Byndex Lead Finder

An Apify Actor for the discovery stage of Byndex: find local businesses for one category, city, and country, then recover public email routes from their official websites with source evidence.

Version 0.3.3 goal

The pilot target is at least 31 clean public emails from 50 accepted businesses. This is a measured target, not a guaranteed yield: some real businesses publish no email at all.

Version 0.3.3 (V3.2.1) does not invent or permute an address. It can evaluate up to 70 accepted businesses for a 50-lead run, crawl their official websites, retry unreachable sites with bounded hostname/protocol variants, and use Chromium only for sites that remain unreachable. clean_public_email_count measures addresses that passed strict syntax/domain rules and appeared in a public source. outreach_eligible_email_count is the stricter subset after domain, page-purpose, role-usefulness, and branch ranking. Neither status proves that a mailbox currently exists or can receive mail.

Workflow

1. Call the configured discovery Actor (default: compass/crawler-google-places).
2. Normalize and deduplicate business identity, address, official website, map URL, rating, reviews, and phone.
3. Reject out-of-city records and records without an official website by default.
4. Build a candidate pool larger than the final output (default: 1.4×, up to 70 candidates for 50 leads).
5. Crawl the starting website page, homepage, /contact, and /contact-us for every pool candidate.
6. Retry fully unreachable websites with bounded HTTPS/HTTP and www/non-www variants.
7. Use a two-page Playwright fallback only when both normal HTTP phases fail.
8. Follow the strongest same-site contact, location, office, support, team, staff, about, privacy, legal, and company links up to the page limit.
9. Recover emails from explicit public evidence channels in this priority order:
   • mailto: links
   • JSON-LD email properties
   • safe email data attributes
   • Cloudflare-protected email attributes
   • visibly obfuscated text such as name [at] domain [dot] com
   • ordinary visible page text
10. Rank automation-eligible addresses first, then compare domain quality, target-branch/location evidence, page purpose, extraction source, and page relevance.
11. Select the final leads from the fully evaluated pool: official-domain eligible email, free-provider eligible email, review email, phone-only, then website-only. Original order is preserved inside each tier.
12. Publish exactly one eligible primary address in public_emails; isolate alternates, possible other-location routes, and review-only routes in separate fields.
13. Store evaluated non-selected records in RESERVE_LEADS, then publish candidate evidence, review status, quality flags, and a run-level yield report.

V3.2.1 quality controls

• Preserves DOM boundaries before reading visible text, preventing labels and phone numbers from being joined to emails.
• Never scans arbitrary JavaScript, styles, SVG, templates, or raw HTML for addresses.
• Requires an ICANN-recognized public suffix and rejects asset extensions, monitoring domains, placeholders, no-reply routes, legal-only routes, and malformed domain suffixes.
• Includes regression coverage for the exact V2 failures such as .comall, .comsitemapbottom, phone-prefixed local parts, Wix/Sentry strings, and image filenames.
• Keeps the correct email source type and source URL for the selected primary address.
• Evaluates a configurable surplus pool before final selection, so a phone-only business discovered early cannot displace a stronger email lead discovered later.
• Treats a city in a source URL as branch evidence only on contact/location pages or the exact official starting path. Unrelated city articles do not receive target-branch status.
• Records crawl_failure_reason, crawl_http_status, crawl_retry_count, and crawl_recovery_method on every lead.
• Moves secondary person/location mailboxes from different office pages into other_location_emails for review.
• Recognizes a narrowly defined official-site brand alias only when an explicit evidence channel publishes it on the official site and both domain stems share a strong prefix. It remains review-only as OFFICIAL_SITE_BRAND_ALIAS_REVIEW.
• Marks person/location mailboxes found only on privacy, legal, terms, HIPAA, or privacy-notice pages as REVIEW_EMAIL_PAGE_PURPOSE.
• Prefers non-legal evidence when the same address also appears on a normal contact, location, or home page.
• Adds explicit confidence labels: PUBLIC_SOURCE_HIGH_CONFIDENCE, PUBLIC_SOURCE_MEDIUM_CONFIDENCE, REVIEW_REQUIRED, and NOT_AVAILABLE.
• Sends automated-response routes such as responses@... and corporate-support routes such as customerservice@... to review instead of approving them for outreach.
• Keeps domain mismatches, low-usefulness routes, alternates, and possible other-location addresses out of the automation-safe public_emails field.
• Deduplicates US phone formats by canonical digits.
• Uses website phones only to confirm the discovery phone. It does not append unrelated branch numbers. If discovery has no phone, it accepts one explicit official-site tel: route.
• Keeps every n8n-facing scalar field present even when empty.

Input example

{
  "category": "dentists",
  "city": "Phoenix",
  "country": "United States",
  "maxLeads": 50,
  "prioritizeOutreachEligible": true,
  "candidatePoolMultiplier": 1.4,
  "minimumOutreachEligible": 31,
  "crawlOfficialWebsites": true,
  "usePlaywrightFallback": true,
  "maxPagesPerWebsite": 8,
  "strictCityMatch": true,
  "requireWebsite": true
}

maxPagesPerWebsite defaults to 8 and can be set from 1 to 12. With the defaults, a 50-lead run evaluates up to 70 accepted websites before selecting the final 50. This can increase discovery and crawl cost by roughly 40%, so first measure it on one 50-lead pilot. minimumOutreachEligible is a reporting target; it never weakens a review rule or guarantees the result.

Important output fields

The dataset keeps the existing leadfinal identifiers and business fields. V3.2.1 adds or stabilizes:

• email, public_emails, alternate_public_emails, other_location_emails, review_emails
• email_candidate_count, email_candidate_evidence
• email_source_type, email_source_url
• email_domain_status, email_verification_status, email_role_type
• email_usefulness_status, email_outreach_eligibility
• email_location_status, email_location_score
• email_page_purpose, email_confidence_status
• contact_review_status
• candidate_pool_order, candidate_pool_size, reserve_count, selected_beyond_initial_limit
• selection_rank, selection_status, selection_reason
• phone, public_phones, phone_evidence_status
• website_crawl_status, pages_crawled_count, crawl_failure_reason, crawl_http_status, crawl_retry_count, crawl_recovery_method
• discovery_quality_status
• quality_flags, quality_flag_count, quality_flags_text
• discovery_source, website_evidence

Recommended n8n gate for the normal clean-email path:

discovery_quality_status = APPROVED
AND contact_review_status = EMAIL_FOUND_CLEAN
AND email_outreach_eligibility = ELIGIBLE
AND email_confidence_status IN (PUBLIC_SOURCE_HIGH_CONFIDENCE, PUBLIC_SOURCE_MEDIUM_CONFIDENCE)
AND email is not empty
AND email_source_url is not empty

Use public_emails[0] or email only after this complete gate passes. A non-empty email can intentionally remain visible on a review record; that is evidence for a human, not permission to send.

For the strictest branded-email subset, also require:

email_domain_status = WEBSITE_DOMAIN_MATCH_UNVERIFIED
AND email_confidence_status = PUBLIC_SOURCE_HIGH_CONFIDENCE

Send REVIEW_EMAIL_DOMAIN, REVIEW_EMAIL_DOMAIN_ALIAS, REVIEW_EMAIL_PAGE_PURPOSE, REVIEW_EMAIL_USEFULNESS, and REVIEW_EMAIL_LOCATION records to manual review. Never send NO_PUBLIC_EMAIL_FOUND records into an email campaign. Do not automatically merge alternate_public_emails, other_location_emails, or review_emails into an outreach list.

Run summary

RUN_SUMMARY separately reports the evaluated pool, selected and reserve counts, public emails found, clean public-source emails, outreach-eligible emails, confidence levels, fully automation-ready emails, role/domain/page/branch review counts, missing emails, and whether the configured target was achieved.

REJECTED_LEADS stores hard-rejected records separately.

RESERVE_LEADS stores fully evaluated candidates that did not enter the final maxLeads dataset. They are evidence-backed backups, not automatic additions to an outreach campaign.

Intentional boundaries

• No guessed addresses or permutations.
• No SMTP mailbox probing or deliverability claim.
• No Hunter, ZeroBounce, LinkedIn automation, subscription forms, support chat, or outreach.
• The source Actor is a replaceable adapter. Follow source terms, website terms, privacy law, suppression requirements, and outreach law for every market.

Deploy and pilot

1. Authenticate the Apify CLI with the intended account.
2. From this project folder, run npm test.
3. Run npx apify push --open.
4. Start a new 50-business pilot with the input example.
5. Confirm both clean_public_email_count and outreach_eligible_email_count in RUN_SUMMARY.
6. Inspect every review bucket before connecting the strict n8n gate.

For lakhs of leads, split work by category and geographic job in n8n. Keep discovery, deduplication, contact recovery, later risk checks, and outreach eligibility as separate stages.