An agent is your own copy of a template — the same tested conversation, with every detail that makes it yours: the voice, the language, what it knows about your business, what it should find out, and what happens after the call.
This guide walks the five tabs on an agent, in the order most people need them.
Quick start
Locales — voice, language and the variables the agent speaks from.
Insights — the results it reaches and the fields it extracts.
Settings — field mappings, if your data uses different column names.
Messages — follow-up SMS triggered by a result.
Webhooks — push results into your own systems.
Test it before building a campaign.
The Actions menu — create a campaign, or clone the agent.
Before you start
Open Agents in the Voice Agent cabinet and open the agent you saved from a template. Two controls sit at the top of every tab:
Enabled — an agent has to be enabled before you can create a campaign from it. Clearing the box takes it out of service without losing any configuration.
Test Call — a real conversation with your agent, using your current settings. It runs in the browser and needs your connections in place, and because it's a real call it's billed like any other.
Changes are saved with Save at the bottom. Nothing you set here alters the conversation flow itself — that stays as the template designed it.
Locales
A locale bundles everything language-specific: which connection to use, which language to speak, which voice, and the variables for that conversation.
Locale #1 is the catch-all. Its Countries field is fixed to all countries and can't be edited. Add a second locale and Countries becomes editable, so you can point specific countries at a different language, voice or connection. The locale used for a call is determined by the country set on the contact list when you upload it to the campaign.
Connection — which AI provider this locale calls. The available voices follow from it.
Language sets the primary language of the locale. STT Language is the language used to transcribe speech in cascaded templates. TTS Voice is the voice used to speak the reply.
How those combine depends on the technology behind the template:
Template type |
What drives the conversation |
Speech-to-Speech (realtime voice) |
Speaks with the voice set in TTS Voice, in the language set in Language. |
Cascaded |
Transcribes using STT Language, then speaks the reply using TTS Voice. |
The voice list is filtered by the selected connection and contains only voices vetted and supported by BSG — so anything you can pick is known to work.
Variables are the values the agent works from during the call: your brand name, the agent's name, the greeting, what your company does, answers to common questions. Each has a name and a value, and the value is yours to write.
You can add and remove variables as your scenario needs. Values can carry placeholders in curly braces — {brand_name}, {lead_first_name} — which are filled in per contact at call time, so one greeting works for a whole list.
Insights
Insights are what the agent hands back after every completed call. Two kinds, on two sub-tabs.
Results
A result is the label that summarises how the conversation ended. After the call, the agent reads the transcript and picks one.
Each result has three parts:
Name — free text, so you can name outcomes in your own vocabulary. A few names are reserved by the platform; if you try to reuse one you'll be told.
Group — how the result is classified for logic further down the line. Groups include success, other, answer_machine and dropped, among others. This matters because campaigns count successes by group rather than by individual result — see Customizing campaign settings.
Instruction — how the agent should recognise this outcome, in plain language. "User says he's busy or asked to call back later" is enough. This is the text the agent reasons against, so being specific pays off.
Add the results your business actually acts on, and delete the ones you'll never look at. A short, meaningful list is easier to build reporting and automation around than a long one.
Two results — the answering-machine one and the dropped-call one — are set by the platform rather than decided from the transcript. They come from system functions that detect voicemail and a dropped connection, which is why their Instruction fields are empty.
Fields
Fields are the data points the agent extracts from the conversation and returns with the call: a budget, a preferred time slot, a summary of what was said.
Each has a Name and an Instruction telling the agent what to capture. Instructions can be strict — "summarise only from information explicitly provided by the contact" — and it's worth being strict, because a field that invents detail is worse than an empty one.
Extracted values come back on the campaign's Contacts tab, in the call export, and in any webhook you configure.
Settings
The Mappings sub-tab connects your agent's fields to the column and property names your own systems use.
Each row has a Target — the agent field — and two sources: Source API for contacts pushed over the API, and Source File for contacts uploaded from a spreadsheet. The two can differ, so the same agent works whether contacts arrive from your CRM or from a file someone exported.
You only need this when your data doesn't already use the agent's names. If your CRM calls it first_name and the agent expects lead_first_name, one mapping row fixes it permanently — cheaper than renaming columns before every upload.
Messages
Messages are follow-up SMS, triggered by how a call ended.
SMS has to be enabled on your BSG account first. Contact the account manager shown in your cabinet to switch it on; until then the tab won't send anything.
Each message block has:
Method — bsg_sms, currently the only option.
Country and Sender — the sender is the name your SMS appears from, entered as free text, and Country selects which country's registered sender applies. Sender registration is per country, so a campaign spanning several needs a message block for each.
Enabled — off until you turn it on, so a half-configured message never sends.
Templates — the actual texts. Each one is tied to a Result, so you can send different wording depending on how the call went: a payment link to contacts who agreed to pay, a summary to contacts who asked for details. Add as many as you have outcomes worth following up. Choose All results to send the same text regardless.
The message text can be written here, or supplied per contact through a column in your file or a field in the API — useful when the content differs by contact rather than by outcome.
Messages are scheduled shortly after the call ends. Delivery shows up in the live campaign statistics and in Campaign Statistics, and the export from the Calls list breaks it down contact by contact.
Webhooks
A webhook pushes the call's outcome to a URL of yours, as soon as the conversation finishes. It's how a completed call becomes a calendar entry, a CRM update, or the first step of a longer automation.
Each webhook has:
Method — POST, PUT, GET or DELETE.
Country and Result — which calls it fires for. Leave both at All to catch everything, or narrow to a single outcome so your endpoint only hears about what matters.
URL — where the request goes.
Content — json or form.
Headers and Parameters — add as many as your endpoint needs, for authentication tokens or extra values.
Enabled — a webhook can be configured and left off, which is handy while you're building the receiving end.
Add several webhooks if different outcomes should reach different systems: bookings to your calendar, opt-outs to your suppression list.
Test your changes
Test Call runs your agent as configured, in the browser. Use it after any change to variables, voice or language — it's the fastest way to hear whether a new greeting lands or a reworded instruction actually changes behaviour.
It uses your connections and bills as a normal call, so it's a real test rather than a simulation.
The Actions menu
Actions at the bottom of the agent is where you go once the settings are right:
Add Campaign — create a campaign from this agent. The agent has to be enabled and you need an active SIP connection in Connections; this is the normal way to get from a configured agent to real calls.
Show Campaigns — every campaign built from this agent, which is the quickest way to see what a change here will affect.
Clone Agent — a copy with all your settings. Worth using before you experiment: clone, change the greeting or the voice on the copy, test both, and keep whichever wins. It's also the fast route to a second agent that differs in only one respect, such as language.
A note on campaigns
Campaigns have their own Locales, Messages and Webhooks tabs. Those add to what you configure here rather than replacing it — so agent-level settings are your defaults, and a campaign can layer on top for a particular run. See Customizing campaign settings.
Comments
0 comments
Article is closed for comments.