Leave no trace (aka do no harm).
- Monitoring shouldn’t affect the system that's being monitored.
- "We must never negatively impact the systems we observe!"
- You must have a reasonable standard for how to define a "negative impact." Monitoring will always have some overhead, which makes it even more vital that such a standard be clearly defined and understood—and never violated thereafter.
- Medical doctors take the oath to do no harm in the course of their work ("primum non nocere"). A monitoring service should do the same.
Just because you can collect certain types of data doesn’t mean you should.
- When you collect data, you're implicitly saying it is important.
- Avoid unrelated data or vanity metrics at all cost.
- Users are communicating that they trust you when they grant you access to their systems; don’t violate that trust by collecting something you shouldn’t.
- And, of course, avoid collecting information that you aren’t allowed to have (e.g. don’t violate security/compliance laws).
Present data clearly and transparently.
- A confusing chart/metric can be just as harmful as an incorrect measurement.
- Users have a right to know what computations your monitoring platform has performed on their raw data prior to their viewing it.
- Your service should do the heavy lifting of isolating and presenting the data that's relevant to the user; if the user has to sift through data and determine what is and isn't valuable, the service should be more deliberate and clear in its presentation.
- Avoid superfluous information; focus your user’s attention on what is most important to them. A monitoring service and the people who design it should have strong opinions on what users should be paying attention to and on what is and is not important to measure. By using a monitoring product, the customers of the service implicitly endorse and invest in the practices the service has established. The monitoring service should, then, ideally, communicate and clarify these principles as part of the product's UX. An especially adept service should work to educate its user base on its opinions and positions in its marketing efforts and community presence.
- It’s important to understand the distinction between focusing your data and censoring the record. If you determine that something is of lesser importance but still potentially significant, the data should be available, just not emphasized.
Don’t interrupt your users' focus.
- Alerts demand attention, and superfluous alerts can quickly desensitize users to their importance. John Allspaw, former CTO at Etsy, has written extensively and very well on this subject, with articles such as his Open Letter to monitoring and alerting companies and his piece on understanding your effect on "users' attention," both worth reading closely.
- When you create interruptions, there must be no tolerance for false-positives/unactionable alerts.
- Users can choose to be alerted by something, but you should strive to guide them so that any alerts they customize still contain the most possible value.
Don’t lose data.
- Users rely on you for a record of what has happened inside their system. You need to preserve that record.
- Lost information is lost trust; without trust, a monitoring service is useless