How does AI outbound prospecting work?

AI outbound prospecting is the automated process by which an AI agent finds companies and contacts that match a defined ICP, researches them, writes personalised outreach, and initiates contact — all without a human doing each step manually. Here's exactly how it works.

Step 1: ICP definition

Before prospecting can run, the agent needs a defined ideal customer profile. This includes company characteristics (industry, size, geography, tech stack, growth signals) and contact characteristics (role, seniority, decision-making authority). The ICP is built during onboarding using the product description and target market, then sharpened over time as the agent's outreach produces replies — positive and negative both signal what to adjust.

Step 2: Multi-source search

The agent queries multiple prospecting sources simultaneously: Apollo for B2B contact data, LinkedIn for professional profiles, Clutch for agency and service business data, YC's company database for startups, Google Places for location-based businesses, and general web research for companies that don't appear in structured databases.

Each result is scored against the ICP criteria. Companies and contacts that meet the threshold are added to the prospect queue.

Step 3: Account research

For each qualified prospect, the agent reads the company's website, recent news, LinkedIn activity, and any available signal data. The goal is to understand their current situation well enough to write a relevant first sentence — not a generic hook, but something that reflects what's actually happening at that company. A SaaS company that just raised a Series A gets a different angle than one that's been flat for two years.

Step 4: Personalised outreach drafting

Using the account research, the ICP, and any coaching stored in memory, the agent drafts a personalised first-touch email. The email references specific information about the company or contact. It's not a template with a name field swapped — it's a message informed by actual research.

Coaching notes loaded from memory shape the copy. Which angle to lead with, which segments to prioritise, which subject lines to avoid — all of it comes from prior cycles, not from a human rewriting the playbook before each send.

Step 5: Sequence enrolment

The prospect is enrolled in a multi-step sequence. Subsequent touches fire on a defined schedule — typically 5–7 days apart — each with a different angle or value point.

Step 6: Reply handling and handoff

When a prospect replies, the response routes to the right next step. Positive replies hand off to the Account Executive with full context — the thread, the research, the sequence history. Opt-outs and hard negatives are logged and removed from active sequences. Ambiguous replies escalate to a human for judgment rather than triggering an automated follow-up that misreads the tone.

Step 7: Memory update and next cycle

After each cycle, the agent writes memory blocks: which segments responded, which messaging angles worked, what patterns emerged. This memory reloads on the next cycle. The targeting tightens. The copy improves. Not because someone updated the playbook, but because the loop closed.