Troubleshooting
On this page
Almost every problem below is diagnosed the same way: open the transcript and find the exact turn where things went wrong. Start there before changing anything.
How to diagnose anything
- Open Call History and read the transcript. Not the recording — the text. You are looking for one specific turn.
- Check what the agent heard, not what the caller said. Mis-transcription makes a good agent look stupid, and it has a different fix.
- Change one thing. Then make one test call. Five changes at once tells you nothing.
- Verify on the other side. An agent saying it booked something is not evidence that it did. Look in the calendar.
Calls
| Symptom | Check, in this order |
|---|---|
| Nobody can reach my number | 1. Is the number attached to an agent? 2. Is that agent deleted? 3. If you are forwarding, is the rule active at your provider? Ring the Vociply number directly to isolate which side is at fault. |
| Outbound calls fail immediately | 1. Number format — it must be +E.164, like +254712345678. 2. Does the agent have a number attached? 3. Is the destination country reachable? 4. Is there a billing block? |
| Calls connect then drop after a few seconds | Usually the caller hanging up. Check the duration and the transcript — a very short call with only the greeting in it is a hang-up, not a fault. |
| Long silence at the start of the call | A very long system prompt costs thinking time before the first word. Shorten it and move reference material to a knowledge base. |
| Every outbound call shows as voicemail | Voicemail detection is working and you are calling at a time nobody answers. Try a different window. |
| A cluster of failed calls appeared suddenly | Almost always a formatting problem in a newly uploaded list. Pause the campaign, open the CSV in a text editor, check the phone column. |
Conversation quality
| Symptom | Cause | Fix |
|---|---|---|
| Talks over the caller | Wait timers too aggressive | Raise wait to ~0.7s and no-punctuation to ~1.8s in Voice. |
| Long pauses before replying | Timers too generous, or instructions too long | Lower the no-punctuation wait, and cut the prompt down. |
| Invented a price or a policy | Fact missing, and no “say you don’t know” rule | Add the fact, and add an explicit “never guess” instruction. |
| Ignores one of your rules | It contradicts another rule in the prompt | Find the conflict. Two “always” rules that cannot both hold produce coin flips. |
| Rambles | No length rule | “Keep every reply to one or two sentences.” near the top of the instructions. |
| Answers a question nobody asked | Mis-transcription | Check what the transcript says it heard. If it is wrong, try a different speech engine on the agent. |
| Will not transfer to a human | No transfer tool, or triggers too narrow | “If the caller asks for a person, is upset, or you have failed twice — transfer.” |
| Hangs up too early | End-call trigger too vague | “Only end the call after the caller has said goodbye or confirmed there is nothing else.” |
| Callers hang up in the first ten seconds | The greeting | Shorter. Name the business, offer to help, stop. |
Integrations
Check this first, always
Connected is not the same as attached. Open the integration panel and confirm this specific agent is selected. This is the cause more often than everything else on this list combined.
| Symptom | Cause | Fix |
|---|---|---|
| The agent never uses the integration | Nothing in the instructions tells it when to | Name the action and the trigger: “use check_availability before offering any time.” |
| “I don’t have access to that” | Connected but not attached to this agent | Attach it on the integration panel. |
| Works on one agent, not another | Only attached to one | Attach it to both. |
| Worked yesterday, not today | Credentials expired or were revoked | Reconnect — as a shared service account this time, so it survives staff changes. |
| Books appointments at times that are not free | Availability never checked, or the calendar is not honest | Add the check-first instruction, and block lunch, admin time and holidays in the calendar itself. |
| Creates duplicate CRM contacts | Creating without searching first | “Never use create_contact without using search_contacts first.” |
| Google Sheets does nothing | No spreadsheet chosen after connecting | Finish the setup step — signing in authorises the API, not a particular sheet. |
| n8n workflow never runs | Workflow not Active, or the test URL was used | Activate the workflow and use the Production URL from the Webhook node. |
Leads and campaigns
| Symptom | Check |
|---|---|
| Meta leads never arrive | The Page was not chosen on the Meta integration after connecting. That step is what subscribes the Page to lead notifications, and it is invisible in Facebook’s own interface. Also confirm the Page ID on the trigger matches. |
| Webhook leads never arrive | Check the trigger’s recent activity to see whether anything is being received at all. If yes, the phone field mapping is wrong. If no, the secret header or the URL is wrong on your form platform’s side. |
| Leads arrive but no call is placed | 1. Is the trigger’s agent attached to a phone number? 2. Is it outside the calling window — the lead is held, not lost. 3. Is consent required and missing? 4. Is the number on your do-not-call list? |
| Everyone qualifies, or nobody does | Read the transcripts. Nine times out of ten the agent never asked the question, which is an instructions problem, not a criteria problem. |
| Transfers ring out | Your calling window is wider than your team’s hours. Narrow it, or set transfers to Never outside working hours. |
| Campaign connect rate is very low | Bad numbers, or an unfamiliar caller ID. Use a local number for the market you are calling. |
Call IQ
| Symptom | Cause | Fix |
|---|---|---|
| No drafts appear at all | The call did not go through the analysis line | Only calls routed through the line are analysed. Check how calls are reaching the rep. |
| An analysis failed | A transient failure, often a timeout | Use Reprocess — it fetches a fresh copy of the recording and runs the pipeline again. Most clear on the second attempt. |
| Extractions are consistently wrong | How the calls are conducted | Full product names, numeric quantities, and reading the order back at the end. One page to the sales team fixes more than any setting. |
| Approved orders do not reach Dynamics | The connection, or a field Dynamics rejects | Check the failure counter on the Call IQ page and the Dynamics connection. Re-approve after fixing. |
| The line does not route calls | The line is not active | Only active lines route. A disabled line is history. |
Contacting support
If none of the above fits, email support@vociply.com. Include these five things and you will get a real answer on the first reply rather than the third:
- The call ID from Call History, or the agent ID from AgentsCopy ID.
- What you expected to happen.
- What happened instead — quote the line from the transcript.
- When it started, and what changed around then.
- Whether it happens every time or occasionally.
Do not paste your whole prompt
Paste the transcript turn that went wrong and the two or three instruction lines that relate to it. Support can work from that immediately; a 900-word prompt with no pointer takes a round trip to narrow down.
Still stuck? support@vociply.com