Agent health

Midwater rates each agent from its recent conversations, separately in the test and live environments. Read it with GET /v1/agents/{agent_id}/health, or a whole group with GET /v1/groups/{group_id}/health.

curl "$MIDWATER_BASE_URL/v1/agents/front-desk/health" \
  -H "Authorization: Bearer $MIDWATER_API_KEY"

{
  "agent_id": "front-desk",
  "name": "Front desk",
  "environment": "live",
  "health_status": "watch",
  "reason": "Resolution 81% vs 87% 30-day baseline (−6 pts) since version 3.2.0",
  "group": { "id": "brightsmile-dental", "name": "BrightSmile Dental" },
  "last_7_days": {
    "conversations": 212,
    "with_outcome": 180,
    "resolution_rate": 0.81,
    "handed_to_person_rate": 0.09,
    "requests_for_person_not_honored": 2,
    "avg_frustration": 0.22,
    "compliance_failures": 0
  },
  "last_30_days": {
    "conversations": 905,
    "with_outcome": 774,
    "resolution_rate": 0.87,
    "handed_to_person_rate": 0.08,
    "requests_for_person_not_honored": 5,
    "avg_frustration": 0.19,
    "compliance_failures": 1
  }
}

Statuses

health_status	Shown as	When
healthy	Healthy	None of the rules below apply.
watch	Watch	The 7-day resolution rate is 5 to 10 points below the 30-day rate, or there was at least 1 compliance failure in the last 7 days.
at_risk	At risk	The 7-day resolution rate is more than 10 points below the 30-day rate, or there were more than 2 compliance failures in the last 7 days.
not_enough_calls	Not enough calls yet	Fewer than 5 conversations with an outcome in the last 7 days, so nothing is measured yet.

Below 5 conversations with an outcome, one thing still counts: high-severity compliance failures. One in the last 7 days makes the agent watch, two or more at_risk, and the reason names the failure. Otherwise it’s not_enough_calls, with a reason such as “No calls yet” or “Needs 4 more calls with an outcome this week”. The window numbers are still returned; they just don’t set the status.

reason is one sentence saying why. It’s written for people: show it, don’t parse it.

Windows

last_7_days and last_30_days are counted in whole days in your workspace’s time zone, by each conversation’s ended_at.

Field
conversations	Conversations received.
with_outcome	Conversations with an outcome, other than “Not a customer call”. The denominator of the rates.
resolution_rate	Share of those that were resolved, 0 to 1, or null with none.
handed_to_person_rate	Share handed to a person: a transfer, or a person requested and provided.
requests_for_person_not_honored	Conversations where the caller asked for a person and didn’t get one.
avg_frustration	Average caller frustration, 0 (calm) to 1 (very frustrated), or null.
compliance_failures	Compliance checks that failed.

Groups

A group’s health_status is the worst among its active agents, and reason names the agent that set it. A group is not_enough_calls only when every agent is. agents lists each agent with its own status, and the windows are totals across them.

Changes as they happen

Instead of polling, subscribe to the agent.health_changed and group.health_changed webhooks. They fire when watch or at_risk is the old or the new status; becoming measured as healthy, or going quiet, sends nothing.

Agent IDs use only letters, digits, ., _, : and -, so they go into the URL as they are. Unknown IDs answer 404.