Detect the third contact before your customer gives up.
Support agent sup-v3 was closing tickets on the first touch and seeing them reopen within 48 hours. Flowlines surfaced the sentiment pattern three sessions before churn and traced it to a missing context field.
The problem
The agent handled every new ticket as if it were the first. Returning users with two prior escalations got the same welcome template as a free-tier sign-up. Sentiment declined across three sessions; nothing in the pipeline looked at the trajectory.
Flowlines watched for a week and found the pattern: 67% of accounts that churned had three consecutive sessions with declining sentiment, and 84% of those sessions had sentiment_trajectory available but not injected. The field was being written and ignored.
The fix
Flowlines drafted a two-line change: inject sentiment_trajectory and issue_history into the support prompt, and route to a senior agent when the rolling score crossed a threshold. The engineer approved behind a flag. Repeat-contact rate fell 23% over the next 400 tickets.
What it catches now
Repeat contact within a 7-day window, same issue category. Sentiment decline across a rolling 5-session window. Resolution failure: tickets marked closed that reopen within 48 hours. Each correlates to a structured memory field and a specific routing decision.
What the team writes
Two lines in the agent entrypoint. Flowlines watches every ticket, the team reviews a Monday digest of drafted routing and memory fixes, and anything they approve ships as a flag-gated change to the injection view.
Structured memory fields
The typed fields Flowlines reads and writes for this domain. Each field is scoped, versioned, and traceable back to the interaction that produced it.
sentiment_trajectoryissue_historytier_rulesaccount_contextresolution_state