Deliver to Farid

Step 1: Draft the narrative

The analysis is done. Now it needs to reach Farid in a form he can act on.

Direct AI to draft a narrative that follows the analytical thread -- what was found, in what order, and what it means for his five requirements. The structure should follow the investigation: combined data picture first, then channel trends, then product profitability, then discontinuation candidates, then data quality issues.

Review AI's draft. AI commonly buries the most important finding behind less important ones. If the key insight is "retail is declining 15% while online grows 40% -- but online represents only 30% of revenue with a different cost structure," that belongs at the top, not in paragraph seven.

AI also overclaims certainty. "Revenue will continue to decline" is a prediction the data doesn't support. "Revenue declined 15% over the fourteen-month period" is what the data shows. Check every claim against what the numbers actually say.

Watch for technical language Farid wouldn't understand. "The GROUP BY aggregation at the channel level reveals a divergent trend" needs to become "your retail shop revenue dropped while your online sales grew -- but the online growth isn't enough to offset the retail decline."

Step 2: Address each requirement

Go through Farid's five requirements one by one. For each, point to the specific section and numbers in your narrative:

1. Combined data picture: How many records total? What cleaning decisions were made? What assumptions are documented?
2. Product profitability by channel: Which products are profitable in which channels? What are the margins?
3. Channel trends: Which channels are growing, declining, or flat? Over what period?
4. Discontinuation candidates: Which products should Farid consider dropping, and why? Which require channel-specific decisions rather than wholesale discontinuation?
5. Data quality issues: What inconsistencies were found? What should Farid fix in his systems?

If a requirement can't be fully addressed because the data doesn't support it, say so. Farid asked for "channel trends over the two years" but the data covers fourteen months. That limitation should be stated clearly.

Step 3: Write the data quality documentation

Farid's fifth requirement is the one most likely to get thin. The student who focuses on analysis often forgets that Farid explicitly asked for a record of what's wrong with his data.

Write a data quality section that documents:

• Product name inconsistencies: How many variant names existed across the three systems? How were they reconciled? What mapping was used?
• Wholesale data structure: The items column required parsing. What assumptions were made about quantities and pricing?
• Null values: Where were they, how many, and what was done about them?
• Date format discrepancies: Three different formats across three sources.
• Missing information: No customer identifiers in retail. No wholesale tier information. No discount amounts online.

Each issue should include a recommendation for Farid: what to fix in his systems and why. "Standardise product names across POS, Shopify, and wholesale records -- use the catalog names as the single source of truth. This would eliminate the reconciliation step entirely."

Step 4: Review the deliverable

Before sending, review the full narrative against the metric definitions. Every number in the deliverable should be traceable to a specific metric definition. If the narrative says "wholesale revenue was MYR 145,000," check: which revenue definition produced that number? Does it match the plain language version?

Check the charts embedded in the narrative. Do they have clear titles, labels, and axis descriptions? Can Farid understand each chart without reading the surrounding text?

Check the edge cases from Unit 3. Did the narrative handle them consistently? If your definition excludes pending wholesale invoices, does the wholesale revenue number in the narrative reflect that exclusion?

Step 5: Push and deliver

Commit your work and push to GitHub. The repository should contain: the unified dataset, the metric definitions document, the cleaning log, the analysis notebook with charts, and the narrative deliverable.

Open the chat with Farid. Send a summary of the analysis along with the key findings for each of his requirements.

Farid reads methodically. He'll confirm that the channel trend matches what he suspected -- retail declining, online growing. He'll appreciate the data quality documentation and say it gives him a clear picture of what to fix in his systems.

He'll also ask about something you may not have highlighted: the wastage difference between his Sabah Gold blend and the seasonal blends. He hadn't considered that different products have very different shelf lives, and the cost of inventory waste could be significant. This is information that enriches the analysis -- a question the data raised that Farid hadn't thought to ask.