Can public records find you customers?
Sometimes — and far less often than the data’s availability suggests. Nearly every regulated industry publishes a dated record of who just entered it, and almost nobody uses it. That gap looks like an opportunity, and occasionally it is. More often the data is public because it is not commercially valuable, or someone has been quietly selling it since before the web. This is the test for telling those apart, written by someone who has applied it to about a hundred markets and killed most of them.
The four conditions
All four have to hold. Three out of four is a market that looks promising right up until you have built the thing.
| Condition | Looks like | Fails when |
|---|---|---|
| A public, dated event | A license issued, a permit pulled, a filing accepted, a well completed — with a date attached. | A roster of everyone who currently holds a license. A list of who exists is a directory, and directories are already sold cheaply by everyone. |
| The event comes before the purchase | A new contractor licence is issued, and general liability insurance gets bought in the weeks after. | The purchase is a prerequisite for the licence. If your product must already be in place before the licence can be granted, the sale was lost before the record ever appeared. |
| A fragmented, quiet buyer | Hundreds or thousands of small operators who each feel the cost of a lead directly and can decide alone. | A handful of large buyers with procurement committees, or an industry where one platform already owns distribution. |
| Volume that justifies the build | Enough events per month, in the geography you sell to, that working them is a real habit rather than an occasional novelty. | A dozen a year nationally. Real, interesting, and not worth a pipeline. |
What tends to be out there
If your market is licensed, permitted, inspected or registered by a state or county, there is almost certainly a public record of it. Quality varies enormously between neighbouring states that regulate the same trade.
| Kind | Examples | Timing quality |
|---|---|---|
| Occupational & trade licences | Contractors, electricians, cosmetologists, home-care agencies | Usually dated; some states publish issue dates, many require diffing |
| Permits & completions | Building permits, water-well completion reports, environmental filings | Dated, but filing deadlines create lag — sometimes 60 days |
| Business & entity filings | New LLCs, nonprofit determinations, registered agents | Well covered by incumbents; check before assuming a gap |
| Facility & site registrations | Care homes, childcare centres, food service, tobacco retail | Varies wildly by state; several publish nothing usable |
The hard part isn’t collecting it
Getting the data is a solved problem. Knowing whether to trust it on any given Tuesday is not, and that is where these projects actually fail.
- Sources go stale silently. A portal that stops updating still returns a perfectly valid file. Without a check against live ground truth, a pipeline will serve last month’s data with total confidence for as long as you let it.
- Published late is not the same as issued late. Some jurisdictions publish well after the event. If you label a week’s worth of records by the week you fetched them, some of them are simply in the wrong week — and they are never recovered, because every later window is later still.
- Most sources need a warm-up period. Where no issue date is published, “new” has to be derived by comparing snapshots, and the first useful output is a month or two away. This cannot be backfilled.
- Coverage is not readiness. A source you can fetch is not the same as a source you should sell from. It is entirely reasonable for a rigorous assessment to clear only a handful of sources out of dozens on the first pass — that is the check working, not the project failing.
- The join is where it really goes wrong. One record says a licence was issued. Matched against the business-filing registry it also tells you the company is four months old and who signed for it — which is the difference between a name and a reason to call. But matching on company name alone produces collisions, and matching on address alone returns whoever the registered agent is. Get it wrong and two businesses silently become one row that looks entirely correct. The defensive version matches on a normalized dual key and leaves anything ambiguous marked as a candidate rather than merging it.
When it doesn’t work
Most candidate markets die on one of these. Each is cheap to check and expensive to discover late.
The prerequisite trap
The single most common way this fails, and the least obvious. If what you sell has to be in place before the licence is granted — insurance, a bond, a compliance plan, equipment — then by the time the record is public, your prospect has already bought from someone else. The public event looks like a starting gun and is actually a finish line.
Someone has quietly sold it since 1998
“Nobody is doing this” is usually wrong, and the incumbent is rarely a funded startup you would find in the tech press. It is more often a thirty-year-old business selling an enriched version of the same list to the same buyers, invisible to anyone searching the way a technologist searches.
The buyer can't capture the value
In some trades, efficiency gains flow straight through to the customer as price compression. Your prospect works faster, earns the same, and has no reason to pay for the tool that did it. Check who keeps the money before building anything.
The government does it free
Agencies increasingly email their own licensees with renewal reminders and bulk services. It is difficult to charge for something the issuing authority already provides at no cost, and worth ten minutes of checking before it becomes a discovery.
The pain is real but mis-attributed
People will describe a problem vividly and still not buy, because the thing they fear is not the thing you fixed. Verify that the expensive event is the one your product prevents, not merely the one that comes to mind first.
For what it’s worth, this list is not theoretical. It is the accumulated wreckage of my own candidate markets — assessed properly, then abandoned, mostly before writing any code. The assessment is cheap. Finding out afterwards is not.
Common questions
Is it legal to use public records commercially?
Usually, but not universally, and the exceptions are specific rather than general. Some states explicitly restrict commercial solicitation directed at people identified through their own rosters; some datasets carry terms of use that differ from the underlying public-records law; a few publish data with an explicit non-commercial disclaimer. The right approach is to check per source before building on it, exclude the ones that don't clear, and say plainly which those were. Anti-spam law applies on top of all of that, independently.
How current can this data actually be?
It depends entirely on the source, and the honest answer is often less current than it appears. Filing deadlines can allow weeks between the real-world event and its publication, and some jurisdictions publish on a lag of ten days or more even when the record itself is dated. That doesn't make the data useless — it makes the difference between saying “issued this week” and “published this week,” and only one of those is defensible.
What if the source doesn't publish issue dates?
Then “new” has to be inferred by comparing snapshots over time, which means the pipeline produces nothing at all for its first month or two while it accumulates a baseline. There is no way to backfill this. It's the single most common scheduling surprise in this work, and it's the reason collection should start well before anyone needs the output.
Can't I just buy a list instead?
Often yes, and sometimes you should. Bought lists are cheap and immediate. What they generally aren't is fresh or exclusive — by the time an aged-lead vendor has a record, it has usually been sold several times over. The case for building is timing and exclusivity; if neither matters in your market, buying is the rational choice and I'd tell you so.
Why would this need maintaining after it's built?
Because government sources change without warning or notice. Portals migrate, fields get dropped, URLs move — sometimes within days — and an unmaintained pipeline tends to fail quietly rather than loudly, still running, still reporting, no longer correct. Assume meaningful annual decay in any set of sources and budget for someone to notice.