Selected work
Work with the numbers attached.
Each of these is a problem that was ambiguous when it landed and provable by the time it closed. The interesting part is usually the diagnosis, not the fix.
These are from eight years owning the data warehouse and reporting function of a consumer marketplace platform, in-house rather than as an outside consultant. The employer and the internal system names are withheld; the numbers are the real ones.
Analytics engineering
A seven-figure block of revenue with no category attached.
After a payment platform migration, seven figures of revenue was landing without category attributes. Tracing it found a data model gap the migration merely exposed.
- Seven figures Revenue correctly attributed
- Byte-identical dev/prod parity Verification standard
Data platform build
The metric had not collapsed. Its attribution had.
A key metric appeared to fall off a cliff and stakeholders feared a real drop in user activity. The events were still arriving — they had lost their attribution upstream.
- ~75% → ~100% Completeness restored
- Backfilled unsampled History
Product analytics
The aggregate metrics did not move. The harm was real anyway.
A moderation bug degraded how some listings were shown. Aggregate engagement looked untouched because demand shifted to substitutes, so impact was measured per item.
- ~9σ Statistical confidence
- One day earlier than reported True onset
Analytics engineering
There was pressure to make the number match. The number was right.
Transactions were missing from reporting and the fastest fix would have massively over-counted revenue. Proving where the gap was mattered more than closing it fast.
- Proven clean by exact reconciliation Transformation layer
- Unsafe fix rejected with evidence Over-count avoided
BI and reporting
A legacy reporting estate, moved without losing anyone’s trust.
Legacy reports migrated to self-serve BI, each validated side by side against the version it replaced, with data-quality fixes folded in along the way.
- Side-by-side against legacy, per report Validation
- Improved during migration Accuracy
Data platform build
The quota limit was the symptom. The absence of attribution was the problem.
Hitting a warehouse query quota threatened reporting. Mitigating it was an afternoon; the durable fix was making it possible to attribute cost to the thing that caused it.
- Per-application service accounts Attribution
- Prevented Recurrence
Data stack audit
Data quality was ad hoc and invisible. It needed a definition first.
A nine-dimension data quality strategy with KPIs and a communications plan, plus a repeatable AI-assisted assessment workflow any teammate can run.
- 9 Quality dimensions defined
- Repeatable by any teammate Assessment workflow
Got a number nobody trusts?
Tell me which metric it is and I will tell you how I would go about finding out why it is wrong.