76% of the vehicle database loads itself
- BRP
- Global Digital Transformation Specialist
- 2026
Staff role, not a client engagement
Technical data for every vehicle in the European lineup was retyped into the database by hand, because no two source files arrived in the same shape. About 76% of it now loads itself.
At a glance
76%
of the database populated automatically
75%
less effort on everything that is left
100%
accuracy on every field the automation writes
The situation
Before a page can be built, every vehicle needs its technical data in the database. That data arrives as spec files and taxonomy files produced upstream, and nothing forces them into a shape: columns move between one file and the next, fields are missing, a value is a number in one place and text in another. So it was read and retyped, field by field, model by model, for the whole European lineup. North America had already been through this. Europe had more sources and less agreement between them.
What I did
Absorb the variation instead of assuming it away
The automation normalizes whatever arrives, matches it against the taxonomy, and writes only what it can be certain of. That turns out to be about 76% of the database, filled with nobody touching it, and the effort on the remaining quarter fell by about three quarters as well.
What it writes, it writes correctly. The corrections that used to come back from UAT were almost all transcription, and transcription is the part that is gone.
Then make the mess visible
A validation script reads the extraction, taxonomy and spec files before anything is loaded, and reports what changed: a schema that moved, data that isn't there, a layout that broke, formatting that drifted. It writes a log I can hand to the team that owns those files, so a problem gets fixed where it was created rather than patched where it landed.
That log is the smaller piece of work and probably the more durable one. Nobody can enforce a standard that has no evidence behind it. The log is the evidence, the first version of a rule you can point at when it is broken.
Results
- About 76% of the database populated automatically from the spec and taxonomy files
- 75% less effort than filling it by hand
- 100% accuracy on the fields the automation writes
- Fewer corrections coming back from UAT
- Input problems surface as a log the upstream team can act on, before anything loads
Skills used
- Data Pipelines & Automation
- Governance & Standards
Systems: Python, Adobe Experience Manager (AEM)
