Before / correction arrives
A corrected file overlaps an earlier submission. Appending both versions would count some sales twice.
Case study 01 / Data engineering / Business intelligence
From distributor files to repeatable sales reporting.
Distributor submissions arrived in different formats, and recurring reporting depended on manual data entry. I designed and deployed the path from file intake to Power BI, giving the Sales team more consistent, granular reporting.
Outcome / Reduced dependence on manual data entry and made recurring sales reporting more consistent.
- My contribution
- End-to-end owner · Design, implementation and deployment
- Primary users
- Sales team
- Project status
- Deployed
01 / The problem
B2B Sales Data Platform
Different channel formats, corrected submissions and rolling reporting windows made it easy to duplicate records or confuse the latest stock position with historical trends.
Separate intake paths for General Trade and Modern Trade converge into governed reporting structures. Raw files remain replayable, while shared transformations prepare sales and stock for analysis.
- Receive & retainDistributor submissions → n8n intake → raw S3 archive. A human submission step remains.
- Validate & transformAWS Glue normalizes dates and identifiers, checks columns and handles duplicate submissions.
- Model for analysisParquet and PostgreSQL structures separate sales, current stock and weekly snapshots.
- Report & operatePower BI reporting, data-health views, correction guidance and user enablement.
02 / See the work
A report that separates sales from stock
Recreated report concept with fictional values. Sales is a period measure; current stock is a point-in-time position.
B2B Sales Data Platform
Illustrative reporting view
Sales units · example week
Both bars use a 0–800 unit scale. These are illustrative quantities, not business results.
Stock position
320 units · latest snapshot
- Previous weekly snapshot280 units
- Change between snapshots+40 units
- Reporting meaningPoint in time
Stock snapshots are compared across dates; adding them would double-count inventory.
After / controlled reprocessing
Retain the original input, identify the affected reporting window and reload that scope using stable keys.
03 / Technical judgment
The decisions behind the result.
Keep raw inputs replayable
Preserving the submitted file and its metadata made corrections and troubleshooting traceable.
Centralize transformation
Thin intake workflows pass files to Glue, where validation and cleansing can be maintained consistently.
Respect reporting grain
Channel-aware identifiers avoid collisions. Current stock and weekly history have separate meanings and loading patterns.
Deliver the operating process
Submission templates, correction procedures, reference-data guidance and testing materials helped users operate the system.
View a simplified processing pattern
# Illustrative pseudocode — not production source
retain_original(submission)
rows = validate_and_normalize(submission)
keys = scope_identity_by_channel(rows)
reload_affected_sales_window(keys)
update_current_stock(rows)
write_weekly_snapshot(rows)04 / Ownership & impact
What I delivered, and what changed.
My work and collaboration
I owned ingestion, transformation, relational modeling, BI reporting, monitoring, infrastructure guidance, testing and user enablement.
The solution still relies on people to integrate and submit distributor data. I documented that handoff and the correction process as part of delivery.
Outcome
The deployed platform reduced manual data-entry dependence, improved reporting granularity and introduced repeatable duplicate handling. Business-impact figures are omitted because they are not approved for public use.
Qualitative outcomes and usage status reflect my project account. Illustrative values above are not impact metrics.