Skip to content

Presenter demo script

Audience: Retail operations, analytics, and technology stakeholders
Duration: 15-20 minutes
Data: Synthetic

For a browser-first tour of the deployed workspace, use the deployed solution walkthrough.

For a standalone 5-7 minute persona walkthrough, use the presenter journeys for retail operations, merchandising, or executive/analytics audiences.

Before the audience arrives

  1. Complete the getting-started guide.
  2. Confirm the required KQL tables and ag/au Lakehouse tables exist.
  3. Open the Eventhouse query editor, Power BI report, and the streaming notebook.
  4. Decide whether dashboard, ontology, agent, and ML surfaces are ready; skip any surface that has not passed its support gate.
  5. Keep the operations guide open for recovery.

1. Frame the architecture

Open the architecture overview.

Talk track:

  • Setup notebooks create repeatable historical data. Silver is the cleaned, detailed layer; Gold is the business-ready summary layer.
  • stream-events.ipynb writes eighteen live event types directly to Eventhouse.
  • KQL (Kusto Query Language) answers recent event questions; Lakehouse stores durable history and summaries.
  • Power BI uses Direct Lake, which reads Lakehouse data without importing a second copy.
  • Optional ontology and agent surfaces add business context after their capability and binding checks pass.

2. Start live data

In stream-events.ipynb, select the Eventhouse sink and run a bounded stream. Show a recent row:

receipt_created
| where ingest_timestamp > ago(10m)
| take 10

Then show live sales aggregation:

mv_store_sales_minute
| where ts > ago(10m)
| summarize sales = sum(total_sales) by store_id
| order by sales desc

Suggested talk track:

Each generated business event says what happened and when. A unique trace identifier connects related receipt, line-item, promotion, and payment events, so we can follow one synthetic transaction across the process. The payload also identifies the store or order used to keep related events together. Eventhouse records when it received each event, which lets us prove that the optional stream is currently active.

3. Ask operational questions

Choose a supported path from the presenter journeys, or use checked-in queryset tabs for:

  • Recent sales and top products.
  • Inventory movements and stockout detections.
  • Omnichannel order creation and fulfillment events.
  • Store presence and zone dwell.

Be precise: recent stockout detections are not the same as unresolved current stockout state. Truck dwell is implemented and covered by cross-layer contract tests, but present it only after a bounded live run produces recent fn_truck_sla() rows in the selected workspace.

4. Show durable history

Run or show the last successful streaming-to-Silver/Gold pipeline. In the Lakehouse:

  • inspect ag.fact_receipts or another mapped fact;
  • inspect a Gold aggregate such as au.sales_minute_store;
  • show run or watermark evidence rather than relying on visual freshness alone.

5. Show Power BI

Open the Power BI Project (PBIP) report and demonstrate historical and operational pages backed by the Direct Lake model. Use explicit measures and current data periods.

Deployment saves each date filter to the month containing the configured history end date. State that month before showing a number. If the report was rerendered with a different history window, its default month changes with the data.

If required ML tables have not been generated and validated, do not present predictive pages as supported output.

6. Close with actions and roadmap

If a validated pricing approval or alert scenario is available, show it as an optional governed action. Otherwise, describe it from the owning backlog:

Do not overclaim

Avoid using these as headline proof points without the stated qualification:

  • Truck dwell behavior is owned by the event contract; do not present it as live-validated until a Fabric smoke run has been recorded.
  • Marketing attribution and return on ad spend (ROAS) are implemented and contract-tested. State the seven-day last-touch rule and selected data period before presenting a result.
  • Required ML publication has been exercised live and remains fail-closed. Still check the current readiness report before presenting predictions, because model evidence is time-bounded.