Product thinking / DIJION
Do you need to collect everything to measure a product?
A measurement plan starting with a product question, defining events and decision boundaries while excluding unnecessary user data.
Measurement tools can record many events; that does not make every event useful. Replace What are users doing? with a narrower decision question. For example, do a guide’s contents links help readers reach the relevant section? Let the information collected follow that question and the decision it supports.
Write the decision before the measurement
More clicks do not always mean a better experience. A person might click repeatedly because they cannot find what they need. Record the intended interpretation, competing explanations and missing evidence. Counting an event and confirming task completion are different kinds of evidence.
Scroll the table horizontally to see all columns.
| Question | Necessary signal | Information excluded |
|---|---|---|
| Are contents links used? | Use of the relevant section link | Identity or free-form text |
| Where do form errors occur? | A controlled error category | Name, email and message content |
| Can people find help? | Context in which help opens | Private user content |
Specify when an event occurs
Submission attempted and Submission confirmed by the server are different events. Do not record failed requests as success. If the same action is measured from both browser and server, investigate duplicate counting. Design and engineering should share the same event name and triggering condition.
Keep free-form input out of event parameters
Google Analytics guidance explains that personally identifiable information such as names and email addresses should not be sent to Analytics. URLs, search fields and error messages can also carry user content accidentally. Prefer predefined categories in this example plan rather than sending the entire form body.
Evaluate permission, access and retention against the actual tool configuration. This card is not a legal compliance document. Keep the absence of unmeasured or non-consenting sessions visible when interpreting reports.
Read the number in context
Traffic sources, devices or content can change alongside an interface update. A small sample does not justify a confident causal claim. Observation, feedback and measurement contribute different perspectives on the same question. Record the missing evidence as a further research question.
Before adding an event
- Which decision could it change?
- Can existing information answer the same question?
- Could parameters contain user content?
- Have deduplication, permission and retention behavior been verified?
This article does not report Dijion conversion or usage results. The useful output is not more data, but a short plan explaining which information will support which decision.