare them.
Procedures are often the hardest logic to read, which makes output comparison the practical test. We run the old procedure and its replacement against the same inputs and compare the rows each one writes, updates, or deletes.
Long enough to cover month-end and any weekly or monthly jobs, usually four weeks. One validation cycle covers a week of scheduled runs.
The Migration Validation Program starts at $10,000 and is priced by scope. For example, 60 SSIS outputs over a 4-week parallel run comes to an estimated $20,000. Your fixed price comes after the Mini Validation.
Start with a free Mini Validation on 10 pipeline outputs. We'll compare what the old and new jobs produce, show you every difference, and give you a fixed price to finish the parallel run.
Free, with no obligation to go further.
Still planning your migration? Get the Migration Validation Checklist →