Every module, from the radar sweep to the signed contract.
StormSight is one platform that carries a storm the whole way: detect it, prove it, map it, score it, alert on it, route it, and sync it to your CRM. Here's how each piece works.
The Bayesian fusion engine
The central piece of intellectual property, and the reason a StormSight event means something a hail map never could.
Public weather data is a firehose: radar sweeps, satellite frames, lightning strikes, spotter reports, all arriving on their own clocks, all disagreeing at the edges. The fusion engine's job is to read all of it and answer one question honestly: how sure are we that damaging hail hit this exact spot?
It does that with a Bayesian log-likelihood-ratio model. Every signal (a MRMS radar estimate, a dual-polarization hail signature, a GOES lightning jump, an NWS warning, a CoCoRaHS hail pad) contributes a calibrated amount of evidence toward the probability of hail. The engine groups signals by the physical thing that produced them, so it never double-counts two views of the same radar scan. The result is a single, calibrated 0–100 confidence score you can trust to mean the same thing every time.
- Six independent measurement channels (radar reflectivity, dual-pol, satellite lightning, satellite IR, forecast, and ground-truth), weighed separately so corroboration is real, not an echo.
- Diminishing-returns math within a channel: the strongest signal counts fully, each additional one at half-weight, so nobody can stack the same evidence to fake confidence.
- MESH bias correction from the published meteorology literature, so giant-hail radar estimates are scaled down to what actually reaches the ground.
- Spatial and temporal coherence bonuses: tight clusters of reports near the radar's timing count for more than scattered noise.
- Every event carries a stamped weights version, so a detection replays identically months later, which matters when a claim goes to dispute.
The Storm Map
Where your sales day starts. Zoom out to see where it stormed; zoom in to see which roofs.
The Storm Map is the working surface for the whole team. Zoomed out, it paints an impact-density heat field (more impacted roofs, more heat), so a rep can see at a glance which part of the metro took the worst of last night's storm. Zoom in and that heat resolves into per-home score dots.
Two color ramps tell two different stories. A warm ramp shows Recent impact: a fresh storm, a live claim lead. A cool ramp shows Lifetime exposure: a roof that's been weathered over a decade, a long-play replacement lead. A single toggle flips between them. Layer on the raw evidence when you want to see the work: colored dots for ground reports, radar, satellite, and warnings; dashed polygons for NWS tornado and severe-storm warnings; storm-day outlines tagged with the date so same-color, same-date means same storm.
- Impact heat field zoomed out; per-home Recent and Lifetime score dots zoomed in.
- Severity presets (1″, 1.5″, 2″) thin the swarm down to only the most severely impacted homes.
- A time slider replays how the storm field built through the event.
- Evidence layers: ground / radar / satellite / warning signals, SPC outlooks, and validated-parcel highlighting for roofs with direct radar hail evidence.
- Storm Map 2.0 adds a continuous Region → Area → Home zoom with a breadcrumb trail, a Hail / Wind / Combined peril switch, and a single Storm Site Impact score that every layer speaks.
Events & time-accurate storm replay
Open any event and get the full evidence trail, plus a radar replay of the exact storm, centered on your customer's home.
Every detected event has its own page: the signals that created it, the confidence score and how it was reached, CSV and PDF exports for the office, and a revocable public share link you can hand to an adjuster or a homeowner without giving away a login.
The replay is the demo that closes doors. StormSight streams the archived NEXRAD reflectivity for the real event window and plays it back frame by frame, with the event-time hail field painted on top and an optional pin on the customer's address. Your rep stands on the porch and shows the homeowner the storm rolling over their roof, at the minute it happened. And it's honest to a fault: if a historical radar frame doesn't exist, the map flags the gap instead of quietly showing today's weather.
- Full evidence trail per event with CSV / PDF export.
- Revocable share links for adjusters and homeowners: read-only, no account needed.
- Time-accurate NEXRAD reflectivity replay with a playhead and a home pin.
- Optimized canvassing route plans from a storm's impacted parcels, using real drive-time isochrones.
- Side-by-side event comparison for overlapping or repeat storms.
On-request address scoring & exposure grades
Type in any address in Oklahoma, Kansas, or Texas and get an objective, insurance-framed read on its storm exposure, instantly.
You don't have to wait for an address to be catalogued. Paste any street address (or drop a pin) and StormSight resolves it, folds its full weather history, and returns a score on the spot. A known address returns exactly what the map paints; a brand-new point is computed fresh from the event and signal record and saved back, so the next person sees it on the map too.
A raw score of 47.3 means nothing to a homeowner, so StormSight translates it into an exposure grade an adjuster's language recognizes: Severe, High, Moderate, Low, or None, each tied to real hail-size and exposure thresholds. Under the hood, a Recent score decays over roughly fifteen months while a Lifetime score accumulates ten years of exposure with no decay, and wind is tracked as its own separate insurance peril alongside hail.
- Score any OK / KS / TX address or lat/lon on demand, with no pre-cataloging and no waiting.
- Five-tier, insurance-framed exposure grade (Severe → None) tied to published hail-size thresholds.
- Separate Recent (15-month, decayed) and Lifetime (10-year, cumulative) scores for claim leads vs. replacement leads.
- Hail and wind scored as distinct perils, then combined into one Storm Site Impact number.
- Out-of-region addresses return an honest "out of coverage," never a false "no impact."
Prospects & the leads pipeline
Turn an impacted map into a per-crew work queue, then push it through a real sales pipeline.
The prospect queue starts empty on purpose. Instead of drowning your team in every address near a storm, StormSight lets a rep click the homes that matter and add them to their crew's queue, a curated list ranked by expected hail size against building value, so the biggest, most-damaged, most-valuable roofs float to the top.
From there it's a straight sales pipeline. Promote a whole storm's ranked parcels into leads in one move, then work each one through a six-stage kanban (New, Assigned, Contacted, Inspected, Contracted, Declined) with a full timeline and the parcel's details (year built, square footage, last sale price, assessor value) attached to every card.
- Curated, per-team prospect queues: reps add the homes worth working, not all 80,000 near the storm.
- One-click promote-from-event: a storm's ranked parcels become leads, sorted by hail size × building value.
- Six-stage kanban pipeline (New → Contracted) with click-to-advance status.
- Parcel context on every card: age, size, last sale, assessor value.
- Every score is team-agnostic underneath, so the same roof reads the same for every crew.
Territories & monitored areas
Draw a circle around the neighborhoods you protect. When real hail lands inside it, the right crew hears about it.
Territories are the front door of the canvassing loop. A manager draws a circle on the map (anywhere from a single subdivision to a whole county), assigns the crew or crew-group that owns it, and sets the thresholds that matter: minimum hail size to bother them, minimum confidence to trust it.
After that it runs itself. Every fifteen minutes StormSight checks for significant hail that actually intersects each territory, dedupes so nobody gets the same storm twice, and emails the assigned crew, often with a short AI-written summary of what hit and how sure the system is. You can monitor by circle, county, city, or state.
- Draw-a-circle territories from a single block up to 200 km across, or monitor by county, city, or state.
- Assign individual crew members or named groups ("North OKC Crew") and StormSight resolves the recipient list.
- Per-territory thresholds for hail size and confidence, so a crew only hears about storms worth a truck roll.
- A 15-minute dispatch loop with per-event dedup: real-time coverage without alert fatigue.
- Member-level saved-site geofences for the individual rep watching their own book of business.
Alerts across every channel
Push, SMS, email, and voice, delivered the minute a storm qualifies, logged so nothing slips and nothing double-sends.
A detection nobody sees is worthless, so StormSight treats delivery as its own discipline. Severe events go out instantly; quieter ones roll into a digest. Crews get a push notification on the mobile app for free; SMS kicks in for the big stones; email carries the branded summary; and for the truly significant events, an optional voice call escalates until someone presses 1 to acknowledge.
It's all built to be trusted at scale. Delivery is idempotent, so the same event, version, and channel never fire twice. Every send is written to an immutable audit log, and the SMS path handles STOP, HELP, and START automatically to stay on the right side of TCPA.
- Four channels (push, SMS, email, and voice escalation), matched to event severity.
- County and state hail-alert subscriptions alongside territory alerts, matched by real county polygons.
- Idempotent delivery: one event · one version · one channel, never a duplicate.
- TCPA-compliant messaging with automatic STOP / HELP / START handling.
- An immutable alert audit log and an in-app inbox with an unread bell.
The AI meteorologist
A reasoning layer on top of the deterministic engine, with fail-closed guardrails and a one-click kill switch.
The math decides what happened. The AI meteorologist decides what to do about it. On a schedule, it reviews each candidate situation (a radar cluster, a spike in social reports, an unresolved warning, a lightning jump), reasons over the evidence with a set of read-only tools, and proposes one of four actions: promote it to an event, wait for more evidence, dismiss it, or send it to a human.
It is deliberately kept on a short leash. Hard guardrails force anything suspicious to human review: a low-baseline promotion, a high-baseline dismissal, anything under the two-source minimum. A cost-and-rate safety brake trips to human review if it starts behaving oddly, and a single click at the admin console drops the whole layer back to suggest-only. Every decision records its full reasoning trace, so you can replay exactly why the system did what it did.
- Promote / wait / dismiss / human-review decisions on every candidate, on a continuous schedule.
- Fail-closed guardrails that force uncertain calls to a person, never past them.
- Autonomy modes from suggest-only up to full-auto, each change gated and audit-logged.
- A safety brake plus a one-click kill switch back to suggest-only.
- Deep Look: a customer-facing AI read on a specific roof's score, blended toward the deterministic baseline so a re-run can't tank a well-corroborated home.
The mobile field app
The whole platform in your crew's pocket: same map, same events, same pipeline, same login. iOS and Android.
Roofing happens on a roof, not at a desk, so StormSight ships a native mobile app that shares the same identity and the same data as the web. Reps carry the storm map, the events list with one-tap CSV/PDF sharing, their monitor zones, and the leads pipeline as a swipeable kanban.
Field Notes is the piece built for the porch: snap a photo with the camera, tag it with GPS, and it uploads straight to the address record. Push notifications land the second a monitored storm qualifies, so the crew closest to the damage is the first one knocking.
- Native iOS and Android app on a shared login and API.
- Storm map, events, monitor zones, and a swipeable leads kanban.
- Field Notes: camera + GPS photo capture tied to the address graph.
- Instant push notifications for qualifying storms.
- Account and billing management, including the Stripe portal, on the device.
CRM sync, integrations & API
StormSight plugs into the software your shop already runs, starting with AccuLynx.
A lead that lives in a silo dies in a silo. StormSight syncs two ways with AccuLynx (pushing storm leads in as jobs, tracking status both directions) and overlays your existing AccuLynx customer book right on the storm map. When a storm rolls over a home you've already sold, it re-pings you: that customer just became a warranty call, a re-roof, or a referral.
Beyond AccuLynx, StormSight posts alerts into Slack and fires HMAC-signed webhooks into JobNimbus, Zapier, and anything else you run, and it exposes a documented API (67 endpoints under a public OpenAPI spec) for the shops that want to build their own connections.
- Two-way AccuLynx sync: leads out as jobs, status back in.
- A "My Customers" overlay that re-pings existing customers when a storm hits their roof.
- Slack alert posting and signed outbound webhooks to JobNimbus, Zapier, and more.
- A documented REST API (67 endpoints, OpenAPI 3.1) for custom integrations.
- US-hosted throughout: customer data does not leave US data centers, and is never used to train models.
Built on free public data. Verified against the record.
StormSight covers Oklahoma, Kansas, and Texas today, fusing eleven free public sources from NOAA, the NWS, and the National Severe Storms Laboratory, which is why the per-event data cost is $0. Detection runs continuously, with a median of about fifteen minutes from a radar scan to a mobile alert, and the confidence engine is published on a public methodology page and backtested against the NOAA storm archive every quarter. Nothing here is a black box.
See it on your own territory.
Start a free trial, or tell us where your crews canvass and we'll show you the last storm that rolled through, scored, mapped, and ready to work.