Executive reporting
and sales forecasting.
In my role at Pro-Racing, I organize inventory, monitor sales performance, and prepare weekly reports directly for the CEO. I built this independent demonstration to show the reporting and forecasting process while keeping company records confidential.
Sales, stock, orders, and the weeks ahead.
A clearer starting point for the weekly review.
I connected channel forecasts from KNIME with sales and inventory reporting in Power BI, bringing current performance and four weeks of expected unit demand into one executive report.
EXPLORE THE FINISHED REPORT ↓- weeks of synthetic history
- 156
- sales channels
- 2
- weeks forecast ahead
- 4
- measure throughout
- Units
Tools and their roles
Current reporting work and the portfolio demonstration.
ExcelCurrent reporting work
SQLCurrent data preparation
Power BIExecutive dashboard
KNIMEPreparation & forecasting
About the evidence. This case study uses synthetic data created for portfolio demonstration. Figures and model results do not represent actual company performance. The finished report demonstrates my analytical work; it is not presented as a dashboard deployed to the CEO.
Four questions for the weekly review.
What sold this week, how much stock is on hand, what wholesale orders were placed this month, and what demand is expected next?
The report puts those questions together. Weekly channel trends distinguish steady online demand from uneven wholesale volume, while inventory by category provides a current stock view. The forecast adds a forward-looking planning window.
Read the date first.
The snapshot is as of 11 Oct 2026, the latest sales date in the demonstration. The next-week forecast of 4,722 units combines both channels for the week starting 12 Oct 2026.
Keep the measures distinct.
60,901 units is shared inventory at the latest snapshot. It stays unchanged when a sales channel is selected. The report shows zero wholesale ordered units month to date: an October shipment does not turn a September order into an October order.
See the four-week forecast values
| Week starting | Online retail | Wholesale |
|---|---|---|
| 12 Oct 2026 | 4,450 | 272 |
| 19 Oct 2026 | 4,465 | 8,567 |
| 26 Oct 2026 | 4,474 | 288 |
| 02 Nov 2026 | 4,454 | 11,038 |
Forecasts are weekly totals by channel. They are not predictions for individual products or customers.
From operational records to a planning view.
I used KNIME to prepare weekly channel data, compare regression approaches, and produce the forecast export used in Power BI.
This demonstration also supports my transition from Alteryx to KNIME. Excel and SQL remain part of my current reporting work; the workflow here shows the completed forecasting exercise separately from that ongoing company transition.
Inside the KNIME workflow
The original workflow keeps the preparation, model experiments, forecasting branches, and backtest outputs visible. The overview shows how they connect; expand the detail views to inspect the final model and recursive forecast stages.
Inspect the final regression branches
Inspect the recursive forecast branches and export
The diagram and node settings were inspected from the supplied .knwf file. The workflow was not re-executed. Detailed views are crops of the original vector export, not recreated diagrams.
- 01
Prepare each channel’s history.
Daily sales were aggregated to weekly unit totals and split into online retail and wholesale branches. Chronological sorting kept each channel’s lag values tied to its own history.
- 02
Compare models in time order.
I compared three linear-regression candidates using lag and time features. The saved final learners use 12 predictors: week index, four sales lags, annual and trend sine/cosine terms, promotion days, and two wholesale shipment-day indicators. Usable weeks through 130 formed the training period, weeks 131–143 supported validation, and weeks 144–156 supported testing.
- 03
Carry the forecast forward.
After fitting on usable historical weeks 5–156, the recursive workflow predicted the next week, shifted the lag values, and used the new prediction to advance again. Repeating this step produced weeks 157–160 for each channel.
- 04
Connect the export to the report.
The combined CSV contains eight rows: four weeks for each of two channels. Power BI brings that export together with Sales, Inventory, Orders, Products, Accounts, Channels, and Calendar. The displayed forecast uses nonnegative units; clipping occurs after recursive feedback.
Channel filters connect to both sales and forecasts. Product filters support operational analysis, but the channel-level forecast is not allocated to individual SKUs.
Evaluate the horizon, not just the fit.
I reviewed actual-versus-predicted plots for the later weeks and checked a separate four-week recursive forecast from origin week 130.
The two views answer different questions. One-step evaluation uses observed prior-week values; recursive forecasting feeds predictions into later steps. Results from one should not be treated as performance for the other.
KNIME Numeric Scorer output for the testing branch, using synthetic data.
KNIME Numeric Scorer output for the testing branch, using synthetic data.
View the online retail validation plot · weeks 131–143
Four-week recursive check
| Channel | Actual units | Total absolute error, units | WAPE |
|---|---|---|---|
| Online retail | 8,608 | 145.799 | 1.69% |
| Wholesale | 18,362 | 883.773 | 4.81% |
Values checked against the supplied final WAPE export. WAPE = total absolute error ÷ total actual units × 100. The wholesale row reflects the raw forecast treatment, not the separate clipped result. These figures are error percentages, not R² or “forecast accuracy.”
This is a small four-week check on synthetic data, not a broad rolling backtest. The synthetic data was revised during development, so these scores demonstrate the evaluation process rather than establish real-world predictive performance.
What this work makes possible.
A consistent executive review
Sales, stock, wholesale orders, and upcoming demand appear together, with an explicit data date and unit-based measures.
Channel-specific planning
Online and wholesale patterns remain visible, helping frame stock reviews ahead of higher-volume wholesale weeks.
Current inventory visibility
The report uses the latest inventory snapshot rather than adding stock balances across historical dates.
A repeatable forecast handoff
KNIME outputs a structured CSV that Power BI can consume, connecting model development to a four-week reporting view.
Scope. The demonstration does not establish time savings, financial gains, stockout reductions, or real-company forecasting performance. Forecasts are by channel, not product. Scheduled refresh and end-to-end production deployment have not been demonstrated.
The dashboard image and KNIME charts document the presented results. The PBIX and KNIME workflow have not been independently re-executed for this case study.