How to Set Up an AI Leads Engine: Scrape → Qualify → Personalize → Send

Outcome: targeted outbound without spammy blasts.

0) The Blueprint (what you’re building)

• Inputs: Your ICP, lead sources, enrichment fields, scoring rubric, first-line prompt, email sequences.
• System: A repeatable workflow that pulls fresh leads weekly, ranks them, drafts personalized openers, and sends through warmed inboxes with A/B experiments.
• Outputs: Meetings booked, reply rate, positive-intent rate, cost per meeting.
• Guardrails: Deliverability, opt-out handling, compliant data sources, human spot-checks.

1) Define ICP & Sources (quality in = quality out)

ICP (Ideal Customer Profile) essentials

• Firmographics: industry, headcount range, revenue band, geography.
• Technographics: stack clues (e.g., Shopify vs. WooCommerce), data tools, cloud.
• Triggers: hiring bursts, new funding, new location/site, tech migration, leadership change.

Where to find leads (mix 2–3)

• Company directories: vendor marketplaces, partner lists, conference/exhibitor pages
• Public signals: job posts, press releases, SEC filings, facility permits
• Social/communities: LinkedIn posts, subreddit flairs, niche forums
• Enrichment vendors: Apollo, Cognism, Clay data sources, Clearbit; also your CRM + past wins

Tip: Document why each source maps to your ICP so your AI can reference it in scoring and personalization.

2) Enrichment (make raw leads usable)

Pull only what you’ll actually use:

• Company: website, HQ/state, NAICS/industry, employee count, revenue band, tech tags, recent news link
• Contact: role/title, department, seniority, LinkedIn URL
• Context hooks: a line from a recent post, case study, hiring need, location, product launched

Stack sketch (example)

• Scrape/collect: Clay, PhantomBuster, Browserless/Playwright, manual CSVs from events
• Enrich: Clay + Clearbit/Apollo APIs; basic site parsing for “About” and “Careers”
• Store: Google Sheet/Airtable or a leads table in your CRM

3) Scoring Rubric (AI-assisted, rules-enforced)

Create a simple, explainable score (0–100). Combine rules + AI justification.

Rules (weighted)

• Industry match (+25)
• Title contains Director/VP/Owner for your buyer (+20)
• Tech fit (e.g., “uses Shopify”) (+15)
• Geography in target states (+10)
• Trigger present (hiring, funding, expansion, recent launch) (+20)
• Negative filters (agency, student, non-commercial) (–20 each)

AI assist (optional)

• Prompt a short “Why this account fits” summary using the enriched fields.
• If the AI’s reason doesn’t cite data you have, drop the score or flag for review.

Cutoffs

• 75–100: priority queue for 1:1 outreach
• 50–74: nurture content or LinkedIn warm touches
• <50: park for later; do not email

4) 1:1 First Lines (no fluff, no guessing)

Your “first line” is the only truly custom sentence. It must be factual, specific, and relevant.

Prompt pattern (use with your model of choice)

“Write a single, 22-to-35-word opener that references one concrete fact from the fields provided (post/news/role/hiring/tech). Avoid praise clichés. Sound human, not salesy.”

Inputs to the prompt

• Company: name, product, recent news link/title
• Contact: title, region
• Context: 1–2 bullet facts you scraped (quote or summarize)

Good examples

• “Your Houston posting for a night-shift warehouse supervisor and the new 24/7 line suggest throughput is spiking—are you exploring shift-handoff checklists that cut rework for new hires?”
• “Saw your Shopify → SFCC migration note; teams usually hit PDP load and content ops issues—open to a 20-minute ‘what to watch’ compare from recent migrations?”

Bad examples (never use)

• “I love what you’re doing!”
• “Hope this finds you well.”
• Anything that could be true for any company.

Workflow

1. Generate first lines in batch.
2. Auto-flag anything lacking a citation or over 40 words.
3. Human skim 25–50 per batch; approve or edit.

5) Message & Sequence (short, useful, skimmable)

Email 1 (value + relevance)

• First line (custom)
• 1-sentence problem statement tied to their context
• 1-sentence “how we solve it” (proof/metric)
• 1 simple CTA (binary choice, 20–25 min)

Email 2–3 (evidence & objection)

• Social proof: similar client + result
• 1 practical artifact: 4-bullet checklist, 90-second loom, 1-page teardown
• Handle 1 likely objection (budget, timing, vendor fatigue)
• Soft CTA: “Worth a quick compare?”

Email 4 (breakup/redirect)

• “Close the loop” + useful resource + opt-out language

Keep each email 60–110 words. Plain text. No images in early sends. Use a single link max.

6) A/B Testing (small bets > spray & pray)

Test one change at a time; run until 200–300 sends/variant before calling it.

High-leverage tests

• First-line style: news-hook vs. role-pain
• Problem framing: time-to-value vs. risk reduction
• CTA: “this week or next?” vs. “worth a 20-min compare?”

Metrics to track

• Deliverability: bounce < 2%, spam < 0.1%
• Opens: only directional (image blockers skew)
• Replies: ≥ 8–12% overall on well-fit lists
• Positive intent: ≥ 2–4% (intros, call booked)
• Cost per meeting: your ultimate governor

7) Warm-Up & Deliverability (protect the domain)

• Use dedicated sending domains (e.g., get.yourbrand.com); keep root domain pristine.
• Warm inboxes gradually (e.g., 15 → 30 → 60 → 120/day).
• Rotate 2–4 inboxes per rep; cap at 120–150/day each.
• Authenticate (SPF, DKIM, DMARC), consistent signatures, no images/attachments early.
• List hygiene: verify emails; never send to role-based (info@, sales@).
• Always include a working opt-out that’s honored within 48 hours.
• Monitor blocklists and domain reputation weekly.

8) Operating Cadence (how this runs weekly)

Monday

• Pull + enrich new leads (goal: 200–500 net).
• AI score + human spot-check top 50.

Tuesday

• Generate first lines; approve in batches.
• Launch Sequence A (two inboxes).

Wednesday

• Launch Sequence B (two inboxes).
• Review deliverability dashboards; pause if bounce > 2%.

Thursday

• Analyze replies; label positive intent vs. objections.
• Update scoring rubric based on patterns.

Friday

• Run A/B analysis; promote winners.
• Refresh opt-outs and suppression lists.

9) Compliance & Respect (no spammy blasts)

• Send only to business-relevant contacts your product genuinely serves.
• Make it easy to opt out and actually remove them.
• Avoid “data exhaust” sources you can’t justify.
• In sensitive geographies, follow local outreach rules.
• Do not claim affiliations or results you can’t prove.

10) What This Looks Like in Tools (example stack)

• Data/Scrape: Clay + targeted scrapers (Playwright), LinkedIn Sales Nav exports
• Enrichment/Verify: Clearbit/Apollo + NeverBounce/ZeroBounce
• Scoring/AI: OpenAI API or Claude for first lines; rules in Clay/Airtable
• Send: Instantly, Smartlead, Lemlist, or Apollo for sequences
• CRM: HubSpot/Pipedrive; auto-log replies and outcomes
• Analytics: Simple dashboard (Airtable/Looker Studio) for reply + positive-intent

(Use what you already pay for—stack choice matters less than the rules and QA.)

11) Success Criteria (green lights)

• 70%+ of weekly sends go to 75+ scored accounts
• Reply rate ≥ 10% on priority segments
• Positive intent ≥ 3%
• Domain reputation steady; bounces < 2%
• Booked meetings per week is rising on steady send volume

12) Common Failure Modes (and fixes)

• Great copy, wrong list: Revisit ICP and sources first.
• Low replies despite opens: Your problem statement is generic; add trigger-based specificity.
• Spam hits climbing: Slow send, prune lists, check authentication and link domains.
• AI hallucinations in first lines: Force citations; reject any opener without a specific, verifiable fact.

Want this done for you?

BoostMyAI will stand up your AI Leads Engine in weeks, not months—ICP, scraping, enrichment, scoring, 1:1 first lines, deliverability, and A/B ops—so you get targeted outbound without spammy blasts.

Ready to book more meetings with fewer sends? Contact us and let’s build it.