Automate Voice Agent for Lead Qualification by Phone with Vapi.ai — n8n Workflow in n8n — Step by Step
A lead fills out your demo form at 11:47 PM. By the time an SDR sees it the next morning, opens HubSpot, and dials, it's 9:40 AM — nearly ten hours of silence. That lead has already booked a call with
A lead fills out your demo form at 11:47 PM. By the time an SDR sees it the next morning, opens HubSpot, and dials, it's 9:40 AM — nearly ten hours of silence. That lead has already booked a call with two competitors. The data is brutal and consistent: contacting an inbound lead within five minutes makes them 21x more likely to convert than waiting 30 minutes. Every hour after that, the odds collapse. Yet most teams still route new leads into a queue that a human works through in business hours, in the order they feel like.
The bottleneck isn't your product or your pitch. It's the gap between "form submitted" and "first human contact." This article shows you how to close that gap to under 30 seconds — automatically, 24/7 — using an AI voice agent built on Vapi.ai and orchestrated by n8n. No SDR headcount, no missed after-hours leads, and every call scored and logged in HubSpot before your team wakes up.
The real cost of slow lead follow-up
Speed-to-lead is the single most under-optimized metric in most sales orgs. Here's what actually happens with manual follow-up: leads pile up overnight and on weekends, SDRs cherry-pick the ones that "look good," and 40–60% of inbound never gets a second dial. The leads that do get called are qualified inconsistently — one rep runs a tight BANT script, another just asks "so what are you looking for?" and forgets to log the answers.
The result is a dirty pipeline. Your CRM fills with contacts that have no budget signal, no timeline, and no authority flag. Your account executives waste hours on discovery calls that should have been disqualified at the door. And your best leads — the ones ready to buy right now — churn to whoever called them first.
You can throw bodies at this. A team of SDRs dialing within five minutes, around the clock, across time zones, is expensive and hard to staff. Or you can put an AI voice agent on the front line: it never sleeps, calls in seconds, runs the exact same qualification script every time, and hands your humans only the leads worth their time.
The solution: an AI voice agent that qualifies before you dial
The workflow is simple to describe and powerful in practice. A new lead submits a form. Within 30 seconds, Vapi.ai places an outbound phone call with a natural-sounding AI voice. The agent introduces itself, confirms it's a good time, and runs a structured BANT conversation — Budget, Authority, Need, and Timeline. It listens, follows up on vague answers, and handles objections from a defined script. When the call ends, Vapi returns a structured transcript and analysis. n8n scores the lead, updates HubSpot with the BANT fields and a qualification score, sets the next step, and — if the lead is hot — notifies a human to take over.
The division of labor matters. Vapi handles the voice layer: telephony, speech-to-text, the LLM conversation, and text-to-speech. n8n is the operator: it catches the form, triggers the call, receives the results, applies your scoring logic, and writes to your CRM. This separation keeps each part replaceable — swap HubSpot for Salesforce, or Vapi's voice model, without rebuilding the whole thing.
Step-by-step: building it in n8n
Here's the concrete build. The workflow has two halves: trigger the call and process the result. You need a Vapi.ai account (with an assistant and a phone number configured), a HubSpot private app token, and your n8n instance reachable from the internet for webhooks.
1. Catch the new lead. Start with a Webhook node (method POST) as the entry point, or an HubSpot Trigger node listening for contact.creation if leads already land in your CRM. Point your form tool (Typeform, Webflow, your own site) at the webhook URL. The payload should carry at minimum the lead's name, phone number, and email.
2. Normalize and validate. Add a Set (Edit Fields) node to map incoming fields to clean variables: phone, firstName, company. Then an IF node to check the phone number exists and is E.164 formatted (e.g. +15551234567). Vapi will reject malformed numbers, so drop or route invalid ones to a Slack alert instead of failing silently.
3. Trigger the Vapi call. Use an HTTP Request node — POST to https://api.vapi.ai/call. Set Authentication to Header Auth with Authorization: Bearer YOUR_VAPI_KEY. In the JSON body, pass your assistantId, your phoneNumberId, and a customer object with the lead's number. Inject lead context into the assistant via assistantOverrides.variableValues so the AI greets them by name and references what they signed up for. Your BANT script and objection handling live in the Vapi assistant's system prompt — keep it tight: one question per turn, confirm answers, and instruct it to extract budget range, decision-maker status, primary need, and buying timeline.
4. Receive the result. Vapi calls back when the call ends. Add a second Webhook node subscribed to Vapi's end-of-call-report event (configure the server URL in your Vapi assistant). This payload contains the full transcript plus a structured analysis object if you defined a structuredData schema in Vapi — which you should, so BANT fields come back as clean JSON instead of you parsing prose.
5. Score the lead. Add a Code node (or an AI Agent / OpenAI node if you want an LLM to grade nuance). Simple deterministic scoring works well: assign points per BANT dimension — budget confirmed (+30), decision-maker (+25), clear need (+25), timeline under 90 days (+20). Sum to a 0–100 score and bucket into hot (≥70), warm (40–69), or nurture (<40).
6. Write to HubSpot. Use the HubSpot node (Contact → Update, or Create-or-Update to be safe). Map the score, each BANT field, the call transcript, and a next_step property. Add a Switch node on the bucket: hot leads trigger a Slack or WhatsApp notification to the AE and create a HubSpot task due immediately; warm leads enter a nurture sequence; low scores are tagged and parked. Every branch ends by logging the outcome so nothing disappears.
Why this changes your funnel
The gains compound. Speed: first contact drops from hours to seconds, capturing intent at its peak. Coverage: 100% of leads get called — nights, weekends, holidays — with zero cherry-picking. Consistency: every lead is qualified against the identical BANT rubric, so your pipeline data is finally comparable and trustworthy. Focus: your human closers only touch pre-qualified, high-intent leads, which lifts their close rate and their morale. And cost: a Vapi call costs cents; an SDR making the same dials costs thousands per month. One workflow replaces the entire first-touch tier of an SDR team while producing cleaner data than any of them.
The strategic shift is that qualification moves upstream of human time. Instead of paying people to find out who's worth talking to, the machine finds out first and only escalates the winners.
Common pitfalls to avoid
Bad phone data breaks everything. Non-E.164 numbers are the number one cause of failed calls. Validate and reformat before the Vapi node — a small Code node that strips spaces and prepends the country code saves hours of debugging.
Skipping structured output. If you don't define a structuredData schema in your Vapi assistant, you'll be regex-parsing transcripts to extract budget and timeline — fragile and error-prone. Define the schema once and get clean JSON every time.
Compliance is not optional. Automated outbound calls are regulated. Honor consent — only call leads who opted in — respect calling-hour windows per region, and have the AI identify itself as an AI assistant. Bake a Do-Not-Call check into the IF node before triggering.
No fallback path. Calls go to voicemail, leads don't answer, Vapi occasionally errors. Add a retry branch (wait 20 minutes, dial once more) and a graceful handoff: after two missed attempts, drop the lead into a human queue with an SMS follow-up rather than losing it.
Over-scripting the AI. A rigid interrogation feels robotic and gets hung up on. Let the assistant acknowledge answers conversationally and ask one question at a time. The goal is a human-feeling qualification call, not a survey read aloud.
Build this once and your speed-to-lead problem is solved permanently — every lead called in seconds, qualified consistently, and delivered to your team ready to close.
Ja construimos isso pra voce
Nao comece do zero. O Voice Agent for Lead Qualification by Phone with Vapi.ai — n8n Workflow e um workflow n8n pronto para instalar — conecta suas ferramentas em minutos, sem codigo.
Instalar por $99.0 →
