Ale CabreraSHIPPED · CLEARBIT · 2023 · 6 MONTHS
Data Guardrails
Preventive guardrails in the data pipeline, moving quality from fixing records after a customer complained to catching whole classes of errors before delivery.
- 70%fewer customer-reported data issues, measured over 3 months
- 1 → 5attributes covered by guardrails, starting with employee count
- ~15validation rules, grown from a handful as quarantine surfaced new issues
Customer-reported data issues
Bars show the relative volume. Before = 100.
Rollout: one attribute first
Each new rule came from an issue found in quarantine.
Where a value is checked now
- New value
- Validation rules
new, before delivery - Production
- Customer
Fails a rule: the value waits in quarantine until it is validated.
Rejected: kept in a "deleted" table for investigation.
Two risks, and what happened
| Risk | How it was handled | Result |
|---|---|---|
| Value. Customers were finding the errors before the company did. | Classes of errors defined from support tickets, SQL, dashboards and calls. Rules applied before delivery. | 70% fewer customer-reported data issues in 3 months. |
| Feasibility. Where to hold a suspicious value without slowing delivery. | A quarantine step. One attribute and a few rules first, then more as issues showed up. | All 5 attributes covered, around 15 rules. |
My role. Defined quality from the customer's perspective. Analyzed the failure patterns, wrote the guardrail design, and decided which attribute to start with and which rules to add next.
The insight. A wrong value can be worse than no value. A global consumer brand was listed as a 10-employee company. Coverage alone was the wrong goal.
Product leader with a data scientist past · Zapier, Clearbit, HubSpot
Product doc: ale-cabrera.com/portfolio/guardrails