A Stale 92% Scorecard on the Executive Dashboard

No. The 92% is a historical snapshot, not a current measure of Customer 360 quality.

Staleness The profile ran six weeks ago on 40 million rows; the table now has 62 million, a 55% increase. Uniqueness and completeness can change materially in that volume growth. The duplicate surge Operations reports is exactly the kind of change a six-week-old profile would miss.

Sampling and null handling The uniqueness rule samples 1% and excludes nulls. A 1% sample may not represent a 22-million-row increase, especially if duplicates cluster in new segments. Excluding nulls inflates uniqueness: a column can look unique among non-null values while many records are missing the key entirely.

Coverage The new source has never been profiled, so any quality problems there are absent from the score. The scorecard’s 12 domains may not include that source at all, which understates enterprise risk.

Execution gap The rules are external and loaded as custom attributes. CDGC is displaying results, not computing them. If the external job failed or was not rerun, the attribute values may be stale even if the metadata scan is current.

The score can be shown only with an explicit “as of six weeks ago” label and a caveat, not as tomorrow’s quality position.

Before tomorrow

• Relabel the dashboard with the profile date and row count: “92%, as of [date], 40M rows; current table 62M rows.”
• Add a data freshness indicator and a known-issue note about the duplicate surge and the unprofiled source.
• Re-run the external uniqueness and completeness rules on the current table if the job can be triggered overnight. Even a targeted run on the customer key is better than no refresh.
• Show the score as “last measured” rather than “current,” and suppress green status.

Longer fix

• Align profiling frequency with business criticality: daily or weekly for Customer 360, not monthly.
• Increase the uniqueness sample above 1% for large tables, or use a full-key scan where feasible.
• Stop excluding nulls from uniqueness without also reporting null rate; a null customer key is a quality defect, not a neutral exclusion.
• Profile the new source and add it to the domain coverage.
• Version the score and store the rule logic, sample size, and run date with each result so the number is reproducible.
• Define a freshness SLA: a score older than the SLA is shown as “stale,” not as a valid percentage.

Priority The labeling and targeted rerun address the immediate credibility risk. The sampling and coverage changes address whether the score is meaningful at all. Doing only the relabel is honest but not sufficient; doing only a full redesign misses tomorrow’s deadline.

Use a three-part message: what is measured, what has changed, and what action is underway.

1. What is measured. “The 92% is the Customer 360 quality score as of six weeks ago, based on a 1% uniqueness sample that excludes nulls. It covers the primary table at 40 million rows.” This makes the number interpretable rather than absolute.
2. What has changed. “Since then the table has grown to 62 million rows, Operations has reported a duplicate surge, and a new source is not yet profiled. The score should be treated as historical, not current.” This validates the operations signal instead of contradicting it.
3. What action is underway. “We are rerunning the customer-key rules tonight, adding the new source to coverage, and will publish a dated score with sample size and null rate. Until then, the dashboard will show the score as stale.”

Why not hide it Removing the scorecard creates a vacuum and invites unmanaged anecdote. Why not present it as current Presenting 92% without caveats would contradict Operations and damage the CDO’s credibility when duplicates surface in executive reporting. The honest framing preserves the scorecard’s value as a trend indicator while making clear that it is not a real-time guarantee.