Business Service Extractor

Extract services that a business has explicitly declared in JSON-LD on one official business website page. The Actor does not infer services from visible page text.

Who it is for

B2B list-quality teams, service-category researchers and sales operations users who need traceable, machine-readable service declarations, not marketing-text guesses. One URL per run. No accounts, logins, LinkedIn collection or personal-profile harvesting.

Input

{"url":"https://firstcapitol.co.uk/debt-collection-agency-london/"}

This is a real publicly reachable business page (independently accessible on 2026-11-01), not a fabricated/example domain. The live HTML's current JSON-LD service coverage could not be verified by the build environment. Accordingly, service results for this URL are not guaranteed; honest not_found output is part of normal operation.

Output

One row per explicitly declared Service (or recognized Schema.org service subtype) from JSON-LD @graph declarations, makesOffer, hasOfferCatalog, nested OfferCatalog entries, or itemOffered. @id references to same-page entities are supported. Output keys remain stable:

• businessName: a published provider business name, or name of the organization declaring an offer; null if not present
• serviceName: published Service.name, otherwise null
• serviceType: published serviceType, otherwise explicit Schema.org Service @type
• description: first 300 characters of the published description with whitespace normalized; null when not provided, never generated
• sourceUrl: fetched input page URL
• status: found or not_found
• error: null for found rows, or readable explanation of not_found

An illustrative fixture result, not live evidence for the prefill target:

{"businessName":"Northstar Trade Services","serviceName":"Business Debt Recovery","serviceType":"B2B collections","description":"Collection service for businesses.","sourceUrl":"https://firstcapitol.co.uk/debt-collection-agency-london/","status":"found","error":null}

If the page has no declared Service, one row is emitted with businessName, serviceName, serviceType, and description all null, sourceUrl set, status:"not_found", and an informative error string.

Honest limits

• No service guessing from headings, navigation, prose, product offers, page titles, or catalog names; a Service node must explicitly exist in embedded JSON-LD. Does not read visible text, Microdata, RDFa, externally linked JSON-LD or other pages.
• Site-authored JSON-LD is self-declared and not proof a service is currently offered. Data may be incomplete, outdated, duplicated, or malformed. Duplicate same-name/type/description declarations within a page are coalesced.
• @graph / @id in-page links and nested offer catalogs are supported. If no provider name is explicitly linked, businessName is null (never inferred from the domain).
• Only recognized Schema.org Service types are eligible; exotic unlisted service subtypes may be missed. Long descriptions are clipped to 300 characters without generating summaries.
• No cross-domain redirects, login, JS rendering, or crawling. Enforces HTTPS and public IPs, robots.txt, TLS validation, a 12-second network timeout and 2 MB HTML limit.
• Public sites may block bots or omit JSON-LD. Failures produce a plain ERROR: line without a traceback. The Actor never extracts contact details or personal data.

Local and Apify use

Python 3.12+ standard library exclusively; requirements.txt is intentionally empty of dependencies.

$ACTOR_INPUT_JSON='{"url":"https://firstcapitol.co.uk/debt-collection-agency-london/"}' python src/main.py

Alternatively put the same object in storage/key_value_stores/default/INPUT.json. Local results: storage/datasets/default/000000001.json, etc. For Apify, input is read using APIFY_DEFAULT_KEY_VALUE_STORE_ID + APIFY_TOKEN and GET /v2/key-value-stores/{id}/records/INPUT; dataset records are written to APIFY_DEFAULT_DATASET_ID via POST /v2/datasets/{id}/items. ACTOR_INPUT_JSON takes precedence for local tests. Live Apify platform operation was not tested; API behavior was verified with a fixture.

Tests and actual script-mode output

python tests/run_direct_smoke.py
python -m unittest discover -s tests -v

The direct runner executes the actual src/main.py script in a subprocess, not an imported main() function, using a valid URL and a simulated offline response. Evidence in tests/SMOKE_OUTPUT.txt, dataset in tests/SMOKE_DATASET.json, full unittest output in tests/TEST_RUN_OUTPUT.txt. The fixture file tests/sitecustomize.py is test-only and activated with ACTOR_TEST_HTTP=1 + test PYTHONPATH; production never uses it.

Store category and price suggestion

Candidates: LEAD_GENERATION, DATA_EXTRACTION. Suggested pricing: $3 per 1,000 dataset items ($0.003 each), model PRICE_PER_DATASET_ITEM. This is a planning suggestion, not deployed billing. Before any future billing activation, decide whether not-found rows should be charged.