What makes it a call agent
Creating one under Calls → AI agents switches everything on for you: the vault and transcription spokes, GHL, the Brain, every call skill your agency has, your MCPs and your integrations. A half-configured call agent is the classic failure, it writes a confident report and saves nothing, because the tool that persists an analysis was never attached.
It carries every sweep skill, GHL, JustCall, VAPI and Retell. Which one runs is decided by the integration you have connected, not by what you ticked at creation. A sweep whose tools are absent says so and stops rather than guessing at call data, so carrying an unused one costs nothing.
Choose which calls it analyzes
Open the agent and go to the Segments tab. Tick the kinds of calls this agent should score: AEP Sales Calls, Plan Enrollment, Plan Renewal, Fact Finder, Cross Sales, TPMO Disclaimer, SOA, Compliance, Customer Service, Appointment Setting, Follow-up or Other / Non-client. New agents start with AEP Sales Calls only. A call gets one type, by its main purpose: a sales call that includes the TPMO disclaimer or an SOA is still an AEP sales call.
Every call is classified from its transcript first. Ticked types get the full analysis and scorecard. Unticked types show on the Calls page as Skipped with their type, and are not scored, so they cost almost nothing. If the type guess is unsure, the call is analyzed anyway, so a real sales call is never lost to a bad guess. Filter the Calls page by Type to see each kind.
Voice AI calls and Human calls
Two switches sit above the call types and decide which calls are scored: Voice AI calls (calls your AI voice agent handled, such as HelloWiseAI, GHL Voice AI or VAPI) and Human calls (calls a person handled). The call types you tick apply to whichever switches are on. Both start on. AI calls are scored on the AI receptionist scorecard; people's calls on the scorecard for their type.
| Voice AI | Human | What gets scored |
|---|---|---|
| On | On | Every call of the ticked types, whether a person or your AI voice agent handled it |
| On | Off | Voice AI calls of the ticked types. People's calls show as Skipped — Human calls off |
| Off | On | People's calls of the ticked types. AI calls show as Skipped — Voice AI calls off |
| Off | Off | Nothing. The page warns you |
A line under the switches always says what the call types apply to. Calls where it is unclear who answered follow the Human calls switch. Voicemail, hang-ups and no-shows are filtered either way. Turn a switch back on and the calls it skipped are picked up again by the next scoring run.
Live vs draft
Routines only run for a live agent. Set an agent back to draft and its routines are paused automatically, a draft agent firing at 6am is the opposite of what "draft" means. Setting it live again does not re-enable them; switch the routine back on yourself, so nothing starts emailing clients without you knowing.
Reporting schedule
The Routines tab has a one-click recurring report, daily, weekly or monthly. Use it rather than writing the prompt by hand: paraphrasing the pipeline prompt is how a job silently stops writing analyses.
If a routine reports success but nothing appears
Open the run in Conversations and read what it actually did. The pipeline is explicit that a step producing nothing must say so. A report that describes what would happen, rather than what happened, means the agent could not reach a tool it needed, check the agent is live, and that your call source is connected under Settings → MCP or Integrations.
Run it on a schedule: the Routine library
Open the agent, go to the Routines tab and click Routine library. A window lists ready-made routines for call agents. You do not write a prompt: each one already knows which skills to run and where to save the results.
| Routine | What it does | Default schedule |
|---|---|---|
| Score new GHL calls | Finds new GoHighLevel calls (people and GHL Voice AI), scores them and saves each analysis | Every hour, 8am–8pm Mon–Fri |
| Score new VAPI calls | The same for your VAPI assistants' calls | Every hour, 8am–8pm Mon–Fri |
| Score new Retell calls | The same for your Retell agents' calls | Every hour, 8am–8pm Mon–Fri |
| Score new calls (any provider) | The same for any call provider you attach as an MCP that can list calls and get one call, such as HelloWiseAI or VAPI. It only reads from the provider, never places calls or changes your voice agents | Every hour, 8am–8pm Mon–Fri |
| Finish stranded calls | Picks up calls an earlier run left waiting for a transcript and finishes them | Every day, 5:30am |
| Compliance alert | Any newly scored call with a compliance problem (TPMO, SOA, identification, PHI) | Every 2 hours, 8am–6pm Mon–Fri |
| Daily call digest | Yesterday in one short read: volume, goal-met rate, compliance flags, the top objection, three calls worth a listen | Weekdays, 7:30am |
| Monday team review | Last week's rollup and a Review → Train → Roleplay agenda | Mondays, 9am |
| Weekly Voice AI check | How your voice agents did last week. Report only: it never changes an agent | Mondays, 8am |
| Monthly rep scorecard | Last month for every rep: trend, biggest problem, how to fix it, strength | 1st of the month, 9am |
- Pick a schedule on the card and click Add (off). The window closes and the routine appears, highlighted, in your routines list, switched off.
- Click Run now once and read the result in Conversations.
- When it looks right, switch it on.
A card that cannot be added says what is missing: a spoke on the Tools tab, a skill, or a connection (GoHighLevel under Settings → Integrations; VAPI, Retell or any other call provider under Settings → MCP, then attached on the agent's MCPs tab). Routines only run while the agent is Live.
Use a provider's own scoring routine or Score new calls (any provider), not both. Scoring routines handle 5 calls per run (per provider for the any-provider routine), oldest first, so running every hour keeps up with a busy day and clears a backlog over a few runs. When a run finds nothing new it is still recorded in the routine's history, but it does not post to Slack, Discord or your notifications. Reports with a channel post every time.
A routine you have already added shows Show in list in the window instead. When we improve a library routine, its card shows Update prompt. Your schedule, on/off switch and channels stay as they are.