Skip to main content

​
Inbox coaching: a supervisor playbook for digital conversations

Inbox → Coaching is the supervisor surface for digital threads — SMS, WhatsApp, email, and the other channels your team works in the inbox. It lists the conversations the QM judge scored below your threshold, lets you review the thread in context, and carries the feedback loop forward: coaching notes on the thread, the agent’s acknowledgement, the agent’s score trend, and a before/after read on their digital KPIs. This guide covers when to use this surface instead of voice coaching, how to review and flag a thread, how to run a weekly coaching loop, and how the loop feeds your QA reports. No screenshots needed — the queue is short enough to describe by its parts, and the guide names every part as you will meet it on the page.

​
1. When to use Inbox → Coaching vs voice supervisor coaching

Both surfaces close the same loop — a score flags a problem, a supervisor responds to it, and a report later tells you whether the response worked. They differ in what they review and how the feedback is delivered:
  • Inbox → Coaching (this guide) — triages conversations: multi-message threads across digital channels, scored by the QM judge after close. Feedback is a coaching note attached to the thread that the agent reads and acknowledges asynchronously. The scorecard reports per-rubric pass rates, a daily score trend, and top failure modes. Effectiveness is measured on digital KPIs — first-response time, resolution rate, and CSAT.
  • Voice → Coaching Plans and Coaching Templates — the voice supervisor loop. It reviews calls, and feedback is a coaching plan with a goal, a due date, and a close-out outcome, often written mid-call as a whisper or post-call note with snippets from the template library. Effectiveness is measured on voice KPIs — average handle time, first-contact resolution, and post-call CSAT. That loop is covered in Supervisor coaching: plans, templates, and effectiveness reporting.
The rule of thumb: the channel that produced the score decides the surface. A low-scoring WhatsApp thread belongs in the inbox coaching queue; a mishandled call belongs in a coaching plan. Both queues appear in Inbox → Coaching (the channel filter shows All / Inbox / Voice rows), but a voice row deep-links into the call surface — the coaching-plan workflow still happens over on the voice side. One scope note: this surface coaches the quality of conversations, not the regulation of them — rubrics about tone, accuracy, and resolution behaviour. Compliance controls themselves (quiet hours, DNC, consent) are workspace-owned settings documented under the compliance guides, not something a coaching note enforces.

​
2. Review a thread and attach feedback

Open Inbox → Coaching. The page needs an owner, admin, or supervisor role; agents never see the queue.
  1. Set the threshold. The queue shows conversations whose average rubric score falls below the threshold — default under 50%, with stricter (30%) and lenient (70%) options. All scored turns the filter off and lists every scored conversation with at least one failed rubric, lowest score first.
  2. Pick the channel. Filter to Inbox for digital threads. Each row is prefixed with a channel glyph (I for inbox, V for voice) so a mixed queue stays legible.
  3. Read the row before opening it. Each row shows the contact, the average score as a percentage, the failed/total rubric count (for example 2/5 means two of five rubrics failed), and when the conversation was last scored. Scan the rows first — a conversation failing four of five rubrics is a different review than one failing a single rubric.
  4. Open the conversation. Clicking a row deep-links into the thread on the inbox conversation pane. Read the thread end-to-end before writing anything: which replies actually failed the rubric, and what the customer saw around them.
  5. Attach a coaching note. Post a note against the conversation. The note has three parts: the free-form feedback, optional references to the specific failed rubrics, and an optional suggested action. Notes are per-reply feedback — you address them against the specific exchanges that failed, not a generic season summary. A note addressed to an agent who was not assigned to the conversation is refused (a 422), which keeps feedback pinned to the right owner.
  6. Close the loop. The agent sees the note as addressed to them and acknowledges it with one action; the acknowledgement stamp is visible back on the note. Supervisors, owners, and admins see every note on a conversation — an agent sees only their own.
The same endpoints exist over the API for scripted review workflows:

​
3. Build the coaching loop: sample → feedback → track

A coaching queue is only useful if it empties and the scores move. The loop is:
  1. Sample. Scores come from your QA program’s sampling — the weekly sampler assigns a random quota of each agent’s conversations for scoring, and scored conversations below the threshold land in the coaching queue. If the queue is empty and you expected traffic, your sampling is the knob to turn, not the threshold. See QA sampling: weekly sample assignment.
  2. Feedback. Work the queue lowest-first. Write a coaching note per thread, referencing the failed rubrics so the note is evidence rather than opinion. Small, specific notes weekly outperform a quarterly write-up — the acknowledgement stamp gives you a hard read receipt.
  3. Track the agent’s trend. Pull the agent’s scorecard — per-rubric pass rates, a daily average-score trend over up to 90 days, and the top failure modes:
Read the trend against your own notes. If you have coached “verify before refunding” for three weeks and rub_verify_identity still sits under the other rubrics’ pass rates, the observation matters more than another note — change the intervention (a macro fix, a pairing session) rather than repeating it.
  1. Measure whether it worked. Judging a coaching loop on note volume is wishful; the effectiveness report compares the agent’s real digital KPIs in a symmetric window before and after the coaching event:
Each KPI reports the before value, the after value, the delta, and a direction (improved / regressed / flat / insufficient_data): response time (lower is better), resolution rate (higher is better), and CSAT (higher is better). The report anchors on your explicit anchor_at, or on the agent’s most recent coaching note if you omit it — a window that has not fully elapsed yet is flagged in_progress so you do not read a partial span as a verdict. An agent with no coaching notes and no anchor returns a 404: there is no intervention to measure against. Agents can read their own scorecard and effectiveness report; owners, admins, and supervisors read anyone’s. A rubric for steady notes — the pattern that makes feedback land: Keep notes to one issue each. A note bundling three complaints gets acknowledged and forgotten; three acknowledged single-issue notes leave a trail the scorecard can actually confirm.

​
4. Wire it into QA

The inbox coaching loop and the QA program read from the same scoring pipeline; the coaching surface is one of its consumers:
  • Upstream — sampling and scoring. The coaching queue exists because your evaluation forms and the QM judge score conversations. The forms, the grading workflow, and the acknowledge/appeal lifecycle live under Quality → Evaluations — see QA Evaluations: scorecards and the appeal lifecycle.
  • Downstream — the public read surface. If your workspace has the quality leaderboard turned on, the scoring the coaching queue consumes also feeds team-wide ranking — the place agents see where coaching has moved them relative to the team. The rules, badges, and visibility scoping are in Quality leaderboard: scoring rules, badges, and thresholds.
  • Parallel surface — the voice loop. Voice rows in the coaching queue are triage; the coaching itself happens as plans and template-backed notes per Supervisor coaching.
The program shape is: sampling feeds scoring, scoring feeds both the supervision queue and the leaderboard, and the coaching note is how a supervisor intervenes between the two.

​
See also