M365Marketing - Schema
======================

Synthetic multi-channel marketing campaign performance dataset. Four tables joined on campaign_id, channel_id, and segment_id. 78 weeks (18 months) ending 2026-05-31.

Tables
------

channels.csv (5 rows)
---------------------

Column              Type    Notes
------------------  ------  --------------------------------------------------------------------------------------------------------------------------------------------
channel_id          string  Primary key.
channel_name        string  Display name.
attribution_window  string  Known imperfection: different channels use different attribution windows. Reconciling conversions across them is a chapter exercise (Ch 11).

segments.csv (10 rows)
----------------------

Column                 Type     Notes
---------------------  -------  ----------------------------------------------
segment_id             string   Primary key.
segment_name           string   Display name.
industry               string   One of 8 industries.
size_band              string   Enterprise+ or Mid-Market.
tam_estimate_accounts  integer  Total addressable market size for the segment.

campaigns.csv (24 rows)
-----------------------

Column              Type     Notes
------------------  -------  -------------------------------------------------------------------
campaign_id         string   Primary key. Format CMP-NNNN.
campaign_name       string   Synthetic name.
channel_id          string   FK -> channels.channel_id.
segment_id          string   FK -> segments.segment_id.
attribution_window  string   Denormalized from the channel (reflects real warehouse design).
weekly_budget_usd   integer  Planned weekly spend; actual varies in performance.csv.
status              string   active, paused, or ended. Ended campaigns have no performance rows.

performance.csv (~1,800 rows)
-----------------------------

Column         Type        Notes
-------------  ----------  --------------------------------------------------------------------------------------------------------------------------------
week_starting  date (ISO)  First day of the ISO week (Monday).
campaign_id    string      FK -> campaigns.campaign_id.
impressions    integer     0 for paused weeks.
clicks         integer     0 for paused weeks.
conversions    integer     0 for paused weeks.
spend_usd      integer     0 for paused weeks.
note           string      paused, platform-outage, or empty. Diagnostic: 3 campaigns have a platform-outage week - useful for anomaly-detection exercises.

Suggested chapter exercises
---------------------------

- Chapter 3 (story): Build the Monday-morning reallocation narrative - which campaigns should grow, hold, or shrink? Pair with the M365Marketing starter BRD.
- Chapter 5 (focus and decluttering): Strip the cross-channel performance dashboard to a single answer per channel.
- Chapter 10 (AI visuals - anomaly detection): Find the three platform-outage weeks. Compare anomaly-detection sensitivity at different thresholds.
- Chapter 11 (prep for AI): Reconcile the attribution-window differences across channels. This is the canonical exercise for "AI data schemas" - the conversion metric is the same name across channels but means different things.
- Chapter 12 (dashboard design): Compose the weekly portfolio dashboard for the Director of Product Marketing.

Reproducibility
---------------

Generated by scripts/datasets/generate-m365marketing.js using a seeded PRNG. Source seed: 0x4D6B7400.