What You Actually Get When You Scrape a Business Directory

Getting the records out is the easy part. Here's how much of a directory extract arrives usable, why the same company shows up three times, how fast it goes stale, and what you owe the people in it.

By Chirag Jakhariya, Co-founder & CEO · 10 September 2026 · 8 min read

The export finished and the list still isn't a list

The export finished. Four thousand rows, and every column has a header. You open it, sort by company name, and start reading - and somewhere around row 200 it stops looking like a list of prospects and starts looking like a spreadsheet you're going to have to do something about.

This is the normal outcome, and the extraction usually isn't what went wrong. Getting records out of a business directory is the easy part of this job. What follows is the part nobody sells you: how much of that file arrives usable, why the same company shows up three times, how fast it goes stale, and what you owe the people in it.

Listings, leads and contacts are three different things

A listing is a business. A name, an address, a phone number that rings a reception desk.

A contact is a person. A named human with a role and a way to reach them directly.

A lead is a person with a reason to talk to you.

Directories reliably give you the first. They sometimes give you the second. They never give you the third. If you're paying lead prices for listing data, that gap is where the money goes - and it's the single most common reason a bought list underperforms.

Worth saying where we stand, because it shapes everything below. We build business-level extracts - the organisation, not the individual - and we decline work that means collecting personal or identity information about people. So read the rest of this as what you'll be dealing with in a directory file, and where the line falls for us.

Cause one: the fields arrive empty

You can extract twenty fields. That's not the same sentence as twenty fields arriving populated, and the difference between them is the whole problem.

The most useful measurement we've found comes from a LocaliQ study published in February 2026. It's small - 100 US home-services businesses, graded with LocaliQ's own listings tool - so treat it as a shape rather than a benchmark. With that said, the shape is stark. Only 15% of the listings they checked had completely correct information.

Look at the two columns rather than the numbers. A profile the business claimed and maintains itself is missing a phone or a website around 10% of the time. The same field across aggregated directories is missing four times as often.

That tells you something more useful than any accuracy percentage: which source you pull from moves your fill rate far more than how well you extract. A perfect scraper pointed at a thin directory returns a thin file. We've had projects where changing the source improved the usable-record count more than every other change combined.

One thing the table can't show you. Most directory records don't carry a named person or a direct email address at all - that's not a fill-rate problem, it's a category problem, because directories exist to list businesses and not staff. We're not putting a percentage on it, because we couldn't find a credible published one and we're not going to invent one.

What to do: before you commission anything, ask for 200 records with the fill rate per field attached. Not a sample of what the fields are called. A count of how many came back populated. Any vendor who has done this work has that number, or can get it in an afternoon.

Cause two: the same business, three times

Pull from two directories and you'll get overlap. That's expected. What surprises people is that reconciling the overlap is a judgement call, not a script.

Here's the case that breaks automatic matching. The same company appears once at a trading address and once at its registered office, under two spellings, with two phone numbers - one of which still works. Name-and-address matching says these are different businesses. A human says they're obviously the same. Postcode matching says they're different too, and postcode matching is what most dedup runs on.

You'll see duplicate rates quoted for this. We've quoted them ourselves. Follow any of them back and they cite each other rather than a study with a method behind it, so the only number worth having is the one you measure on your own file.

What to do: decide the match rule before extraction, not after. Which field wins when two records disagree, and what a human reviews. Writing that down takes twenty minutes and it's the difference between a dedup you can defend and one you'll redo.

Cause three: it was already out of date

Every vendor in this market quotes you a decay rate. Follow one to its source and it gets interesting.

A Cleanlist post from January 2026 puts B2B data decay at 22.5% a year and credits Dun & Bradstreet, with a link. That link returns a 404.

The figure is real, but it isn't D&B's. It comes from HubSpot's Database Decay Simulation, which states that "Marketing Sherpas research shows that B2B data decays at a rate of 2.1% per month." No date. No methodology. No link to the study. A footnote naming a research firm is the entire provenance of the most-quoted number in this industry.

Meanwhile ZoomInfo publishes a range of 22.5% to 70% for the same measure, and an older industry deck puts it above 5% a month. A threefold spread isn't disagreement about an answer. It's a sign that these aren't measurements of the same thing - different fields, different sectors, different definitions of "wrong".

None of that means data doesn't go stale. It obviously does. It means the industry number can't tell you how fast your file will, because nobody has measured your segment. A directory of long-established trade businesses behaves nothing like a list of contacts at venture-funded startups, and one annual percentage covering both is worth very little.

What to do: take a 100-record slice, put a date on it, and re-check it at 90 days. That gives you a decay rate for your sector, measured on your data, in about an hour of work. It's the only version of this number that should change what you do.

The one with no clean fix: the notice you owe

Here's the part that gets skipped, and it's the reason this section has no tidy answer.

When you collect someone's details from a directory instead of from them, you take on an obligation you wouldn't have had otherwise. Under GDPR Article 14, where personal data is obtained from a source other than the person, you have to tell them - what you hold, where you got it, and why - generally within a month, and before or at the point you first contact them.

Legitimate interest does cover B2B outreach. It's a real legal basis and it's the one most B2B prospecting runs on. It isn't a free pass, though. It needs a documented assessment weighing your interest against theirs, and it needs the contact to be genuinely relevant to that person's job. Companies have failed that test on relevance while believing they had it covered.

The direction of travel is tightening, not loosening. In July 2026 a major CRM vendor changed its terms to pool customer contact data for enrichment by default, and reversed it four days later after its own customers objected. Their chief product officer's line was "we made a mistake." That's the expectation you're operating inside now.

So why no clean fix? Because the obligation doesn't disappear, and most outreach tooling gives you nothing to meet it with. The workable answer is dull: a privacy notice you can point to, a source recorded against every record so you can say where it came from, and the notice going out with first contact rather than being promised and forgotten. We record the source against every record as a matter of course, because retro-fitting provenance means going back to the source and re-running the job. The rest of it - the notice, the assessment, the decision to contact anyone at all - sits with you. That isn't us dodging it. It's where the obligation lands.

None of this is legal advice, and the specifics change with where you and your prospects are. If your outreach is regulated or you're working at real scale, get it looked at properly. What we say about the legal side of collection is in our FAQ, and it's the same answer we give on a call.

When a directory is the wrong source

Three cases where we'd tell you not to do this.

• You need a named decision-maker. Directories list businesses. If your outreach only works when it reaches a specific role, you're starting from the wrong place and you'll spend more on enrichment than the extract cost.
• Your market is small enough to count. Under a few hundred targets, a person with a browser and two days will beat a pipeline, and the file will be better. We've said this and lost the work, which is the correct outcome.
• The directory is a copy of a copy. Some aggregators republish others. Extracting from one means inheriting every error in the original, with a longer lag. If you can't establish where a directory's data originates, that's a reason to look elsewhere.

Where a directory is the right source, the work is genuinely worth doing, and we do a lot of it - that's what our directory extracts look like, and matching and cleaning records is usually the larger half of the job.

The decision isn't whether to scrape a directory. It's whether you know what's going to come back before you pay for it. If you want to find out for your sector, ask us for 200 records with the fill rate per field attached, and judge the job on that.