Sample AI Data Readiness Report

Mask all 6 PII columns immediately: credit_card (stg_payments), email/first_name/last_name (customers), ip_address (events), shipping_address (orders). Use BigQuery dynamic data masking before any AI pipeline or analyst touches this estate.

0 of 6 PII columns are currently masked. Any AI training job, feature store export, or portfolio-triage dashboard that reads these tables exposes credit card and personal identity data in plaintext, triggering PCI-DSS scope and GDPR Article 25 violations. This single gap can block the entire AI investment program pending a regulatory audit.

-- BigQuery masks via Data Catalog policy tags. Create a taxonomy + a PII policy tag
-- with a data-masking rule, then attach the tag to each unmasked column:
ALTER TABLE `acme-analytics.warehouse.customers`
  ALTER COLUMN email SET OPTIONS (policy_tags = ['<your-pii-policy-tag-resource-name>']);
ALTER TABLE `acme-analytics.warehouse.customers`
  ALTER COLUMN first_name SET OPTIONS (policy_tags = ['<your-pii-policy-tag-resource-name>']);
ALTER TABLE `acme-analytics.warehouse.customers`
  ALTER COLUMN last_name SET OPTIONS (policy_tags = ['<your-pii-policy-tag-resource-name>']);
ALTER TABLE `acme-analytics.warehouse.events`
  ALTER COLUMN ip_address SET OPTIONS (policy_tags = ['<your-pii-policy-tag-resource-name>']);
ALTER TABLE `acme-analytics.warehouse.orders`
  ALTER COLUMN shipping_address SET OPTIONS (policy_tags = ['<your-pii-policy-tag-resource-name>']);
ALTER TABLE `acme-analytics.warehouse.stg_payments`
  ALTER COLUMN credit_card SET OPTIONS (policy_tags = ['<your-pii-policy-tag-resource-name>']);

Apply BigQuery policy tags to all 31 columns (currently 0/31 tagged), starting with the 6 confirmed PII columns, to enable column-level access control and data-lineage tracking across AI pipelines.

Without classification, every AI consumer (model training, feature pipelines, BI dashboards) has undifferentiated access to sensitive fields. The compliant factor scores 0; closing this gap is prerequisite to demonstrating EU AI Act Article 10 data-management practices and to any enterprise data-sharing agreement that depends on this estate.

-- Attach a sensitivity label to each table (repeat per table):
ALTER TABLE `acme-analytics.warehouse.<table>` SET OPTIONS (labels = [('data_classification', 'pii')]);

Document the 16 undocumented columns (currently 15/31 documented, 48% vs. 50% threshold). Prioritize columns in the customers, orders, and events tables that will feed AI feature pipelines first.

Sixteen ambiguous columns in a portfolio-triage AI model will produce mis-labeled features with no visible error: the model may appear to run while producing meaningless signals. This is also a direct gap against EU AI Act Article 10's requirement for documented data-management practices for high-risk AI.

-- Add a description to each undocumented column (repeat per column):
ALTER TABLE `acme-analytics.warehouse.<table>`
  ALTER COLUMN <column> SET OPTIONS (description = 'what this column holds and its unit/meaning');

Partition or cluster the 4 unoptimized tables (only 1/5 currently partitioned/clustered) to control query costs before AI workloads scale scan volume.

At 1/5 tables optimized, full-table scans are the default access pattern. As AI feature pipelines and training jobs run repeated large queries, BigQuery on-demand costs will scale linearly with data volume, a predictable budget overrun that will surface in the first sprint of any serious AI build-out on this estate.

Declare an identifier for the 1 remaining unidentified table (4/5 tables have identifiers) to complete entity-resolution coverage across the estate.

The missing identifier on one table prevents reliable entity joins in AI feature construction. For a portfolio-triage workload, this means one table's signals cannot be linked to customer or order entities, creating a blind spot in any cross-entity investment signal the model tries to generate.