Organised retail crime patterns
Coordinated push-out crews, fence-channel chatter, cargo diversion, and gift-card draining — surfaced from Telegram, Signal, and X before the morning brief.
Example leaf · orc.push_out → alert_nowUse case 03LP · ORC · Asset protection · PR
External public-media intelligence for retail LP, ORC, and AP teams.
Telegram resale boards, fenced-marketplace listings, regional news, and viral store incidents — classified against a versioned taxonomy and routed into Auror, ThinkLP, Sensormatic IQ, or your internal case-management tool. We sit next to your investigation stack, never inside your cameras.
Talk to us about retail intelligenceCAPABILITY MAP · TAXONOMY V0.1
Five categories. One stable routing contract.
A versioned public-media signal taxonomy for retail LP, ORC, AP, and PR — designed so the rationale on every alert is reproducible, auditable, and reviewable by an analyst on the way out. The full taxonomy document is published with this release.
Fenced-merchandise listings
High-shrink resale across detergent, formula, OTC pharma, cosmetics, designer apparel, power tools, and electronics — patterned, not one-shot.
Example leaf · merch.formula → alert_nowStore-incident severity
Verbal aggression, physical aggression, weapon-involved, worker injury, customer injury, and fatality — severity-ordered, leaf-routed.
Example leaf · incident.weapon_involved → alert_nowViral-incident narrative templates
Parking-lot fights, shoplifting clips, brand-shaming threads, hoax incidents, and influencer call-outs — reputational events that belong in PR, not LP.
Example leaf · viral.brand_shaming → pr_onlyParking-lot incident classes
Vehicle break-ins, lot robberies, hit-and-runs, shootings, and cart-yard dumping — distinct from in-store because the response is usually local-PD-led.
Example leaf · parking.robbery → alert_nowWORKFLOW
Signals → taxonomy → routing → evidence → your tool.
Five steps, end-to-end. The labels are tuned to retail-LP playbook actions; the engine is general.
Public sources only.
Telegram resale boards, fenced-marketplace listings (eBay, Facebook Marketplace, OfferUp, Vinted, Allegro, OLX), regional news, RH-ISAC-style chatter, and public-records bulletins. No in-store CCTV. No POS exception data.
Versioned classification.
Every signal classified against the retail-LP signal taxonomy v0.1 — five categories, leaf-pinned, with the version stamped on every chip so the rationale is reproducible.
Four buyer-side actions.
Each leaf carries a default action: alert_now (minutes — LP duty / AP on-call), file_and_watch (hours — ORC investigator), morning_brief (next business day — regional AP), or pr_only (corp comms).
What the analyst gets.
Source URL, publisher, region, dated excerpt, the taxonomy leaf, the modifiers applied (named_retailer_brand, known_orc_cell, high_severity_overlap), and the rationale — exportable as a self-contained case file.
Into the tools you already run.
Structured events into Auror, ThinkLP, Sensormatic IQ-style investigation queues, or your internal case-management system over webhook. We sit next to your existing stack — we do not replace it.
WORKED EXAMPLE
Open the demoThree real public sources, one Northern California ORC pattern.
A reproducible demo built against this taxonomy: a regional broadcast bust, a flea-market resale write-up, and an X-amplified social clip — classified, rolled up, and routed into a mock investigation queue. No Auror partnership implied.
TRUST POSTURE
Public sources. No biometrics. EU coverage you actually use.
The trust posture is the product. Public-media classification only, never per-individual identification, with first-class EU and Polish-language source coverage as a structural moat over US-only social-listening stacks.
- Non-biometric by designNo facial recognition, no per-individual identification, no suspect attributes. We classify public-media signals, never people. The trust posture is the product, not a feature flag.
- EU and Polish-language source coverageNative coverage of Allegro, OLX, Polish and EU-regional Telegram channels, and central-European local news — a structural moat over US-only social-listening stacks that miss the cross-border ORC layer.
- Reproducible classificationEvery alert ships with the taxonomy version, the leaf, the modifiers, and the rationale. Reviewable by an analyst, defensible in procurement, and auditable on request.
- Public sources onlyNo scraping behind a login, no purchased data brokers, no synthetic surveillance. If a source is not already public, it does not show up in your queue.
WORKER-SAFETY CONTEXT
We surface the signal. Your compliance team owns the program.
Two recent statutes have made workplace-violence prevention a board-level item for US retailers. We do not certify your program — but incident.worker_injury and incident.weapon_involved leaves in our taxonomy are designed to feed the incident log your compliance team is already required to maintain.
Public-media signals will never substitute for an in-store incident report. They will surface the regional context — adjacent stores, copycat windows, cargo-corridor escalation — that the in-store log alone cannot.
NY Retail Worker Safety ActEffective 2025. Requires retailers with 10+ employees to maintain a written workplace-violence prevention program and incident log.dol.ny.gov / retail-worker-safety-act →
California WVPP (SB 553)Effective July 1, 2024. Mandates a workplace-violence prevention plan, training, and a logged incident record for nearly all employers.dir.ca.gov / workplace-violence-prevention →
HONEST LIMITS
What MediaIntel does not do.
The shortest path to a real customer engagement is being explicit about the lane we are not in. These limits are structural, not a roadmap.
No in-store computer vision
We do not watch your cameras. In-store concealment, sweethearting, and self-checkout exception classification stay with Everseen / Veesion / Sensormatic IQ / StrataVision / Trigo — that is their lane.
No facial recognition or biometric watchlist
Per-individual identification is never the output. "Reported suspect" framing is the maximum we surface, and only when the upstream public source has already published it.
No cashier-less or autonomous-store framing
We do not claim to enable, replace, or augment cashier-less store formats. Those are unrelated platform problems.
No "AI shoplifting detection"
We are not in the in-store-CV detection lane. Public-media intelligence is the layer above your investigation tool, not below your camera fleet.
No blanket compliance product
We do not certify your store operations as GDPR / EU-AI-Act / OSHA-compliant. We surface signals; your compliance posture remains yours.
No POS / CCTV replacement
EAS, POS exception management, and CCTV stay in place. Structured events feed in over webhook — we sit next to your existing LP stack, not on top of it.
Questions retail teams ask first.
From the conversations that come up in LP / AP procurement and during the first taxonomy review.
Is this an in-store loss-prevention or shoplifting-detection product?
No. We classify public-media signals — Telegram resale chatter, fenced-marketplace listings, regional news, viral store-incident clips. In-store concealment and self-checkout exception classification are a different lane (Everseen, Veesion, Sensormatic IQ, StrataVision, Trigo). The honest framing is “a dedicated externalpublic-media intelligence layer for retail LP, ORC, AP, and PR teams.”
Does this replace Auror, ThinkLP, or Sensormatic IQ?
No. Those are the investigation queues. We pre-classify external signals against a stable taxonomy and feed structured events into your existing tool over webhook — Auror, ThinkLP, Sensormatic IQ, or your internal case-management system. We sit next to the investigation tool, not in place of it. No partnership with those vendors is implied.
How does the routing actually work?
Every leaf in the taxonomy carries a default action: alert_now (minutes — LP duty / AP on-call), file_and_watch (hours — ORC investigator), morning_brief (next business day — regional AP director), or pr_only (CMO / corp comms / legal).
Modifiers like named_retailer_brand, known_orc_cell, and high_severity_overlap can escalate the default action per customer policy.
Do you do facial recognition or biometric watchlists?
No. Never. We classify signals, not people. The maximum suspect framing we surface is “reported suspect” — and only when an upstream public source has already published it. The non-biometric posture is structural, not a feature flag we can flip.
Why does EU and Polish-language coverage matter?
Most US-built social-listening stacks treat Allegro, OLX, central-European Telegram channels, and Polish-language regional news as long-tail. For retailers operating across the Polish border, the Baltics, and the German-speaking DACH region, that long tail is the ORC layer. Native coverage of those sources is a structural moat — not a translation toggle.
Are you a compliance product for the NY Retail Worker Safety Act or CA WVPP?
No. NY Retail Worker Safety Act (2025) and CA SB 553 / WVPP (July 2024) require a written workplace-violence prevention program — that is a policy, training, and logging problem your EHS / compliance team owns. We surface incident.worker_injury, incident.weapon_involved, and adjacent public-media signals that inform the program. We do not certify it.
How is this not over-claiming versus competitors?
We are not “first AI loss prevention” — Everseen and Veesion have been in that lane for years. We are not “first visual analytics for retail” — Talkwalker has shipped that since 2017. We are not “first retail crime intel” — Auror has run that since 2014. The honest claim, and the only one we make: a dedicated external public-media intelligence layer for retail LP teams, with a versioned taxonomy and a non-biometric trust posture.
Retail intelligence · LP · ORC · AP · PR · taxonomy review in week one
Talk to us about
retail intelligence.
Bring your three highest-shrink SKUs and your top regional Telegram and marketplace sources. We will return a classified two-week sample against the published taxonomy inside three weeks.