All use cases

Prevent / Care

Negative-mood self-help routing

A user taps "bad" on today's mood check and the app just colours the square red — the agent turns that low moment into something they can actually do about it.

Next best thingGet aheadBehavioral triggerRisk triggerIn-appPushChatWellness (B2C apps)

What it does

Mood-tracking apps capture the hard days but mostly leave them there: a user logs "meh" or "bad," the chart records another low point, and nothing happens at the exact moment a small intervention would land best. A logged low mood that is not a crisis is still the moment a low-grade slide either gets met or quietly compounds into disengagement and churn. The agent reads a sub-threshold negative mood as a cue to surface the right coping tool in-thread — a short CBT reframe, a breathing exercise, the app's own first-aid kit — so the user moves from logging a feeling to taking a small action. This is the layer below the crisis rail by design: it handles the everyday low, and the instant anything reads as crisis it hands straight to the crisis pathway.

How it works

  1. 1
    Trigger. a mood log lands in the negative-but-sub-clinical range (a "bad" or "meh" entry, a dip from the user's own recent baseline) without any crisis language.
  2. 2
    Decision. the agent confirms the signal is below the crisis threshold, reads recent context from memory (what helped this user last time, what they are working on), and picks the single most fitting coping tool rather than a generic feed; every unprompted surfacing passes a quality and tone judge.
  3. 3
    Action. it offers one tool in-thread — a CBT reframe prompt, a two-minute breathing or grounding exercise, or a direct link into the app's first-aid kit — framed as an invitation, not a prescription, on the surface the user is already on.
  4. 4
    Follow-through. the suggestion and whether it was taken write back to memory so the next low day builds on what worked; if the conversation surfaces self-harm or crisis language at any point, the agent stops self-help routing and hands to the crisis pathway.
Example
AgentI saw today landed on the rougher end — sorry it's a hard one. Want a quick reset? I can walk you through a two-minute breathing exercise, or pull up the reframe that helped last time you logged a low day. Either is one tap, no pressure.
Userthe breathing one i guess
AgentLet's do it — find a comfortable spot and I'll pace it with you. Afterwards I'll check whether it took the edge off, and we can note what works for next time.

Configuration

How the agent is wired for this use case.

Triggera mood-log event in the negative sub-clinical band (mood_logged low, or a dip below the user's recent baseline) raised by the app, with no crisis language present.
Tools & actions
  • Mood classifier · place the logged mood on the user's own scale and confirm it is below the crisis threshold before anything is offered.
  • App backend / user memory · recall what coping tool helped this user before and what goal they are working on, to pick the most fitting one.
  • Knowledge base / content library · retrieve the right CBT reframe, breathing exercise, or first-aid-kit entry for this moment.
  • In-app surface · offer one judge-gated tool in-thread as an invitation and open it on a tap.
  • App backend · write the suggestion and the user's response back to memory for the next low-mood touch.
Autonomymood classification, tool selection, and the single in-thread offer run unattended under judge gating and frequency caps. The agent surfaces self-help content only and gives no therapy, diagnosis, or clinical advice; under a wellness-safety policy class, any self-harm or crisis language is detect-and-escalate to the crisis pathway, handled by card #256, never by this flow.
Channelsin-app · push · chat
Escalationany crisis or self-harm language, a sustained worsening pattern across logs, or an explicit request for human help hands off to the crisis-detection pathway with full context rather than being met with a coping tool.

What you need

The inputs this use case runs on. Your channels stay yours; the agent supplies the judgment.

Signals

mood-log events with a value, a baseline to compare against, and user replies; crisis screening running alongside on every message.

Data

the user's mood history and baseline, what coping tools they have engaged with before, the content library, consent state, channel preference, and persistent memory.

Guardrails

self-help content surfacing only — no therapy, diagnosis, or clinical advice; a hard threshold line where anything at or above crisis goes to the crisis pathway (#256), not to a tool; judge gating and tone-scoring on every unprompted message; frequency caps and quiet hours so a low day is met once, not nagged; visible opt-out; full memory write-back.

Metrics it moves

  • dau-mauup: a logged low mood becomes an in-app action instead of a dead-end red square, giving the user a reason to stay in the session.
  • adherenceup: meeting the hard days with the right tool keeps the daily logging-and-coping loop the app depends on alive.
  • save-rateup: low-mood days are a leading churn signal, and turning them into a helpful moment heads off the quiet drop-off before it starts.

Related use cases

Prevent / Care

Crisis-language detection and safe handoff

the crisis rail above this; everything at or past the crisis line routes there instead

Engage / Lifecycle

Daily mood check-in companion

the logging ritual that produces the mood signal this routing acts on

Prevent / Care

In-conversation de-escalation and recovery

the related tone-shift pattern for a frustrated customer, where this is a low-mood self-help cue

See it on your own customer journey

Bring one drop-off, one churn cliff, or one silent segment. We will show you what a proactive agent with memory and judgment does with it.

Book a demo