Why Accessibility Is a Quality Dimension
A dataset can be complete, accurate, consistent, and timely — and still be inaccessible in practice. Accessibility measures the degree to which data is available to authorized users in a usable format when they need it.
Inaccessible data is data that effectively doesn't exist for the people who need it.
The Three Accessibility Barriers
Technical barriers:
- Data is in a system only IT can access
- Query performance is so slow the data isn't practically usable
- Data is stored in proprietary formats that require specific software
- No API or export capability for downstream systems
Knowledge barriers:
- Data exists but nobody knows it does
- Data dictionary is missing or outdated — users don't know what fields mean
- No documentation of where to find specific data
- Training requirements too high for non-technical users
Governance barriers:
- Permission structures that over-restrict access (data is available in principle but not in practice for most users)
- Request-and-approval workflows that take so long users work around them
- No self-service analytics capability — all reports must go through data team
Sohovi profiles your datasets for quality issues in minutes — see what's broken before it breaks your pipeline — try Sohovi free.
Measuring Accessibility
Unlike other dimensions, accessibility is harder to measure numerically. Indicators:
Time to data: How long does it take a user to get from a business question to data that answers it? (Benchmark: under 1 hour for routine queries, under 1 day for complex requests)
Ticket volume for data access requests: High volume indicates a governance or self-service problem
Data utilization rate: What % of available datasets are actively used? Low utilization often means data exists but isn't accessible or known about
User survey: Do stakeholders report that they can find and use the data they need when they need it?
Improving Accessibility
- Build a data catalog (Alation, DataHub, Google Data Catalog) — the index of what data exists and where
- Enable self-service analytics for common use cases (Looker, Tableau, Mode Analytics)
- Adopt a data mesh or federated ownership model that reduces central bottlenecks
- Review permission structures annually — over-permission where risk is low, protect where risk is high
