Build your own list

Decide who you are actually selling to

The register records what a company does, not who works there. Getting this step right is most of the work, and it is the cheapest place to change your mind.

The 3 ways to describe a market

By	What it means	Use it when
Industry code	The 5-digit code a company declares for its trade	You know exactly which trades you want and need every company in them
Trade or industry name	Plain words ("plumbers", "care homes") matched to the codes behind them	You know the trade but not the codes. This is how most people start.
Company number	A specific list of companies you already have	You have a list of names and want the contact details filled in

What the codes are

Every company on the register declares at least one 5-digit Standard Industrial Classification code (SIC code) describing its trade. It is how the government classifies what a business does. The full list is published by the Office for National Statistics and there are just over 700 codes in active use.

A company picks its own codes when it registers and can declare up to 4. Nobody checks them, so they are a good guide and not a guarantee.

Two more worked examples, so you can see the shape:

Code	Trade	Companies
43220	Plumbing, heat and air-conditioning installation	47,075
45112	Sale of used cars and light motor vehicles	32,965
45200	Maintenance and repair of motor vehicles	54,591

If your buyer is a job title

You can target "used car dealers" because that is a thing a company is. You cannot target "marketing directors" or "companies with more than 10 staff" directly, because the register holds neither.

That does not make this useless to you. The register is still the fastest way to get a complete, current list of the companies worth approaching: every active business in your trade, in your area, with its website. Finding the right person inside those companies is a second job, usually done on a professional network, and it is far easier once you have a defensible list of which companies to bother with.

Download the register

The whole United Kingdom company register is published free, in full, once a month. Most people do not know this, and it is the single most useful fact in this guide.

It is here: download.companieshouse.gov.uk/en_output.html

Take the single-file version. It is a zipped CSV of roughly 2.6GB unpacked, about 5.7 million rows, and it contains every company currently on the register: active, dormant, in liquidation and proposed for strike off. Dissolved companies have been removed and are not in it. It is a complete snapshot, not a list of changes, so you never need to stitch months together.

What is in it, and what is not

You get	You do not get
Company number and registered name	Any email address
Registered address, town and postcode	Any website
Up to 4 trade codes	Any phone number
Company status and type	Turnover, headcount or any trading detail
Incorporation date	Names of staff below director level

A note on the registered address

The address on the register is where the company is registered, which for a large number of small companies is their accountant's office. It is not necessarily where they trade. Treat it as a rough geography, useful for "within 25 miles of here" and not much more.

Filter it to your market

A spreadsheet will not open 5.7 million rows. Here is how to do it without one, and it is about 10 minutes if you have never done it before.

Load it into a database: the whole thing, 3 commands

SQLite comes preinstalled on macOS and most Linux systems and is a free download on Windows. From a terminal, in the folder with the unzipped CSV:

Type this	What it does
sqlite3 register.db	Creates an empty database and opens it
.mode csv	Tells it the file is comma-separated
.import BasicCompanyData.csv companies	Loads all 5.7 million rows into a table called companies. Takes a couple of minutes.

The columns you actually need

The file has over 50 columns and you will use 6 of them. They are named exactly like this, dots included:

• CompanyName: the registered name
• CompanyNumber: the unique identifier
• CompanyStatus: Active, Dissolved, Liquidation and so on
• CompanyCategory: the company type
• SICCode.SicText_1 through _4: the trade codes, as "69201 - Accounting and auditing activities"
• RegAddress.PostTown and RegAddress.PostCode: geography

The 4 kinds of company to remove first

1. Anything not actively trading. Keep CompanyStatus = 'Active' only. Dissolved, liquidation and in-administration companies are all in the file and all a waste of a send.
2. Dormant companies. They show as Active and trade nothing. They declare code 99999, and across the trades above they are roughly 7 to 10% of each code.
3. Company types you cannot sell to. Limited Partnerships, and depending on your offer, companies limited by guarantee and community interest companies. That is CompanyCategory.
4. Companies whose match is incidental. A company can declare 4 codes. One whose fourth code matches yours is probably not in your trade. Matching on codes 1 and 2 only is usually the right call.

The query, in full

For the 3 accountancy codes, active companies only, no dormant, in Dorset:

Change 6920% for your own codes and DT% for your own postcode area. To write the result to a file, type .mode csv then .output mylist.csv before running it.

Find each company's website

This is the first of the 2 steps that stop being a spreadsheet job. There are 3 ways to do it and they differ by about a factor of a hundred in effort.

Route	What it costs	Sensible up to
By hand	About 20 seconds a company, searching and pasting	~200 companies
Build it	A day or so of code, then a few pounds per 2,000 in search fees	any size
Use a tool	Ours, or a competitor's	any size

200 companies by hand is about an hour and there is nothing wrong with it. Below that number, stop reading this step and just do it.

The obvious shortcut, and its ceiling

The obvious move is to guess: take the registered name, strip "LIMITED", remove the spaces, try .co.uk then .com. It is free and instant, and it works less often than anyone expects.

Trade	Guessing the domain from the name
Sale of used cars and light motor vehicles	7.5%
Accounting and auditing activities	13.3%

Both figures are end-to-end (name in, working contactable website out), measured on real cohorts. VANSTOCK NORTHWEST LIMITED trades at vanstocknw.co.uk, and no rule working from the legal name arrives at that abbreviation.

What actually works

Searching for the company by name plus town and reading the first page of results. It finds the sites guessing cannot, because it uses the same signal a human would. It is also the first thing in this process that costs money: a search API is roughly a seventh of a penny per company, so 2,000 companies is a couple of pounds.

The barrier is not the money. It is that this is one request per company, and you cannot do 2,000 of them in a spreadsheet.

Using an assistant for the judgement

Deciding whether a domain belongs to a company is the part a rule cannot do, and it is the part an assistant like Claude is good at. Paste in a batch of names with the candidate domains you found, and ask it to judge them. A prompt that works:

Get the contact details off the site

This is the step that needs code. There is no spreadsheet formula and no reliable no-code tool for it at volume, and pretending otherwise would waste your weekend.

What to build, in 6 lines

1. Read robots.txt for the site, and stop if it disallows you.
2. Fetch the home page.
3. Find links whose text or URL contains contact, about or get in touch.
4. Fetch at most 4 more pages from that list, never the whole site.
5. Pull out every mailto: link and every string matching an email address.
6. Keep the ones whose domain matches the site you are on. Throw the rest away.

In Python that is the requests library to fetch, BeautifulSoup to read the page, and the built-in re module for the address pattern. All 3 are free. It is roughly 80 lines, and the last of the 6 rules above is the one that matters most.

How to fetch without getting blocked

• Honour robots.txt. Not only for legal reasons; it is how you stay unblocked.
• Identify yourself in the user agent string, with a URL explaining what you are and how to stop you. A blank or fake user agent is what gets a whole address range banned.
• One request per second per host, maximum. Run many different hosts at once instead; that is where throughput comes from.
• At most about 5 pages from any one site. If the address is not on the home page, the contact page or the about page, it is not there.
• Set a timeout, and move on. A dead site should cost you 10 seconds, not a stalled run.

What you are looking for, in order

1. An address on a contact page whose domain matches the site you are on
2. An address in the footer, which is usually the general enquiries one
3. A phone number, which is worth keeping even when you find an email

What it costs in time: the number that decides everything

Measured on a real run: 2,773 companies took 459.6 minutes end to end, with 8 workers running against different hosts and a 1-second-per-host limit. That is about 10 seconds a company, or about 360 companies an hour, and it does not get faster by adding workers, because the limit is politeness, not processing.

List size	By hand	Running continuously
200 companies	~2 hours	about 33 minutes
2,000 companies	~2 weeks	about 5.5 hours
10,000 companies	not realistic	about 28 hours, more than a day

And what you end up with

Most of the companies you started with will not have an email address you can actually use. They publish a contact form, a phone number, or nothing at all, or nothing on their site proves the site is theirs.

That is not a failure of method. That is what the register looks like. Anyone quoting you a 90% contactable rate is quoting you something that was not measured.

What you owe the people on your list

The file is yours now, and so is the legal duty for it. Not your supplier's, not your sending tool's. And buying the list from somebody else does not move it.

1. You are the data controller. You decided who to collect and why, so the responsibility is yours.
2. A lawful basis is required, and it has to be written down. For business-to-business outreach that is normally legitimate interest, which means actually balancing your interest against theirs and keeping the reasoning, not simply asserting it. A paragraph in a file is enough. Nobody asks to see it until something goes wrong.
3. Article 14 applies, because they did not give you the data. When details are collected from a third source there is a duty to tell people where they came from, and a time limit on doing it. In practice: one line in the first email and a privacy notice they can reach.
4. A named individual's work address is personal data. jane.smith@firm.co.uk is a person. info@firm.co.uk is a weaker case, but weaker is not none.
5. Sole traders and partnerships are individuals, whatever the register calls them, and they get the stronger protection.
6. Keep a suppression list and honour it immediately. One unsubscribe removes that address from everything you will ever send, not from one campaign.

Whether to build it or buy it

Every number required to decide is now in one place.

Step	What it costs you
Finding your trade codes	An afternoon, once
Downloading the register	Free
Loading and filtering it	About 10 minutes, with the commands in step 3
Finding the websites	A few pounds per 2,000, plus building the matcher
Fetching the contact details	about 5.5 hours per 2,000, about 28 for 10,000
Staying inside the law	Yours either way. It does not transfer.

Most people who read this far will not do it, and that is a reasonable decision rather than a lazy one. The method is not difficult, it is just long, and the long part is the part a computer has to sit through.

If you want to do it yourself, everything you need is above and we are glad to have helped. If you would rather have the finished file, that is what we sell, and you now know exactly what you would be paying for.