System flows
Five of Plugboard’s AI actions start through flows. They are installed on, marked Built in on the Flows list, and you can open each one to see exactly when it fires, add conditions to narrow it, watch it, or turn it off.
The five system flows
Section titled “The five system flows”| Flow | When it starts | What it does | Feature in Admin, AI |
|---|---|---|---|
| Sort new tickets | A ticket is raised | AI triage suggests or sets the kind, category, priority, campus, device and assignee | Triage |
| Send new tickets to the right queue | A ticket is raised, and again when its category changes | AI routing suggests or sets the queue or desk for a ticket with no queue | Ticket routing |
| Resolve on the spot | Someone raises a ticket they have already fixed and marks it resolved on the spot | Resolves it, with what they did as an internal note | Fill in new tickets and repairs |
| Suggest a device fix | Someone asks for a suggestion on a ticket | AI chooses at most one fix from the device fix catalogue for the ticket’s device; a person approves it before anything runs | Device fixes |
| Spot access requests | A minute after a ticket is raised | Looks for a request for a group, licence or routine item and offers it on the ticket | Access requests |
Tickets a flow raises are sorted and routed as well, because these flows can be started by changes other flows make.
Where they come from
Section titled “Where they come from”Each school gets the five once, installed on for the whole-school desk, published as version 1 and owned by the school’s first active whole-school administrator. The install is in the audit log. A system flow you turn off, change or archive is never installed again, so your choice holds.
The same five are in the template gallery. A desk that adds its own copy gets that copy deciding for its tickets.
What the flow controls, and what the feature still does
Section titled “What the flow controls, and what the feature still does”The flow decides when a feature fires. The feature still decides what happens, using its own mode, thresholds, AI limits and accuracy record in AI settings.
| You do this to the flow | The feature on that desk |
|---|---|
| Leave it on | Fires as it always has, in the mode set in Admin, AI |
| Add conditions, such as “Category is not Printing” | Fires only for tickets that match |
| Publish it as Watching | Does not fire; each run records what it would have done |
| Turn it off, or it is paused | Does not fire |
| Pause all flows | Does not fire, along with every other flow |
| Archive it | Fires as it did without a flow, governed only by its mode |
Both have to say yes. A flow that is on does nothing for a feature set to Off, and a feature set to Auto does nothing while its flow is off. Turning the Automation module off returns every feature to firing without a flow.
System flows are not counted towards the hourly limits or the storm pause, which these features never had. The school’s AI limits bound them.
What stays a feature
Section titled “What stays a feature”Composer autofill (apart from resolving on the spot), reply drafts in the composer and the ticket brief are features only. You manage them in AI settings and on the composer; there is no flow behind them.
The line in Admin, AI
Section titled “The line in Admin, AI”In Admin, AI, under Features, each feature behind a system flow says which flow runs it and the flow’s state (On, Watching, Off, Paused), with Open flow. For Fill in new tickets and repairs the line names only resolving on the spot, because only that part is a flow.
Learning from your own answers
Section titled “Learning from your own answers”Routing and access request recognition learn from your desk’s own corrections, with no training run. When either is asked a question, the model is shown a few recent tickets that a person on the same desk decided, under the answer the person chose: at most three per answer and twelve in all, from the last 180 days. Only explicit human answers count, never an automatic one. The examples are masked and wrapped like the rest of the decision, and a desk never learns from tickets it cannot see.
Triage, school tags and flow questions do not take examples this way. Flow questions learn through their own record; see decision models.
Repairs and documents are also indexed for similar-record search when they are saved. The cost of this background learning shows on AI activity.
For the weekly list of repeated desk work that could become a flow, see offers and flow ideas.