Why three tiers

Data quality coverage model · about 2 minutes

Enterprise data quality program

Why we use three tiers

How we cover every critical data element (CDE) without writing a rule for each one.

Where we started

Policy asked for one rule per critical data element

30,000 rules needed, each one built and maintained with a steward
325 rules in place today Rules the policy would still need

It gets worse

The CDE tag follows the data into every copy

CDECRM applicationCustomer enters their emailOwn rules, own steward
CDELake raw zoneDaily load, untouchedOwn rules, own steward
CDELake curated zoneCleaned and joinedOwn rules, own steward
CDEPower BI dashboardWhere people read itOwn rules, own steward

If every tagged copy needs its own checks, one customer email becomes four sets of rules. The workload grows every time data moves.

The insight

Not all data carries equal risk

Wrong value has consequencesCustomer name in a regulatory submission

Regulatory, financial or customer-facing impact when it fails.

Wrong value is an inconvenienceInternal product code in one team's report

Worth watching, not worth a steward's weekly attention.

One rule per CDE spends the same effort on both. So we stopped counting rules and matched the depth of control to the risk.

The model

Automation covers everything. Judgment goes where the risk is.

Tier 1 · Curated rulesHighest-risk CDEs only

Steward-set thresholds, a named owner and a remediation path.

Tier 2 · Automated completenessEvery CDE

One line of configuration per CDE. No rule-building, and the policy is met.

Tier 3 · ObservabilityEvery CDE

Learns what normal looks like for each field and flags unusual change.

Tiers 2 and 3 apply to every CDE automatically. The only placement decision is which CDEs earn Tier 1, typically 10 to 20 percent.

Why Tier 1 stays small

A curated rule costs more than it looks

Steward timeA strong rule takes expert hours

Thresholds, valid values and a remediation path all come from a steward. Their time is the bottleneck.

Reporting noiseMore results, less attention

Every Tier 1 failure goes to a named steward. With too many, the ones that matter get lost.

Processing costHeavier checks cost real money

A curated rule tests several dimensions per CDE. That compute lands on the bottom line.

Finding the balance is the job. Too few CDEs in Tier 1 leaves high-risk failures without an owner. Too many overwhelms the people who fix them.

Why all three

Each tier catches what the others miss

Tier 1 · CuratedUNKNOWN typed into a regulatory field. The field is filled, so only a validity rule with an owner catches it.
Tier 2 · CompletenessA field that has gone blank, on any CDE, including the ones nobody had time to write a rule for.
Tier 3 · ObservabilityA null rate drifting from 1.1% to 4.2%. No threshold was crossed, but something changed.

The authoritative data source

Check once at the source. Every copy inherits the result.

CDECRM applicationAuthoritative data sourceAll three tiers run here
CDELake raw zoneConsumption pointTransfer check only
CDEMarketing curated zoneInherits the result
CDECampaign dashboardsDisplay the result
CDEFinance curated zoneInherits the result
CDEBilling dashboardsDisplay the result
CDEService curated zoneInherits the result
CDEService dashboardsDisplay the result

and every curated zone added later

The tag travels. The obligation does not. One raw zone element is copied into many curated zones and many dashboards, but the rules run once. For a vendor feed we cannot reach, the source is the raw landing zone.

Placement is not permanent

CDEs move up and down as risk changes

Tier 1 · curated rules
Tiers 2 and 3 only
Customer phone ↑ Internal code ↓
PromoteRepeated completeness failures or recurring anomalies show the risk is higher than first judged.
DemoteNo regulatory, financial or customer-facing risk. Stepping down frees steward time for CDEs that need it.

Tiers 2 and 3 keep running either way. Stewards and data owners make the call, because they know the business consequence.

The point

Rule count is not a measure of maturity

  • Three tiers give every CDE baseline coverage and give the riskiest ones real accountability.
  • The authoritative data source means each CDE is checked once, not once per copy.
  • Promotion and demotion keep steward effort pointed at current risk.
Policy met

Every CDE has documented coverage that fits its risk.

Stewards in control

Their time goes to the CDEs that matter, with alerts they can act on.

What we measure instead: coverage quality, actionable alerts and the capacity to fix what they find.