Self-service analytics: why it keeps failing, and what changed
5 min readUpdated
Self-service analytics has repeatedly stalled because it moved the difficulty rather than removing it: dashboards required someone to anticipate the question, drag-and-drop tools required users to understand the underlying data model, and both left the definitional disagreements unresolved. Natural language removes the interface barrier, but it only works where the definitions and permissions are handled underneath. Otherwise it produces fluent answers that disagree with each other.
Three waves, three bottlenecks
Each wave removed a real barrier and exposed the next one. Dashboards removed the need to write a query and revealed that the hard part was anticipating the question. Self-service tools removed the need to wait for a dashboard and revealed that the hard part was knowing what the fields meant.
| Wave | The promise | Where it stalled |
|---|---|---|
| Dashboards | Answers at a glance | Someone had to predict the question in advance |
| Drag-and-drop tools | Build your own view | Users had to understand the data model to be right |
| Natural language | Just ask | Fluent answers that quietly disagree with each other |
What is genuinely different now
The interface barrier really has fallen. A person can ask a question in their own words and get a structured query out the other end, and that was not true five years ago.
What has not changed is that the underlying disagreements are still there. If two departments define revenue differently, a language interface will confidently give you both, on different days, with no indication that anything is inconsistent. The technology removed the typing, not the arguing.
The condition for this wave working
- The definitions are settled and storedOnce, by someone accountable, in a form the tool reads every time rather than re-deciding per question.
- Access is automatic and per personThe moment self-service means everyone sees everything, it stops being rolled out.
- Answers can be checkedNot because everyone will check, but because the ones that matter will be, and a tool that cannot be audited never graduates from pilot.
- The tool admits its limitsThe failure that ends adoption is not "it could not answer". It is one wrong answer, presented confidently, discovered in a board meeting.