You're building out a data quality strategy and you're realizing that some of what you do catches problems before they happen, while other parts find problems that already exist. This isn't a coincidence — it reflects two distinct approaches to data quality.
Preventive data quality stops problems before they enter a system or dataset. Detective data quality finds problems after they're already in. Think of prevention as the lock on the door; detection as the security camera inside.
What Preventive Data Quality Looks Like
Prevention happens at the point of data creation or entry. Common examples:
- Constrained input forms — dropdown menus, required fields, format masks that make it structurally impossible to enter invalid data
- API schema validation — rejecting records at ingestion that don't conform to defined structure or allowed values
- Pre-import file validation — checking a CSV or Excel file against quality rules before loading it into a database
- Data contracts with suppliers — agreeing upfront on what quality standards data must meet before you accept it
- Training and process — teaching teams who enter data to follow consistent standards
Sohovi lets you set up validation rules for any column and instantly see which rows fall outside them — no code or SQL required.
Prevention's advantage is leverage: one decision made at the design stage eliminates an entire category of errors before they accumulate.
What Detective Data Quality Looks Like
Detection happens after data is already in a system. Common examples:
- Data profiling — scanning an existing dataset to discover what quality problems are present
- Quality monitoring — watching defined metrics over time and alerting when they degrade
- Audits and reconciliation — comparing datasets against each other or against an authoritative source
- Anomaly detection — flagging records that deviate from expected patterns
- Reporting and dashboarding — surfacing quality scores and trends for review
Sohovi runs a full data profile on any CSV or spreadsheet in under a minute — completeness rates, type distributions, outliers, and more.
Detection is reactive. It finds what prevention missed. For many organizations — especially those dealing with legacy data, external data sources, or data they didn't create — detective work is unavoidable.
The Real Tradeoff: Cost vs. Control
Prevention is cheaper per problem resolved, but requires you to control the point of data entry. If you don't control the source — you're receiving data from a partner, an inherited system, or a form you can't change — prevention isn't an option for that data stream.
Detection is more expensive per problem (each error must be found, diagnosed, and corrected after the fact) but works regardless of where data came from.
Sohovi gives you the data quality picture you need to make the case for fixing it — and to track improvement over time.
The principle most quality practitioners follow: apply prevention everywhere you have control over data entry, and deploy detection as a backstop for everything you can't prevent.
The Cost Argument for Prevention
Errors caught at the point of entry cost almost nothing to fix — the person entering the data simply corrects the value. Errors caught downstream may have already been used to generate reports, trigger automated processes, or inform decisions. The cost of fixing a downstream error includes not just the correction but undoing its effects.
This cost differential is significant and well-established in practice — which is why strong quality programs invest heavily in preventing errors at their source, even when detection is easier to implement quickly.
When Detective Work Is Most Valuable
Detective quality controls are especially valuable for:
- Auditing data you've inherited or imported from external sources
- Monitoring pipelines where small, gradual changes can have outsized impact
- Discovering novel quality problems you didn't know to prevent
- Regulatory compliance, where you need documented evidence of quality checks
- Root cause analysis — tracing where a detected problem originated
Frequently Asked Questions
Q: Which approach is more common in practice? Most organizations use both, but detective approaches are more commonly the starting point — because you're often reacting to a quality problem you've discovered rather than designing from scratch. Prevention is more commonly adopted as a mature quality program develops standards and governance.
Q: Is it possible to achieve pure prevention — eliminating the need for detection entirely? No. Even the most constrained systems allow some errors through — people find workarounds for mandatory fields, external data sources bypass your controls, and novel error types emerge that existing rules don't cover. Detection remains necessary as a backstop even in mature quality programs.
Q: How does data governance relate to preventive data quality? Data governance provides the organizational framework — ownership, standards, accountability — that makes prevention possible at scale. Without governance, prevention efforts are ad hoc and inconsistent. Governance defines what "correct" looks like; prevention enforces it.
Q: What is "data quality by design" and is it the same as preventive quality? "Data quality by design" means building quality constraints into systems from the start rather than adding them later. It's the architectural manifestation of preventive thinking — making good data quality the default outcome of how systems are designed, not an afterthought.
Q: How do you prioritize prevention vs. detection when you have limited resources? Prioritize prevention for high-volume, recurring data entry where the cumulative cost of errors is significant. Prioritize detection for data you don't control at the source, or for auditing historical data that already exists. Don't invest in detection for data streams where prevention is feasible — you're paying to find problems you could have stopped.
Q: What role do data stewards play in preventive quality? Data stewards own the quality standards for specific data domains. In prevention, they define what rules should be enforced at entry points. In detection, they triage alerts and investigate anomalies. Prevention without stewardship is fragile — rules degrade if no one maintains them as business requirements change.
Q: Is manual review a preventive or detective control? It depends on when it happens. Manual review before data enters a system (reviewing a file before import) is preventive. Manual review after data is in a system (spot-checking records in a database) is detective. The timing determines the category.
Q: What is a "data quality gate" and how does it combine prevention and detection? A data quality gate is a checkpoint that data must pass before moving to the next stage of a pipeline. It combines elements of both: detective checks identify problems (flagging or scoring records), and the gate prevents problematic data from advancing until issues are resolved. It's a staged-prevention model.
Q: How does preventive data quality apply to AI/ML training data? Preventive quality in ML contexts means enforcing schema validation, completeness requirements, and label consistency at data collection time — before training data enters a pipeline. Poor-quality training data that's detected only after a model is trained is expensive to fix; prevention eliminates entire categories of model degradation.
Q: Can small teams implement preventive data quality without dedicated tooling? Yes. Many effective prevention measures are low-tech: using dropdowns instead of free text in spreadsheets, adding required-field indicators in forms, using data validation in Excel or Google Sheets, and establishing team norms about how data should be entered. Prevention doesn't require a dedicated platform to start.
Prevention is more efficient where you have control; detection is necessary where you don't. The strongest data quality programs invest in both — designing prevention into their data entry and collection processes, and maintaining detective monitoring for everything downstream.
For organizations starting with detective quality work on data they've already collected, Sohovi makes profiling and scoring a dataset fast and accessible — upload your file and get a full quality report in under a minute, with no technical setup required.