Four programs, in detail

Each of these started the same way, with a client explaining who they needed on the phone and why the lists they had bought before were not getting them there. How each program is built stays between us and the client who paid for it, so what follows is the brief, what was delivered against it, and what we measured on the way.

Roofing contractors in Florida

A roofing call center in Florida wanted homeowners worth calling, which is a narrower thing than homeowners. Public records will give you the address and the owner’s name, and both will be correct. What they will not tell you is whether the roof needs replacing, which is the only reason for the call.

So the qualifying work happens before anyone gets a list. A property only becomes a lead once there is reason to believe the roof is a live problem, and the contact details get attached to the ones that survive that.

• In the largest county we built, 214,335 properties were reviewed and 94,398 qualified homeowners came out.
• 245 of 246 records matched correctly when cross-checked against a second independent public source, a 99.6% join.
• Phone append lands on 63.9% of records. That is measured across real batches, and it is why roughly 1.57 records enter that step for every one we need to deliver.
• The batch we delivered came to 5,000 records across seven ZIP codes with recent wind events, 714 in each.

In some counties one of the qualifying checks cannot be completed at all. We could leave those records looking identical to the rest and no buyer would ever know. Instead the file carries a column saying the check was not available there, so nobody assumes a roof passed a test it never took.

Appended phone numbers also get a consistency check against where the property actually is. Append providers are measured on hit rate rather than accuracy, so given a common name they will hand you a live number belonging to the wrong person in another state before they will hand you an empty field. On this program anything that fails that check is dropped instead of delivered.

Insurance agency owners across the US

This client sells offshore staffing to small agencies, both B2B and B2C writers. The buyer is the owner, and an agency’s published phone number reaches a front desk whose job includes stopping that exact call. The owner’s name is usually nowhere on the agency’s own listing either.

That made it two jobs rather than one: build the universe of agencies, then find the person who owns each one, which is a separate problem with a separate answer.

• 82,700 agencies across the first two states in scope.
• 58,893 of them with phone, email and postal address all present on the same record.
• 41,264 with the agency website identified.
• 21,062 matched to a named owner.
• 802 owner-direct contacts.

That last figure deserves a note, because it started at 4,007. Once we tightened what counts as a real person’s name, three quarters of it fell away. The 3,205 rows we dropped were the ones that would have gone out addressed to a filing status or a department.

Two of the things this client wanted to filter on do not exist as public facts. Whether an agency has hired offshore before, and whether it could afford to, are inferences. We score them and deliver them as a flag to sort by, and we do not use them to quietly delete half a list. Headcount is an estimate as well, confirmable on somewhere between half and two thirds of the file and marked as unconfirmed on the rest.

Toll-free numbers, with the business attached

A toll-free number tells you nothing about where its owner is. That is the appeal for the business buying one and the entire problem for anyone building a list off them. City and state can only come from the company behind the number, no public file pairs toll-free numbers with their owners, and the official administrator holds that database without publishing it.

So attribution is the whole job here. A toll-free number on its own is worth nothing to a sales team, and the part nobody will hand you is which business is on the other end of it. That is what this program delivers: the number, the company, the city and the state, together on one row.

• 827,049 unique toll-free numbers found so far.
• 581,679 of them tied to an identified business.
• 91,286 in the top tier, where the number is still in active public use by that business today.
• In the pilot we measured, 81% of records resolved to a company and 53% all the way to a state.
• Delivery currently runs at 10,000 records a week, on a standing schedule.

Yield per unit of work falls as the program advances. Early on it returned around 15 usable rows; across the last stretch we measured it was closer to half a row. That is why this runs continuously against a weekly delivery schedule rather than as an extraction you do once and then sell from.

A number ships only when it can be pinned to a business with a name. At one checkpoint that rule took the deliverable set from 27,783 rows down to 20,252. Everything that came out was a row where the company name turned out not to be a company: a category label, a government helpline, or, in 124 cases, a question a customer had typed somewhere.

Residential energy in Pennsylvania and Massachusetts

A BPO selling energy supplier switching needed two things that are not fields you can request: the person who actually pays the bill, and some indication of who would rather take the call in Spanish.

The first is confirmable. The second is not a fact anywhere, so it ships as a score rather than a filter, and the client’s team decides where to draw the line.

• Around 161,000 owner-occupant households across the target cities in both states.
• Around 20,000 of them carrying a bilingual signal.
• By market: 43,412 in Lancaster, 35,510 across Massachusetts with 7,117 bilingual, 31,942 in Harrisburg, 14,831 in the Philadelphia target ZIP codes, and 7,815 in Reading, where close to half the file carries the signal.

We told this client at the start that the universe covers a fraction of the market they were aiming at. It is built from owner-occupant records, the target ZIP codes hold a lot of renters, and renters pay energy bills too. Reaching them takes paid consumer data, which became a decision for later rather than something to hide inside the count.

One source in this build arrived with an invalid postal code on 83% of its rows. New sources get audited for field completeness as they load rather than at delivery, so it surfaced the same afternoon and was repaired that day.