- Oct 20, 2026
- 3 min read
Matching Businesses to Companies House Without False Positives
In the UK, cold email to companies (Ltd, LLP, PLC) is allowed; to sole traders and partnerships it isn’t without consent (the rules). Google Maps names rarely say “Ltd”, so each business needs checking against Companies House, the official register.
The API
Free, no identity checks for read access, about 600 requests per five minutes. Register at the Companies House developer hub, create a Live application, then a REST key. Authentication is HTTP Basic with the key as the username and an empty password:
url = "https://api.company-information.service.gov.uk/search/companies?" + urllib.parse.urlencode({"q": name, "items_per_page": 5})
auth = base64.b64encode(f"{CH_API_KEY}:".encode()).decode()
items = json.load(urllib.request.urlopen(urllib.request.Request(url, headers={"Authorization": f"Basic {auth}"})))["items"]
# each item: title, company_number, company_type (ltd/llp/plc/…), company_status, address
The naive match, and what it got wrong
First attempt: split both names into words, measure the overlap, accept above a threshold. With a few fictional examples in place of the real ones, it produced:
| Google Maps name | Matched to | Right? |
|---|---|---|
| Example Vets | EXAMPLE VETS LLP | Yes |
| Bright Vet Care - Exeter | EXETER VET CARE LIMITED | No, a different business |
| The Quay Street Veterinary Practice | CENTRAL QUAY VETERINARY PRACTICE LIMITED | No |
| Paws on Mill Road | BIRCH COURT (MILL ROAD) LIMITED | No, it’s a street name |
The words doing the matching were “vet”, “care”, “practice”, the city and a street name. They say what a business does or where it is, not which one it is.
Rules that held
1. Ignore generic words. Trade words, place and street words, legal suffixes and the business’s own city don’t count:
GENERIC = set("""ltd limited llp plc the and co uk group services
vet vets veterinary care practice clinic centre surgery studio
road street lane avenue park court north south east west
of in at for by on""".split())
def norm(name, city=""):
words = set(re.findall(r"[a-z0-9]+", name.lower()))
return words - GENERIC - set(re.findall(r"[a-z0-9]+", city.lower()))
2. Accept only strong evidence. A candidate passes if either:
- the postcode matches the listing’s and they share at least one distinctive word, or
- the names are identical once legal suffixes, “the” and a trailing ”- City” are removed.
same_name = plain(maps_name, city) == plain(ch_title, city)
if not (same_postcode and shared_words) and not same_name:
continue # not the same business
3. Don’t require the same city. I tried “registered office in the same city” as a fallback. It lost real companies: small businesses often register at their accountant’s address, in another town.
After these rules, every remaining match was correct: four confirmed by postcode, two by identical names. The cost was a few real companies left unmatched. That’s the right trade.
Use the trading name in the email
Companies House returns registered names in capitals: “EXAMPLE VETS LLP’s
website” reads like a robot wrote it. The email uses the trading name from
the listing (“Example Vets”); the registered name is kept in a
company_evidence column, as the record of why this business could be
emailed.
Takeaways
- Generic words create false matches. Remove trade, place and street words before comparing.
- Accept postcode or identical-name evidence only.
- A missed lead is better than the wrong company name in an email.
- Trading name for people, registered name for the record.