Platform
Putting an owner and a clock on every incident
Alerts without an assignee become a backlog. What a four stage case lifecycle looks like in practice, and how to set an SLA nobody games.
AI in the real worldKey takeaways
- An alert without an assignee is a backlog with better formatting.
- Four stages is enough: new, acknowledged, in progress, closed with a reason.
- Set the clock on acknowledgement, not on closure, or the target rewards guessing.
- Every close needs a reason code, and 'no action' must be one of them.
Detection is the easy half. The half that decides whether any of it mattered is what happens in the ninety seconds after the alert lands, and that is a process question wearing a technical costume.
Four stages, and no more
Every stage you add is a stage somebody has to be trained on and a column somebody has to fill in. Four covers the real lifecycle: it arrived, a person has seen it, a person is doing something about it, and it is finished with a reason attached.
| Stage | Owner | Clock |
|---|---|---|
| New | The queue | Running |
| Acknowledged | Named operator | Stopped |
| In progress | Named operator | Resolution target |
| Closed | Named operator | Not applicable |
Where to put the clock
Time the acknowledgement, not the closure. A target on closure teaches people to close things, and the cheapest way to close an incident is to decide it was nothing. A target on acknowledgement measures the thing you actually want, which is that somebody looked.
Closing with a reason
- Confirmed, actioned on site.
- Confirmed, no action required.
- False detection, camera or lighting.
- False detection, model.
- Duplicate of an existing case.
Split the two kinds of false detection. Merged, they produce a figure that condemns the model for a problem that is two degrees of camera tilt, and nobody can tell from the report which fix to buy.
Frequently asked
Should low severity alerts have an SLA at all?
They should have an owner and a review, not a clock. A target on something nobody is expected to drop what they are doing for is a target that trains people to ignore targets.
Who owns an incident overnight?
The shift, not the person. Assignment should follow the rota automatically, or the morning starts with a queue addressed to someone who went home at six.
Keep reading
All articles
Platform
The audit trail nobody asks for until they do
When someone asks what happened and who signed off, the answer should already be written down. What to log, and what to keep out of it.
Read article
Platform
Does a model trained on Terra hold up on your cameras?
The gap between a generated scene and a wet lens at four in the morning, measured rather than argued about, and what closes it.
Read articleDeployment
What a hundred cameras cost your uplink
Inference at the edge, clips to the centre. A worked bandwidth budget for a single site, and the two settings that halve it without losing evidence.
Read article
