Telling an automation firm from a copier dealer

“Automation” is a crowded word. On Google Maps it belongs to copier dealers (“office automation”), RPA software houses that sell bots for office work, marketing agencies, training institutes and estate agents. Meanwhile plenty of real automation firms have names that say nothing about what they do.

Layer 1: the name and Google’s category

The cheap checks run first, on what the listing already says. A place is left out if it is:

• permanently closed;
• a school, course or training centre, because it teaches automation rather than sells it;
• another trade by name: copiers and printers, pest control, RPA, test automation, marketing automation, chatbots, real estate, even flats for rent and homestays. A name that also names a core service, like “Home Automation, Office Automation”, survives this rule;
• in a Google category that rules it out, such as estate agents, apartment blocks and schools;
• a shop or unrelated business, like a restaurant, salon or tyre dealer, whose name has no automation word in it.

On the positive side, a name with a strong pattern (gate automation, motorised curtains, building management, KNX, PLC, SCADA, AV integration and so on) is listed straight away, with a category taken from the name.

Layer 2: the firm’s own website

Everything still undecided gets one more look: its own home page, fetched once.

• Only the firm’s own words count. Social pages, business directories, site builders and link shorteners (22 hosts) are skipped, because those pages are templates.
• One bounded fetch: a 15-second timeout, then the page title, meta description and visible text, with scripts and styles stripped.
• Twelve groups of phrases are counted: six physical trades (home automation, building management, industrial controls, gates and shutters, curtains and blinds, audio-visual), control panels, instrumentation, security, software and IT, training, and the word automation itself.

The verdict needs a margin, not just a mention:

trade, hits = best_of(scores, TRADES)   # home, building, industrial, gate, curtain, av
promote = (hits >= 3
           and hits >= scores["it"]
           and scores["training"] <= max(hits, 4)
           and not junk_title(title))

In words: a firm is promoted only when one physical trade gets three or more hits, at least as many as software and IT language, and training language doesn’t dominate. A title that reads like a software house, a marketing or AI agency, a course, a login page, a betting site or a domain for sale blocks the promotion, unless the title itself says automation or smart home.

Labels are not promotions

Two hits are enough to label a firm’s trade, but not to let it in. Control-panel builders and instrument suppliers can be labelled industrial automation, but never promoted on that alone. Three or more security hits (CCTV, access control, intercoms) give a security label, nothing more.

The website also corrects labels for firms already listed by name. A firm called “Firm 0412 Automation” whose site is all lighting control and smart switches is filed under home automation, not the vague “automation”.

What it changed

In the Mumbai list, 318 of 1,176 firms (27%) are there because their own website said what they do, not their name. Outside Mumbai, another 641 got in the same way. Every row in the workbook says which it was: “Name” or “Website text”.

It also shows the limit. Nearly half of the national list, 5,218 of 11,007, is labelled “Automation (type not stated)”: the name says automation or controls, but nothing says which kind. Only 58% of the national list has a website on record at all, and a check can’t read a site that isn’t there.

Rules I’d keep

1. Run the cheap rules before any fetch.
2. Read the firm’s own words, never a directory’s template.
3. Ask for a margin: a trade has to beat software and training language, not just appear.
4. Keep labelling and admitting as separate decisions.
5. Record why each row got in.