skip-tracer

Skip Tracer

Overview

This skill takes a list of people or properties and returns, for each row, the best current phones, emails and mailing address it can find, a match band with the evidence behind it, the tool and identifier that produced the match, a call status for every phone, and a report that prices each step before it runs.

It prevents four mistakes:

1. Searching the weakest identifier. A name alone matches many people and People Finder bills every result. An address with a ZIP or a phone number narrows the search to a household or a subscriber.
2. Paying $0.25 a row for Property Lookup with owner contact on every address when a $0.0199 People Finder search answers many of them. This skill runs the cheap search first, measures its match rate on a pilot, and escalates only weak matches and rows with no known owner.
3. Calling the wrong person. An address search returns the people tied to that address, which can be a tenant or a former owner. Every match is banded by what the input already knows before anyone dials.
4. Paying twice for the same search. Two filings on one property, or one owner on several properties, are deduped before the run and grouped after it.

When to Use

• "Here is a list of property addresses. Find the owners and their phone numbers."
• "Skip trace my pre-foreclosure list."
• "We have names and old addresses for past customers; find current contact info."
• "Reverse these phone numbers to names."
• "Pull distressed properties in Columbus and get me owner contacts."

Use a different skill when:

Prerequisites

Authoritative Foundations

Details and routing: references/framework-notes.md.

Workflow

1. Decide whether a skip trace is needed

For mail to property owners, the property address (and the owner mailing address from Property Lookup) may be enough. Run a skip trace when the plan includes calls or email, or when the owner's mailing address is unknown.

2. Profile and normalize identifiers

Normalize addresses to USPS format ("4410 W Linda Ln, Tampa, FL 33629": number, directional, street, suffix, unit, city, two-letter state, ZIP), phones to E.164, emails to lowercase. Dedupe search values. Then class each row by its strongest identifier:

Each array entry is its own search, and every returned result is billed, so max_results is 1 wherever the identifier points to one household or subscriber. For name searches, set max_results to the number of same-name candidates you are willing to pay to compare. The catalog does not document pairing a name and an address in one search, so search the stronger identifier and use the other to band.

3. Price it and pilot

find_people ceiling   = searches x max(max_results, 1) x $0.0199
property lookup       = properties x $0.15, or x $0.25 with includeOwnerContact
distress list         = maxItems x $0.05

Pilot a random sample per class (the size is your call; 25 People Finder searches cost $0.50). From the pilot, compute cost per usable owner for each path: People Finder cost per search divided by its share of rows reaching probable or better, against $0.25 divided by Property Lookup's share. The cheaper path per usable owner runs on that class. Run check_wallet, show the estimate, and wait for a go-ahead before the paid run.

4. Run the strong searches

Send every search of a class in one run. Check check_run_status once, report the status, and check again on the next turn; then download_results the same day. Keep a run log: date, tool, class, inputs, runId, billed results, cost.

5. Band every match by its evidence

Compare each returned candidate with what the input already knows, using only fields present in the downloaded CSV header. Record the evidence in match_evidence and band from it:

Then check a random sample of confirmed rows against the county auditor or assessor's owner of record and write the result as one QA line. When missed rows share one evidence type, move that evidence type down one band and re-band the file (references/skip-trace-runbook.md section 9).

6. Rescue and escalate

The plan budgets one rescue per row. A second pass is another billed search, so price it before adding it.

7. Phone order and suppression

Normalize every phone to E.164 and keep line type where the data gives it. Order each row's phones by the recency field the CSV holds, newest first; once the first call block has outcomes, order line types by the connect rate you measured for each. A phone that shows up on two or more unrelated owners is a shared line (property manager, office): flag it and dial it last. Remove every number on the suppression list. Mark each phone callable, suppressed, estate_mail, or mail_only when the row's band is below probable.

8. Group owners, hand off, read the outcomes

1. Group rows by matched owner. An owner on two or more input properties is flagged PORTFOLIO, takes the strongest match across their rows, and goes to the top of the call queue with every address in one call.
2. Write three files: the call queue (callable rows, portfolio owners first, then confirmed, then probable, phones in order, an outcome field), the mail file (every input row, owner name or "Current Owner", the owner's mailing address when it differs from the property), and an email file only when email is a planned channel. Columns: references/skip-trace-runbook.md section 11.
3. Set up the dialer outcomes before the first call block: right_party, relative_or_tenant, wrong_number, disconnected, no_answer, remove_request. A remove_request goes on the suppression list the same day, every number on the row, in every tool. After the first block, compare the wrong-party share of probable with confirmed and decide from those numbers whether probable rows keep their calls next run (section 10).
4. Set a re-trace age before each new campaign from your own wrong-number rate on older rows. Write the report from templates/output-template.md and run scripts/check-output.py.

ScraperCity Implementation

Tools and prices (from references/scrapercity-catalog.md)

Calls

{"tool": "scrape_distress_list", "args": {"search": "Columbus, OH", "distressTypes": ["preForeclosure", "auction"], "maxItems": 150}}
{"tool": "find_people", "args": {"street_citystatezip": ["2217 Elmwood Ave, Columbus, OH 43211", "88 S Harding Rd, Columbus, OH 43209"], "max_results": 1}}
{"tool": "find_people", "args": {"name": ["Darnell Hughes"], "max_results": 3}}
{"tool": "find_people", "args": {"phone_number": ["+16145550173"], "max_results": 1}}
{"tool": "scrape_property_lookup", "args": {"addresses": ["1406 Oakridge Dr, Columbus, OH 43224"], "includeOwnerContact": true, "fileName": "cbus-distress-owners"}}

Skip bankOwned in distressTypes for owner outreach: a bank-owned (REO) property belongs to the lender, so a skip trace reaches former owners. CLI for People Finder: npx scrapercity people-find with the name, email, phone or address options (several addresses separated by |), and npx scrapercity distress-list "Columbus, OH". Package 1.0.17 has no Property Lookup CLI command; use MCP or POST /api/v1/scrape/property-lookup.

Worked cost estimate

300 county pre-foreclosure filings with owner name and property address, plus 150 Distress List properties with no owner name, Columbus, OH.

Ceiling before the run: rescue at 71 x 3 = 213 results ($4.24), so $55.21. The alternative of Property Lookup with owner contact on all 450 addresses would cost 450 x $0.25 = $112.50 plus the Distress List.

Errors

Output Format

1. skiptrace-[segment]-[date].csv: input columns, match_evidence, match_band, matched_by (class and tool), matched_name, current_address, phone_1 to phone_3 (E.164), phone_n_status, email_1, email_2, owner_type (person, llc, trust, bank, estate), flags, source, pulled_at.
2. calls-[segment]-[date].csv (call queue) and mail-[segment]-[date].csv (every input row with the mailing name and address).
3. A report from templates/output-template.md: purpose and channels, input profile, identifier plan, cost, match bands with the QA line, funnel, contact results with portfolio owners and shared lines, sample rows, next steps with the dialer outcome setup.

What the reader receives is the deliverable only: product names (People Finder, Property Lookup, Distress List, Email Validator), never tool ids, internal file paths or reference section numbers; another skill is named only in a handoff line that tells the user what to run next; no tables of checks that were not run (one "Not checked:" line instead); no repeated identical rows where one template plus a list will do.

Quality Check

• Purpose and channels written in one table.
• Every search used the strongest identifier the row had, and max_results is 1 for address, phone and email searches.
• Cost lines equal billed results x $0.0199 for People Finder and properties x $0.15 or $0.25 for Property Lookup.
• Every row has evidence and a band; the bands sum to the rows searched; each sample row's band matches its evidence.
• No weak or no_match row has a phone marked callable.
• Every callable phone is E.164 and off the suppression list.
• Entities (LLC, trust, bank) and deceased flags are routed away from person outreach.
• The QA line shows how many sampled confirmed rows match the county owner of record.
• Portfolio owners and shared lines are counted in Contact Results, and the call queue lists portfolio owners first.
• Next Steps sets up the six call outcomes and the band review before the first call block.
• The report names products, not tool ids or file paths.
• scripts/check-output.py report.md prints PASS.

Common Pitfalls

Execution Artifacts

Related Skills

• scrapercity-real-estate-data for distress lists, valuations and agent lists.
• mobile-appender for B2B mobiles from LinkedIn URLs or work emails.
• cold-calling for scripts, identity checks and cadence.
• crm-enricher to write results back to a CRM with audit columns.
• email-list-cleaner to validate found emails before any email step.
• scrapercity-mcp for MCP setup.