Demand dashboard: DAX, visualization, validation & findings

Live Power BI dashboard

Published report walmart_visualization — Executive, Store, Product, and Demand pages on Gold aggregates. Open in Power BI ↗

Analysis pages: 08 Executive · 09 Store · 10 Product · 11 Demand · Overview

Part A — DAX measures used on Demand (Page 4)

Full catalogue: 08 Executive · Part A.

Daily Units =
SUM ( 'agg_demand_trends'[daily_units] )

Daily Sales Value (Demand) =
SUM ( 'agg_demand_trends'[daily_sales_value] )

Rolling 7D Units =
SUM ( 'agg_demand_trends'[rolling_7d_units] )

Rolling 7D Avg Units =
AVERAGE ( 'agg_demand_trends'[rolling_7d_avg_units] )

/* Use Average — never Sum across all days for a KPI card */
Rolling 28D Avg Units =
AVERAGE ( 'agg_demand_trends'[rolling_28d_units] )

Event Units =
SUM ( 'agg_event_sales'[units_sold] )

Event Sales Value =
SUM ( 'agg_event_sales'[sales_value] )

Part B — Dashboard 4 screenshot (updated)

Latest Power BI DEMAND page: period total card labeled, 7D/28D averages, units line, event columns, sales combo, and event table with title and no bogus grand-total row.

1. Scope & method

2. Visualization & interaction

2.1 Layout

┌──────────────────────────────────────────────────────────────┐
│ [ year ] [ month ]   [ Total Units (period) ] [ 7D Avg ] [ 28D Avg ] │
│ [ Units line: daily vs rolling 7D ] [ Event by cat ]                 │
│ [ Sales $ vs 7D avg (combo) ] [ Event vs non-event table ]           │
└──────────────────────────────────────────────────────────────┘

2.2 Visual-to-field map

2.3 Interaction

• Year / month slicers filter demand trends and KPIs via dim_date.
• Event visuals are from an unconnected agg — they do not always shrink with the year slicer unless you relate or filter them explicitly.
• Sync year with Executive / Store for a consistent planning window.
• Click-path: set one full year → read 7D/28D cards → scan unit trend → check event vs non-event rates in the table.

3. Key pattern findings

3.1 Demand rises on a daily basis

Mean rolling 7-day average units ≈ 34.4K; mean rolling 28-day unit total ≈ 957K. Both trend upward across 2011–2016 on the daily charts. The KPI card is correctly labeled Total Units (period) (67M = sum of daily units in filter), distinct from the ~34K rolling 7-day average.

3.2 Most volume is on non-event days (expected)

Absolute event bars look small because there are far fewer event days. Fairer lens — units per day by category:

In this M5 calendar definition, labeled event days are slightly weaker on average units/day than non-event days (holidays / closures like Christmas pull the event average down). Use the table’s avg_units_per_row and per-day rates — not only stacked totals.

3.3 Active SKUs is not a trend metric

active_skus is 3,049 every day (full catalog in the long fact). It was correctly removed from the time combo; do not bring it back as a line series.

3.4 FOODS still dominates event and non-event volume

On both event and non-event days, FOODS is the largest unit contributor — consistent with Executive / Product pages.

4. Dashboard quality notes

• ✓ Rolling 28D card uses Average (956.66K), not Sum (2bn).
• ✓ KPI renamed to Total Units (period) — clearer than “Daily Units”.
• ✓ Event table titled “Event vs non-event by category”; bogus Total row (5823 days / 5.76 sum of averages) removed.
• ✓ Active SKUs removed from time visuals (constant 3,049 by design).
• ! Left line: still pairs daily_units (~35K) with rolling_7d_units (~250K), so daily looks flat near the bottom. Prefer rolling_7d_avg_units (Average) as the companion series.
• ! Combo chart: ensure Secondary y-axis On for Rolling 7D Avg Units; if both series share the $ axis, the units line sits too low under the sales area.
• ! Optional: split combo into sales area + units line for maximum clarity.

5. Recommendations

1. Switch left line companion to rolling_7d_avg_units (Average).
2. Confirm combo secondary Y-axis for units (or split into two charts).
3. Interpret events with units/day, not only absolute stacks.
4. Keep forecasting focus on FOODS baseline demand + holiday zero days.
5. Sync year slicer with Executive.

6. Reproduce

python - <<'PY'
import pandas as pd
from pathlib import Path
P = Path('data/gold_export')
d = pd.read_parquet(P/'agg_demand_trends.parquet')
e = pd.read_parquet(P/'agg_event_sales.parquet')
print(d['daily_units'].sum(), d['rolling_7d_avg_units'].mean(), d['rolling_28d_units'].mean())
print(e.groupby('event_label')['units_sold'].sum())
PY

Screenshot: assets/demand-dashboard-page4.png. Related: 08 · 09 · 10.