01 · The break
Three systems that should agree, mostly didn't.
The integration spanned dealer data, vendor data, and platform data. On a good month the three systems agreed. On most months reconciliation was a multi-day manual chase to figure out where the numbers diverged. The break was not always in the same place: duplicates, missing fields, and cross-system inconsistencies all surfaced in different parts of the pipeline, often only after leadership asked why a metric did not add up.
The team needed two things at once: a way to surface the breaks earlier, and a way to show leadership reconciliation health without forcing them to read three reports.
02 · The work
Catalog the breaks. Catch them at ingest. Show leadership one view.
Walked the pipeline end-to-end with SQL, tracing root causes of each class of failure. Reconciliation problems usually look like one big ugly thing; they are actually small, repeated breaks at specific join points, and they cannot be fixed in aggregate. Cataloged each break, named it, and wrote the smallest possible check that would catch it the next time.
Built validation checks at every join point: duplicate detection, missing-field alarms, cross-system consistency rules. The checks now run on every batch, so the team catches breaks at ingest instead of discovering them weeks later when leadership asks why the numbers do not add up.
Translated the findings into Tableau dashboards designed for leadership reading rather than for analysts. One view showing reconciliation health by vendor, by week, with the failure modes broken out so a non-technical reader can see where the issue lives without asking. Informed process changes around how the team handled the upstream sources so fewer breaks reach the integration in the first place.
03 · Outcome
Fewer mismatches, less manual chase, one view of the truth.
Reference
"We'd been living with reconciliation problems for a year. Katherine spent the first month inside our SQL, found every place the systems disagreed, and built validation that catches the breaks at ingest. Mismatches are down 25%, our manual reporting effort is down 40%, and leadership reads one Tableau view now instead of three. She works through the boring details most people skip."Scott Anderson · CEO, InDesign Data
04 · Behind the work
Notes from the engagement.
The hardest part was not the SQL. It was getting comfortable that the answer to "why don't these numbers match" was going to be a long, boring list of small specific reasons, not one elegant root cause. Once the catalog existed, the checks practically wrote themselves.
The Tableau view mattered more than expected. Leadership had been receiving reconciliation context as tables-inside-emails. The dashboard gave them a single place to look and a vocabulary for the failure modes, which changed what they asked the team about and made process-level fixes negotiable.