Support Signal Design Sprint
Sprint · Hosted by Mindline. Risk flags are produced faster than anyone reads them, sitting in a dashboard while the agent works from a silent queue.
About this open call
What this event is about
Churn flags land in a dashboard nobody has open, so the agent replying to the customer never sees the risk sitting on the ticket in front of them. The work is the delivery: the alert surface, what evidence it carries, and what it asks the agent to do next.
What we are looking for
Product and design teams who have shipped inside a support agent's workflow, with an alert surface that earns the interruption and records what happened after it.
Problem statements
The flag never reaches the conversation
Risk is raised on a live conversation and delivered to a screen the agent does not have open. By the time anyone reads it the ticket is closed and the reply that mattered has already gone out.
An interruption that has to earn itself
Agents work a queue against a handle-time target. An alert that arrives without the quote, the account and the reason it fired gets dismissed on sight, and a feed dismissed twice is never opened again.
Nothing records what happened next
When someone does act on a flag, the outcome is written nowhere. Nobody can say which signals were worth an interruption, so the threshold gets argued about instead of measured.
Evaluation criteria
100% weighted total
Timeline
- Jun 22, 2026
Entries open
The brief and a redacted week of support transcripts went out to everyone who joined.
- Aug 1, 2026
Entries read as they arrive
Every entry gets a first read within a week of landing. Nothing is ranked yet, so sending early costs you nothing.
- Sep 11, 2026
Shortlist
Three teams are shortlisted and walked through a live queue with two support leads.
- Sep 25, 2026
Pilot decision
One entry goes to the paid pilot; the rest get written notes on what did and did not land.
Sponsors & partners
Mindline
Host
Plug and Play
Programme partner
FAQs
Do entrants see real customer conversations?
No. You work from a redacted sample set — transcripts with names, accounts and identifiers stripped — and the live queue stays closed. Nothing here gives access to a real customer record.
Do we have to build against the scoring model?
No. Treat the signal as given: a conversation, a risk score and the phrases that raised it. What is judged is what the agent sees and does next, not how the score is produced.
Does a design-only entry count?
Yes. Annotated screens and a click-through prototype are both accepted, provided every state is there — no signal, a weak one, a false positive, and one an agent acts on.
Take part
Submit your entry before the deadline. Every entry is scored against the criteria above.