New Business Opening Lead Intelligence — New York

Find potential upcoming businesses using public New York liquor applications, business-license applications and commercial construction records. Export deduplicated leads with deterministic evidence scores and the original public signals behind each lead.

Private preview. Opening signals are not confirmations. Scores are not probabilities or estimates of an opening date. This independent tool is not affiliated with New York State, NYC, SLA, DCWP or DOB. Verify public records before acting on a lead.

Coverage and sources

Source	Coverage	Publisher cadence	Role
Current SLA Pending Licenses	New York State premises	Daily	Pending liquor application
DCWP License Applications	NYC licensing; only NY premises retained	Weekly	New premises application; individual licenses and renewals excluded
DOB NOW Job Application Filings	NYC	Daily	Commercial construction support
DOB NOW Approved Permits	NYC	Daily	Initial commercial permit support

Publisher cadence does not guarantee current records. On September 17, 2026, DCWP's latest New application date was April 17, 2026, although renewal records were more recent. A 30-day DCWP query can legitimately return zero. SLA is a current pending snapshot, not a complete historical archive. Separate DOB electrical/elevator/LAA datasets are outside this release.

examples/input-dcwp-history.json uses a 180-day window and a zero score threshold to verify older DCWP records. It is a source diagnostic, not a fresh-lead benchmark.

Input

{"state":"NY","city":"New York","lookbackDays":30,"maxResults":100}

Field	Default	Meaning
state	NY	Only New York State
city	New York	NYC's five boroughs; empty string selects statewide NY coverage
boroughs	[]	Optional Manhattan, Bronx, Brooklyn, Queens, Staten Island filter
lookbackDays	30	Inclusive New York calendar-day age, 1–365; excludes future dates
businessCategories	[]	Exact case-insensitive source category labels, e.g. Restaurant; no invented category mapping
signalTypes	All four	liquor_license, business_license, building_application, building_permit
minimumSignalScore	30	Minimum score, 0–100
maxResults	1000	Final deduplicated result cap, 1–10000; fewer results are possible
includeLowConfidence	false	Include construction candidates without an identified operator
includeRawSourceData	false	Include allowlisted original business/source fields, never unrestricted raw rows

Statewide liquor search:

{"city":"","signalTypes":["liquor_license"],"lookbackDays":30,"maxResults":100}

Commercial construction review, including unknown operators:

{"signalTypes":["building_application","building_permit"],"lookbackDays":7,
 "includeLowConfidence":true,"minimumSignalScore":0,"maxResults":100}

Matching and scoring

Names normalize Unicode, spacing, punctuation and trailing legal suffixes. Displayed names and source addresses remain original. Matching retains house-number hyphens, ZIP codes, boroughs and unit information. Different units, conflicting ZIP/government business IDs or missing versus known units prevent a merge.

Records can merge at the same normalized premises using matching government identity, an exact legal/DBA relationship or conservative name similarity (95/100, long names, same first token and matching numeric tokens). Every group member must be compatible; ambiguous and transitive-only relationships remain separate. A shared building never proves a shared business. Conservative matching can miss real relationships.

DOB property owners, contractors and applicants are not treated as future tenants. Job and permit records can join by exact filing number and compatible premises. DOB records without operator evidence do not join a liquor/DCWP business on address alone. They retain businessName: null and low identity confidence. Use the construction input above to review them. This protects against assigning a landlord's or architect's name to a potential opening.

Initial weights: liquor 45, new premises license 35, permit 20, job application 10. Only the most recent signal of each type contributes its base weight. Ages 0–30, 31–60, 61–90 and 91–365 days receive 100%, 80%, 60% and 30% of that weight. Two independent agencies add 20; a third adds 10. Business identity adds 5; commercial work adds 5; unknown identity subtracts 10. DOB's two datasets count as one agency, not independent corroboration. Scores clamp to 0–100.

Levels: low 0–29, medium 30–49, high 50–69, very_high 70–100. Renewals, denied/withdrawn/expired/closed records, future/stale records, individual licenses, residential-only work and construction without commercial evidence are excluded. No LLM, contact enrichment or estimated opening date is used.

Output and failure behavior

One default Dataset item is one deduplicated potential opening. Fields include leadId, business/DBA/category, address, score/level/reasons, identity confidence, signals, dataset/agency counts, first/latest signal dates and retrieval timestamp. Each signal preserves its source record ID, source link, date, status, original name and address. Unknown phone, website, email and opening date remain null.

examples/output.json is fictional test data, not a real business lead. leadId is deterministic for the same evidence and independent of input order or exact duplicates. It is anchored to a source record and premises; it can change when that anchor disappears, changes or is superseded by new source evidence. This is not a persistent change-monitoring service.

DOB permit source IDs include filing number, work permit, sequence and work type, because the official export can repeat the first three across different work types. Those evidence rows do not create extra leads or repeated base-score contributions. Job record IDs include filing number and floor/unit scope. If a source still supplies conflicting versions of the same record key, all versions are excluded and counted as ambiguous in OUTPUT; coverage is marked partial rather than choosing one status.

The default Key-Value Store's OUTPUT contains source counts, source update times, failures, exclusions, duplicate count, qualified count and actual Dataset count. ambiguousSourceIdentities counts quarantined contradictory identities; per-source rejectedRecords includes their excluded rows. If some requested sources fail, remaining sources produce partial results with explicit coverage gaps. A failed source count is null, never zero. If all sources fail, the Run fails. Check sourceCoverageComplete before treating results as comprehensive.

Dataset writes are sequential and never automatically retried. If a write fails after partial delivery, saved items can remain accessible. Uncertain delivery counts are null in failed OUTPUT. Restart of a run with prior output intent is rejected; start a fresh run. No automatic cross-run deduplication is promised.

Privacy and limitations

Only public business and regulatory evidence is used. Personal license applicants, owner/architect names, phone numbers and mailing addresses are not selected for output. Optional raw fields are allowlisted; construction descriptions remain public work descriptions, not independently verified claims. No email, website, social profile, Google Maps or automated outreach enrichment is performed.

An application can concern an existing business or a change of operator. Commercial renovation does not guarantee a new opening. Missing units, incomplete identities, source delays and conservative matching affect recall. Nationwide coverage is absent.

Pricing and usage

Approved launch price: $0.005 per delivered qualified lead ($5 per 1,000), with platform usage included. 100 leads cost $0.50; 500 cost $2.50; 1,000 cost $5.00. Apify's standard apify-actor-start charge is separate: $0.00005 at 1024 MB. There is no custom base/run fee. Private Cloud billing verification has passed; publication remains a separate step.

Every default Dataset item is one final lead. Apify automatically bills apify-default-dataset-item; no custom qualified-lead event is charged. Source records, signals, duplicates, merged-away entities, rejected/conflicting/below-threshold candidates, source failures and OUTPUT are never result events. Zero leads means zero result charges; the standard start event still applies. Optional low-confidence leads must still satisfy the user's filters and are billed only when delivered.

The Actor respects the remaining run budget after startup, delivering only the affordable prefix of the ranked results, always within maxResults. OUTPUT reports actual datasetItems/returnedLeads and budgetTruncated=true when the spending cap removes otherwise selected leads. Source coverage status is independent of budget truncation. A smaller live population can also underfill maxResults without budget truncation.

Failed source/process runs may fail with no result charges. Failed appends can leave successfully delivered, billable items; no automatic POST retry occurs. Interrupted or restarted writes to an already used Dataset are rejected rather than replayed. A failed delivery reports count null when its committed count is uncertain; inspect the Dataset and platform event ledger. Fresh runs are independent purchases, not cross-run deduplication.

Use Apify API or schedules with the same input JSON. Export nested provenance as JSON; CSV is useful for the flat overview. Scheduling repeats a fresh source scan and can return previously observed leads. Contact the Actor's Store Issues/support route once published; during private preview, report run ID, INPUT, OUTPUT and stage logs to the owner.

Development and private deployment

Python 3.12; independent dependencies and storage, no imports from the NDIS Actor.

python3.12 -m venv .venv
.venv/bin/python -m pip install -r requirements-dev.txt
.venv/bin/python scripts/generate_schemas.py
.venv/bin/pytest -q
.venv/bin/ruff check .
.venv/bin/ruff format --check .
.venv/bin/python scripts/smoke_local.py
.venv/bin/python scripts/smoke_local.py --ppe
.venv/bin/python scripts/smoke_local.py --live examples/input.json
apify run
docker build -t new-business-opening:local .

apify run uses storage/key_value_stores/default/INPUT.json; start with clean storage. The smoke script uses temporary storage. Optional owner secret SOCRATA_APP_TOKEN improves public API quotas; core operation requires no token.

Transport: 10s connect/pool/write, 45s read inactivity, three GET attempts for transport, 429 or 5xx failures, 1s/2s backoff, 180s per-source deadline. Four sources fetch concurrently. Pages contain at most 500 rows and responses at most 8 MiB. More than 25,000 rows per source fails that source explicitly; narrow the date window. Stable row ordering, repeated-row checks and before/after publisher update checks guard pagination, but the API does not provide a transactional snapshot. Entity comparisons cap at 500,000; exceeding this fails rather than silently truncating results.

Create a second independent private Apify Actor from:

git@github.com:Ken0002/apify-intelligence-actors.git#dev:actors/new-business-opening

Use a read-only deploy key, hide source files, LIMITED_PERMISSIONS, 1024 MB and 600 seconds: Cloud-verified and approved production settings. Keep the Actor private. For private billing verification, configure PPE with Pay per event + usage OFF, primary apify-default-dataset-item=$0.005, and apify-actor-start=$0.00005 only. Do not add custom events or tier discounts; runtime rejects unapproved effective prices. See the repository release checklist for spending-cap and event-ledger tests before main or Store publication. Existing NDIS Actor source/settings must remain unchanged.