← Back to guide
AgentDukaanFree guide

Marketing, Sales & Lead Gen · Intermediate · 7 min read

How to run the full B2B lead lifecycle with AI agents: ethical sourcing, personalised outreach, auto-qualifying, booking, and the metrics th

Normally ₹1,499 — free while we launch.

Lead Generation with AI Agents

Most lead-gen breaks not because you lack tools, but because the work between "found a name" and "booked a meeting" is tedious, repetitive, and easy to drop. That is exactly the work an AI agent does well: enrich, personalise, qualify, route, and follow up—on WhatsApp, Telegram, email, or in-app chat—without you babysitting a spreadsheet. This guide shows you how to wire that up responsibly and measure whether it actually works.

The lead lifecycle an agent can run

Think of your agent as a worker that moves a lead through fixed stages, doing a specific job at each one. Map your pipeline to these stages before you automate anything:

StageAgent's jobChannel
SourceFind people matching your ICP from public/opt-in lists
EnrichAdd company size, role, location, intent signals
OutreachSend a personalised first-touch messageEmail / WhatsApp
QualifyAsk 2-4 questions, score the replyWhatsApp / in-app chat
RouteAssign to a rep or self-serve flowInternal + email
BookOffer slots, confirm, add to calendarWhatsApp / email
NurtureFollow up on non-responders and "not now" leadsEmail / WhatsApp

The discipline here matters more than the AI: define the entry and exit condition for each stage. A lead only moves to "Qualify" when it has a valid contact and a reply; it only moves to "Book" when it scores above your threshold. The agent enforces these rules so nothing leaks.

Finding and enriching leads ethically

Under India's DPDP Act, you need a lawful basis to process someone's personal data, and "I scraped it" is not one. Stick to sources where people have a reasonable expectation of business contact: inbound form fills, event/webinar opt-ins, your own WhatsApp Business inbox, LinkedIn connections you actually made, and verified B2B directories you have a licence to use. For cold outreach to a business email or number, keep it relevant, identifiable, and easy to opt out of.

Enrichment is where an agent earns its keep. Feed it a raw list (name, company, email) and have it append the fields you actually use to qualify:

  • Company size band (1-10 / 11-50 / 51-200 / 200+)
  • Industry and city/state
  • Role seniority (decision-maker vs influencer)
  • A recent public signal (hiring, funding, new product, GST registration tier)
  • Best channel (does the company prefer WhatsApp? Is there a generic info@ to avoid?)

Keep a source and consent_basis column on every record. When someone asks "where did you get my number?", you want an honest answer in one click—that is both DPDP hygiene and good manners.

Personalised outreach at scale (without spam)

Spam is not defined by volume; it is defined by irrelevance and absence of consent. You can send 500 messages a day and not be a spammer if each one is specific and each recipient can leave. The agent's job is to make 500 messages feel like 500 individual notes.

Avoid the obvious tells: no "Dear Sir/Madam", no walls of text, no five links in the first message, and on WhatsApp, respect the rule that cold first-contact should use an approved template, not a free-form blast. Lead with one concrete, true observation about their business, then a single small ask.

Here is a copy-paste prompt to generate first-touch messages at scale. Give the agent your lead's enriched fields and your offer:

You are writing a first-touch B2B message. Rules:
- Max 45 words. One observation about THEIR business, one soft ask.
- No "Dear Sir/Madam", no flattery, no more than one link.
- Tone: warm, direct, peer-to-peer. Hinglish OK if the lead is India SMB.
- End with an easy opt-out: "Reply STOP and I won't message again."

Lead: {name}, {role} at {company} ({industry}, {city}, {size_band})
Signal: {recent_public_signal}
My offer (one line): {your_value_prop}
Channel: {whatsapp | email}

Write 2 variants. Do not invent facts about the company.

The "do not invent facts" line is load-bearing—an agent that hallucinates a fake compliment will torch your credibility. For deeper copy patterns, see Prompt Engineering for Real Business Tasks and the 40 WhatsApp Customer-Support Prompts (Copy-Paste).

Qualifying and routing leads automatically

When someone replies, the agent should ask 2-4 short qualifying questions in a natural back-and-forth, not a form dump. The classic frame works fine: need, timeline, budget band, and who decides. Each answer maps to a score.

A simple, transparent scoring rule beats a black box you can't debug:

SignalPoints
Decision-maker role+3
Replied within 24h+2
Stated a timeline under 90 days+2
Budget in your range+2
Generic/info inbox or "just browsing"-2

Set thresholds: 7+ goes to a human rep immediately (this is your hot lane), 3-6 enters a nurture sequence, below 3 gets a polite self-serve link and exits. Routing should be deterministic—post the lead to your CRM, ping the right rep on Telegram or email, and stamp the reason ("scored 8: decision-maker, 30-day timeline"). The rep walks in already knowing why.

Build a graceful handoff: when a lead asks something the agent shouldn't answer (custom pricing, contract terms), it should say so and pull in a human rather than bluffing.

Booking meetings and follow-up sequences

A booked meeting is the only conversion most B2B teams actually care about, so make booking the path of least resistance. The agent should offer two or three concrete slots ("Thu 4pm or Fri 11am IST?") instead of a bare calendar link—specific options convert better than open-ended ones. On confirmation it adds the event, sends a calendar invite, and fires a reminder the morning of.

For the majority who don't reply, a disciplined sequence does the heavy lifting. A sane cadence:

  • Day 0: first touch (the personalised message above)
  • Day 3: short bump with a different angle or a relevant resource
  • Day 7: a "should I close your file?" message—surprisingly effective and respectful
  • Then: stop, or move to a low-frequency monthly nurture if they said "not now"

Hard rule: every "not now" gets a real follow-up date, and every "no" is honoured permanently. The agent should maintain a suppression list and never re-contact an opt-out. Three to four touches over two weeks is firm; eleven touches in five days is harassment.

Metrics that matter and a sample funnel table

Vanity metrics (messages sent, "impressions") tell you nothing. Track the conversion between stages so you know exactly where the funnel leaks. Here is a hypothetical funnel for one month—use the shape, not the numbers:

StageCountConversion to next
Leads sourced1,000
Enriched + contactable82082%
Replied18022%
Qualified (score ≥ 3)9553%
Meeting booked3840%
Deal won924%

Watch two numbers above all: reply rate (is your first touch relevant?) and booked-to-qualified rate (is your follow-up working?). If reply rate is low, fix the message, not the volume. Also track cost per booked meeting—if your agent subscription is ₹2,000/month and it books 38 meetings, that is roughly ₹53 per meeting, a number you can defend to anyone holding the budget.

Next steps

  • Write down your ICP and the exit condition for each of the seven lifecycle stages.
  • Pick ONE channel (WhatsApp or email) and one segment of ~100 opt-in leads to start.
  • Run the outreach prompt above on 10 leads by hand first; only automate once the messages read well.
  • Add source, consent_basis, and a suppression list before you scale a single message.

If you'd rather not build this from scratch, browse ready-made lead-gen and outreach agents on AgentDukaan, or read Build Your First AI Agent: Idea to Live in a Weekend if you want to roll your own. Start small, measure the reply rate, and let the funnel tell you what to fix next.

Source: agentdukaan.in/guides/lead-generation-with-ai-agents · © 2026 AgentDukaan · Shared free during launch.