BANT Outbound Assistant Walkthrough

Overview

This is a full worked example of building an outbound sales-qualification assistant, using an Assistant-based approach (a single focused assistant, not a squad) and the same underlying techniques covered in Custom Function Tools, Files API, and Scheduling Outbound Calls. The assistant places outbound calls to prospects, gets permission to continue, qualifies the lead using BANT (Budget, Authority, Need, Timeline), handles common objections, and books a meeting when the lead is a fit.

Note on the structured-output schema: Real Estate Lead Qualification: Structured Output Template is a different resource — a ready-made JSON schema for a real-estate-specific lead qualification call. This page is a full assistant build for a general BANT sales-qualification flow, from knowledge base and tools through system prompt and outbound trigger; use it as the walkthrough pattern, and adapt the Real Estate template (or build your own schema per Structured Outputs (Post-Call Data Extraction)) for the structured-output side of your own use case.

Upload three CSV files through the Files API before building anything else:

• leads.csv — columns: lead_id, name, phone, company, prior_call_notes
• products.csv — product/plan details the assistant can reference when a prospect asks what's on offer
• call_outcomes.csv — a reference list of standard outcome categories to log the call under

curl --location 'https://api.sulus.ai/file' \
--header 'Authorization: Bearer <YOUR_API_KEY>' \
--form 'file=@"leads.csv"'

curl --location 'https://api.sulus.ai/file' \
--header 'Authorization: Bearer <YOUR_API_KEY>' \
--form 'file=@"products.csv"'

curl --location 'https://api.sulus.ai/file' \
--header 'Authorization: Bearer <YOUR_API_KEY>' \
--form 'file=@"call_outcomes.csv"'

Save each returned file id — the tools built in Step 3 reference these files. See Files API for the complete field reference.

import { CoreSystemClient } from "@core-system/server-sdk";

const client = new CoreSystemClient({ token: process.env.CORE_SYSTEM_API_KEY });

const assistant = await client.assistants.create({
  name: "Outbound SDR",
  firstMessage: "Hi, this is Alex from TechFlow. Is now a good time to chat for 2 minutes?",
  model: {
    provider: "openai",
    model: "gpt-4o",
    messages: [
      { role: "system", content: systemPrompt }
    ]
  }
});

systemPrompt is built out in full further down this page.

The assistant uses four function tools, defined the same way described in Custom Function Tools:

const updatedAssistant = await client.assistants.update(assistant.id, {
  model: {
    toolIds: [
      lookupLeadTool.id,
      scoreLeadTool.id,
      updateCrmTool.id,
      bookMeetingTool.id
    ]
  }
});

As with any assistant update, this replaces the entire model object — include every field you want to keep, not just toolIds.

A good outbound BANT system prompt has five distinct parts:

1. Identity & purpose — who the assistant is, who it's calling on behalf of, and what the call is for.
2. Permission gate — asking whether now is a good time before qualifying anything.
3. BANT qualification — a clear sequence for surfacing Budget, Authority, Need, and Timeline.
4. Objection handling — how to respond to common pushback without being pushy.
5. Booking & wrap-up — when to offer a meeting, and how to close the call either way.

# Outbound SDR Agent Prompt

## Identity & Purpose
You are Alex, an outbound sales development rep calling on behalf of TechFlow. Your goal is to qualify the prospect using BANT and book a meeting with a closer if they're a fit.

## Permission Gate
Always start by asking if now is a good time to talk for two minutes. If not, offer to call back at a better time and end the call politely -- do not attempt to qualify an unwilling prospect.

## Data Sources
- leads.csv: lead_id, name, phone, company, prior_call_notes
- products.csv: product and plan details
- call_outcomes.csv: standard outcome categories

## Available Tools
1. lookup_lead - retrieve the prospect's record and history
2. score_lead - score the qualification once BANT is gathered
3. update_crm - log the outcome and next steps
4. book_meeting - schedule a meeting once qualified

## BANT Qualification Flow
1. Budget - ask about current spend or budget range for this type of solution
2. Authority - confirm who else is involved in a decision like this
3. Need - understand the problem they're trying to solve
4. Timeline - ask when they'd want a solution in place
Call score_lead once all four are gathered.

## Objection Handling
- "Not interested" -> ask one clarifying question before disengaging gracefully
- "Send me info" -> offer a brief call instead, or confirm the best email and log it
- "Too expensive" -> acknowledge, then ask what budget range would work

## Booking & Wrap-Up
- If qualified: call book_meeting and confirm the date/time back to the prospect
- If not qualified or not interested: call update_crm with the outcome and end politely
- Always call update_crm before ending the call, regardless of outcome

## Style & Tone
- Friendly, concise, no more than two sentences per turn
- One question at a time
- Never pressure a prospect who has said no

Attach a structured-output schema (see Structured Outputs (Post-Call Data Extraction)) to pull these fields cleanly off every call:

await client.calls.create({
  phoneNumberId: "your-phone-number-id",
  customer: { number: "+15551234567" },
  assistantId: assistant.id
});

Or via cURL:

curl -X POST "https://api.sulus.ai/call" \
  -H "Authorization: Bearer $CORE_SYSTEM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "assistantId": "your-assistant-id",
    "phoneNumberId": "your-phone-number-id",
    "customer": { "number": "+15551234567" }
  }'

To run this against a large lead list instead of one prospect at a time, see Outbound Campaigns; to delay the call to a future time window rather than dialing immediately, see Scheduling Outbound Calls.