This section is for operations users and implementation teams who run the platform after engineering has deployed it. The goal is repeatability:
  • create bots the same way every time
  • test before calling real customers
  • upload clean campaign CSVs
  • monitor retries, callbacks, and failures
  • know when to escalate to engineering
Campaign dialling has three first-class concepts to keep in mind throughout this section:
  • Retries — failed contacts re-dialled after a delay.
  • Callbacks — customer-requested “call me back later” appointments, dialled with top priority once due (see Callbacks).
  • AMD — answering-machine detection that can tear down machine answers before they reach the bot (see AMD).

​
Ops responsibilities

AreaOps ownsEngineering owns
Bot setupBot name, opening message, prompt/policy, voice pipeline choices, tools, CRM mapping, test callsProvider keys, runtime bugs, transport failures
Campaign setupBot selection, outbound number pool, CSV upload, concurrency, time window, retry rulesDialler health, fleet capacity, SIP trunk health
MonitoringCampaign status, failed calls, attempt reports, obvious bad dataStuck states, infra failures, provider outages
CRM integrationAPI key handoff, variable names, post-call mapping checksEndpoint failures, auth bugs, webhook retries

​
Day-to-day flow

1

Prepare provider settings

Confirm system settings contain working STT, LLM, TTS, LiveKit, storage, carrier, and fleet configuration.
2

Build or clone a bot

Use the bot builder to configure identity, instructions, policy, voice pipeline, behavior, tools, knowledge, and CRM mapping.
3

Test the bot

Run a test call. Confirm the opening message, customer variables, response style, tools, recording, transcript, analysis, and CRM push.
4

Create a campaign

Choose the bot, outbound numbers, CSV contacts, time window, concurrency, and retry rules.
5

Start and monitor

Start the campaign, watch progress, inspect failed attempts, and download reports when needed.

​
Escalate to engineering when

  • /health/fleet shows no capacity but workers should be free.
  • Campaign calls stay in leased, ringing, attaching, dialling, or in_progress too long.
  • Many calls fail with the same SIP code or provider error.
  • Recordings are missing or playback returns access errors.
  • Bot config fetch fails for a valid active bot.
  • A CRM push is accepted in UI config but repeatedly fails at runtime.