Mobile appDocumentation

Escalation, SLA & CSAT

Three safety nets: the agent recognising a conversation it shouldn't handle alone, a clock that notices when a customer has been waiting too long, and a rating that tells you whether any of it landed.

Escalation triggers#

Triggers are evaluated on the customer's message, in this order. The first match wins, and it happens before confidence is considered - a furious customer gets a human even if the agent had a perfect answer ready.

TriggerFires when
Negative sentimentThe conversation reads as negative or frustrated.
Human requestedThey asked for a person - "speak to a human", "transfer me to an agent", "real person", and the many ways people phrase it.
Sensitive topicRefunds, chargebacks, money back, cancellation, lawsuits or lawyers, complaints, GDPR and data deletion, fraud or scam accusations, and opt-out phrases like 'remove me' or 'stop contacting'.
Low confidenceNothing above matched, but the reply scored below the auto-send threshold.

You can extend the sensitive list with your own keywords.

What escalation does#

  • The reply is written but kept as a draft rather than sent.
  • The conversation is tagged as escalated, with the reason recorded.
  • It appears under the inbox's Escalated filter, and the draft shows the reason instead of a confidence number.
  • The workspace owner is notified in-app, by email, and by WhatsApp where that's configured.

Escalation flags, it doesn't freeze

The agent doesn't pause itself or change the conversation's ticket status. A human reviews the draft, edits or replaces it, and sends. If you want the agent to stop replying there entirely, take the conversation over.

Human takeover#

Taking over stops the agent from auto-sending in that conversation and marks the row so the rest of the team can see a person has it. Hand it back when you're done and the agent resumes. This is per conversation - agent-wide settings are untouched.

First-response SLA#

The clock starts when a customer's message arrives and stops when someone - agent or human - replies. The window is 60 minutes.

  • The clock only runs while the last message is inbound. Once you reply, it stops.
  • Resolved conversations - solved or closed - are out of scope entirely.
  • A breach flags the conversation and notifies the assignee.
  • Replying to a breached conversation clears the flag, so the inbox reflects reality rather than accumulating red badges.

SLA states#

StateMeaning
OKInside the window with time to spare.
Due soon15 minutes or less remaining - the point at which to act.
BreachedPast 60 minutes with no reply.

Adherence appears as SLA met in Analytics. A separate 24-hour resolution target exists for how long conversations take to close out.

Auto-send is your SLA strategy

A 60-minute target is easy when the agent answers in seconds and hard when every reply waits for a person. If adherence is poor, the honest fix is usually to let the agent auto-send on at least one channel - not to work faster.

CSAT#

Customers rate individual replies with thumbs up or down. Ratings are stored on a 1-5 scale, with thumbs recorded as 1 and 5, and can carry a written comment.

  • Ratings come from the surfaces that show the control - the website widget and help center. Email, WhatsApp and voice customers aren't prompted today, so CSAT reflects those surfaces rather than everyone.
  • Negative ratings feed the knowledge gaps list in Analytics, which is the fastest route from an unhappy customer to a fixed answer.
  • The score is per reply, not per conversation, so you can see exactly which answer disappointed someone.

Tuning it#

If everything escalates

  • Check your added keywords - a broad word like 'cancel' catches far more than cancellations.
  • Low confidence is the most common trigger. That's a knowledge problem, not an escalation problem.

If nothing escalates

  • Read a few auto-sent conversations end to end. Escalation only catches what customers say plainly.
  • Add the phrases your customers actually use when they're unhappy with you specifically.

If SLA looks great but customers don't

  • The clock measures a first response, not a good one. Read CSAT and knowledge gaps alongside it.
  • A fast unhelpful reply stops the clock and starts a complaint.
PreviousAuto-Send PolicyNextFollow-up Automation